Validation of the Hi-sAFe and yield-safe models for simulation of Northern European alley cropping systems

Table of contents

1 Introduction
1.1 Agroforestry as a solution
1.2 Simulation of alley cropping systems
1.3 Research question:

2 Materials and method
2.1 Selection of the models
2.2 The modelled growing systems
2.3 Modelling principles and procedure
2.4 Use of Hi-sAFe
2.5 Use of yield-safe

3 Results
3.1 Final calibrations
3.2 Validation runs

4 Discussion
4.1 Performance of the models
4.2 Sources of error
4.3 The role of models in the development of alley cropping systems

5. Conclusion

6. Literature

Appendix A - values of crop and tree parameters

Appendix B - parameterisation of crop growth phenology

Appendix C - crop species simulated in different years

Appendix D - values of parameters for growing system details

Appendix E - details on weather parameters

Abstract

Alley cropping is an agroforestry practice with considerable potentials of providing both environmental and economic benefits in Northern Europe. However, these potentials are rarely quantified, and the knowledge required to design new alley cropping systems is lacking. Computer simulation of alley cropping systems may provide solutions through estimation. The objective of this study is to validate the Hi-sAFe and yield-safe models regarding predictions of tree growth and crop yields in Northern European alley cropping systems.

Data from the Ministry of Agriculture, Fisheries and Food-sponsored alley cropping systems (MAFF- sponsored trials) and the Combined Food and Energy system, Copenhagen (CFEC) were used. For the MAFF-sponsored trials, data from one (Cirencester trial) out of three sites (Cirencester, Silsoe and Leeds trials) were used for calibration. For the CFEC, data for a subset (years 2012-2019) of the recorded period (2000-2019) was used for calibration. Then, validation simulations were run with the remaining data from each set. For the English growing systems, both models showed substantial error in crop yield predictions (r2 = 0.02-0.57), but successfully simulated tree growth (r2 = 0.94-1.00). In the Danish systems, yield-safe produced better predictions of crop yield (r2 = 0.01-0.35) than Hi- sAFe (r2 = 0.03-0.05), and none of the models reproduced the variability of tree growth measurements (r2 = 0.02-0.09). For all simulations, yield-safe was substantially easier to use, and thus more efficient, than Hi-sAFe, which on the other hand has been used to deliver critical insights in transdisciplinary studies.

Suggestions for further model developments are given, and appropriateness of using the models for different tasks is discussed. This may have relevance to researchers of agroforestry modelling and potential model users.

Acknowledgements

I would like to thank my supervisors for their guidance. Furthermore, I would like to thank the management of Institute of Plant and Environmental Science, which financed my journey to Montpellier, where I met Marie Gosme, Christian Dupraz, Guillaume Blanchet and the rest of the Hi-sAFe team. I am very grateful for their tireless work of developing the Hi-sAFe model, and for the hospitality, education and guidance they have given me. Finally, I would like to thank Rose, Jesper and Katharina for being such lovely roommates and companions during the writing process, and for putting many smiles on my face.

1 Introduction

Agriculture globally faces a host of challenges. These include global climate change and increased incidence of extreme weather conditions (IPCC, 2014), the need to feed an increasing population (Pretty, 2018), and the biodiversity crisis (Rockstrom et al., 2009), which only represents one aspect of the necessity to improve the environmental impact of most human enterprises. Agroforestry (AF) holds solutions for some of these challenges, and the potential to ameliorate the rest, giving it a vital role in the future of agriculture (Nair and Garrity, 2012; Seiter et al., 1995).

On a European scale, 8.8% of the total agricultural land is used for AF (Burgess and Rosati, 2018). The Mediterranean region has the highest ratios of AF to total used land, likely because different practices of silvopastoralism are widespread, but also due to silvoarable systems with, e.g., olive trees (Olea europaea) (Correal et al., 2009; Mosquera-Losada et al., 2012). However, the adoption of AF practices could provide many benefits in Northern Europe as well (Kay et al., 2019). Newman and Gordon (2018) state that adoption is inhibited by practical limitations, such as questions of how to design a system that is easy to manage, as well as perceptional barriers, such as the view that AF is less productive or less profitable than monoculture farming. What is needed is examples that clearly show that AF is more efficient and profitable than monocultures. These examples may be conceived of, designed and studied using computer models. This study studies two modelling programs designed explicitly for simulating agroforestry systems (AFS), namely Hi-sAFe (Dupraz et al., 2019) and yield­safe (van der Werf et al., 2007). Specifically, the potential of the software to aid in the process of designing new AFSs and studying existing systems is examined.

This chapter starts by defining the concepts of agroforestry and alley cropping and discussing their potentials. Then, the argument that simulation is needed is unfolded, and the two models are presented. Finally, the study objective is stated.

1.1 Agroforestry as a solution

1.1.1 Definition of agroforestry and alley cropping

The following criteria were set up as critical for identifying an agroforestry system (AFS) (Nair, 1993, p. 14):

- “agroforestry normally involves two or more species of plants (or plants and animals), at least one of which is a woody perennial;
- an agroforestry system always has two or more outputs;
- the cycle of an agroforestry system is always more than one year;
- even the simplest agroforestry system is more complex, ecologically (structurally and functionally) and economically, than a monocropping system.”

This definition makes it possible for the term to encompass a vast diversity of different kinds of growing systems, including tropical, multi-storey home gardens (Nair, 1993) , silvopastoral systems, multistrata plantations, and agrisilvicultural systems, such as alley cropping systems (ACSs). Alley cropping is a type of agroforestry growing system that comprises the two elements arable farming and the growing of trees in rows. Usually, rows of trees are planted in a grain field at regular spacings to accommodate the use of agricultural machines.

1.1.2 Potentials of agroforestry and alley cropping

Multiple benefits of using agroforestry (AF) practices have been reported. Here, they are grouped into three categories:

- Ecosystem services
- Increased system stability
- Enhanced efficiency, i.e. increased yield or decreased reliance on external inputs Ecosystem services

Abbildung in dieser Leseprobe nicht enthalten

Figure 1.1: A range of ecosystem services that AFSs provide, and the scale at which they manifest. Reprinted from Jose (2009).

Figure 1.1 presents a list of the most important ecosystem services provided by AFSs in a broad sense of the concept. Some of the elements are expanded upon in the list below.

- Mitigation of climate change through carbon sequestration: Carbon is stored in AFSs in aboveground and belowground biomass and as soil organic matter (SOM) (Nair, 2012). In growing systems composed of only annuals, due to the annual decomposition of most of the biomass and differences in the chemical composition of the plants in the growing system, these processes do not occur at the same rates. The higher production of ligninous compounds in woody plants makes for higher input of lignin to the soil in AFSs. In the soil, it can be transformed into relatively stable compounds of SOM (Whalen and Sampedro, 2010). One study reported that a mixed species AFS in Puerto Rico stored 15.21 Mg C/ha/yr (Nair et al., 2009).
- Increasing soil fertility: Inputs of tree leaf litter and prunings that are regularly given to the soil mean that levels of SOM increase in many systems (Hoosbeek et al., 2018). Generally, AFSs have more architecturally diverse root systems, and therefore, they have a larger potential for feeding soil biota at a range of soil depths. This soil biota comprises organisms such as symbiotic fungi (Barrios et al., 2012). Thus, the microbial biomass is generally higher in an AFS, compared to a conventional system producing the same crop. Higher stability of soil aggregates is reported in comparisons between AFSs and treeless systems (Udawatta et al., 2008), while other soil characteristics, such as soil porosity (Seobi et al., 2005) and water infiltration (Bharati et al., 2002) can also be enhanced with AF.
- Conservation of biodiversity by habitat creation: AFSs will generally have a more diverse spatial arrangement when compared to similar systems without woody species. Furthermore, the higher number of cultivated species in certain AFSs (e.g., multistrata cacao systems), increases the potential of the growing systems to host and feed diverse forms of wildlife (Harvey and Villalobos, 2007). Cohesive patches of trees and soil that may be left undisturbed to a higher degree under trees enhance the connection between nearby habitats, which may exist scattered across the landscape (Johnson and Beck, 1988). This is specifically relevant to ACSs (Best et al., 1990; Rosenberg et al., 1999), because long, cohesive rows of trees and undisturbed soil may act as corridors for wildlife. Finally, AFSs may decrease nutrient leaching and sediment loss from growing systems, because the tree elements have deep, long- lived roots that can hold the soil and take up nutrients that may have been leached to a lower layer in the soil profile. This generally has a beneficial effect on surrounding natural habitats, which might otherwise suffer from eutrophication and sedimentation of water bodies and streams (Nahayo et al., 2019).

The benefits mentioned are generally cited to be found in AFSs, not specifically in alley cropping systems. However, these benefits are likely to occur also in ACSs. Albeit, depending on the tree planting density and pattern of the ACS in question, the effect on soil fertility may not be substantial, especially in situations where a large portion of the crop fields are far away from any trees.

Increased system stability

Growing multiple types of crops (e.g., arable crops and trees) instead of just one type is an example of income diversification. Income diversification is a widely applied strategy used to protect against economic shocks and market instability (Krishnamurthy and Reddiar, 2011; Sultana et al., 2015). The strategy is applied due to the belief that the combined income from two different enterprises will be more stable than the income from any of the two alone. AF often implies income diversification, because a wider variety of plants and animals is produced.

AF practices, such as ACSs, can effectively and significantly decrease soil water erosion by providing higher water infiltration rates and slowing the movement of run-off water across the ground (Muchane et al., 2020). Furthermore, wind erosion of soil can be ameliorated in AF because trees reduce wind speeds (Kay et al., 2019; Sanchez and McCollin, 2015). For these reasons, AFSs are generally more resistant to extreme weather events with high levels of precipitation and high wind speeds. Resistance to soil erosion also benefits soil fertility, compounding the effects noted above and additionally increasing system stability.

Increased yields and decreased need for external inputs

In the right circumstances, higher total yields are attainable by integrating multiple species than by growing separate monocultures of the same species on their own. The term ‘complementarity of resource capture' is used to infer that there might be less competition between two plants of different species than two plants of the same species. Growing nitrogen-fixing plant species in AFSs is one obvious example where higher overall yields can be obtained (Mortimer et al., 2015), and dependence on external inputs is lowered (Seiter et al., 1995). Furthermore, different plant species can support the productive capabilities of each other in other ways. E.g., the provision of shelter for crops by reducing wind speeds in and in the vicinity of growing systems is an important potential benefit of ACSs, depending on the appropriate layout of the growing system (Jose, 2009).

Reports of higher total yields in AFSs abound, when the yields of all useful outputs (multiple plant or livestock derived materials) are considered and combined. It is not straightforward to compare monoculture yields with those derived from an AFS, as the latter will have at least two different kinds of yields. The concept of the Land Equivalence Ratio (LER) is used in such comparisons (Mead and Willey, 1980). This figure represents the yield of a given AFS divided by the yields of monocultures of the same component plants:

Abbildung in dieser Leseprobe nicht enthalten

For instance, in an alley cropping system of wheat and poplar trees, yield A would be wheat yield and yield B would be poplar yield. A figure of less than 1 means that the AFS is less area effective than corresponding monoculture systems, while a figure of more than 1 signifies that the AFS is more effective. For a range of ACSs, values of higher than 1 have been documented (de Jalon et al., 2018; Graves et al., 2007; Sereke et al., 2015; Xu et al., 2018).

1.1.3 Agroforestry in contrasting climates

The practice and science of AF have mainly been developed and successfully implemented in tropical areas. Here, the benefits reaped have been manifold (Harvey and Villalobos, 2007; Paudel et al., 2012; Rahman et al., 2012).

When it comes to temperate climates, by and large, AF as a science and a practice has not seen much use, albeit vast potentials are reported to exist in temperate climate contexts as well (Kay et al., 2019). There has recently been much interest in AF in general, including in temperate and cool temperate climates. The development of, and interest in, the field seem to be progressively increasing. In warm temperate areas, such as the Mediterranean region, examples of the successful implementation of AF are plentiful (Mosquera-Losada et al., 2012). However, in cool temperate climates, such as Northern Europe, much more knowledge and know-how is still lacking if innovative agroforestry practices such as ACSs are to become mainstream.

In some coastal regions of Canada (such as Québec), agroforestry has been trialled and researched to some extent. Some findings made in these areas may be of relevance to Northern Europe, but the projects and efforts made in these areas are relatively young (Anel et al., 2017). In temperate climate regions, ACSs are one of the types of agroforestry that significant amounts of research effort have gone into, and significant potentials could be realized if ACSs saw broader adoption here (Wolz and DeLucia, 2018).

The question arises whether many of the reported benefits of agroforestry practices are irrelevant in cool, temperate climates. Table 1.1 gives a brief comparison of benefits from agroforestry practices in contrasting climates. This makes it evident that many of the benefits of ACSs seem very relevant in cool, temperate climates as well, and specifically in Northern Europe.

Table 1.1: Overview of alleged benefits of AF practices, the kind ofpractice that can bring them about, and an indication of whether or not this likely happens in tropical and cool temperate climates.

Abbildung in dieser Leseprobe nicht enthalten

1.2 Simulation of alley cropping systems

If AF practices, such as ACSs, are both ecologically desirable and economically sound in Northern Europe, why is the implementation of these not happening at a higher speed in this region? Undoubtedly, one reason is that existing and previous examples of, e.g., alley cropping systems, are not common here. For this reason, yield performance of the systems and the amounts of inputs, such as labour, demanded by a given system for implementation, running and maintenance cannot be quantified with any high degree of certainty.

Thus, the application of ACSs in Northern Europe is caught in a vicious spiral: Few Northern European ACSs exist today. Therefore, the performance and requirements of these systems cannot be reliably quantified. Therefore, there is not much general knowledge of the performance and requirements of these systems. Therefore, the design of ACSs cannot happen on an informed basis, and the produced designs will lack credibility by the farmers and land managers that are to manage and maintain them. Therefore, few new ACSs are established. Therefore, the density of Northern European ACSs will continue to be low, and little knowledge of the systems will be produced.

One conceivable way of overcoming this challenge might be through the development of effective modelling software for use in the Northern European context. This software would be able to predict properties of projected agroforestry systems, such as crop and wood yields. It could be used both to assess how appropriate the adoption of AF would be and to guide in the process of growing system design. In these ways, models could give us more realistic expectations of ACSs, as well as the estimates required for it to be attractive to establish them.

1.2.1 Descriptions of the models used

Hi-sAFe

The objective of the Hi-sAFe (HS) model design was to combine the two existing French modelling programs Capsis and STICS to create a mechanistic model for the specific purpose of modelling temperate, and specifically European, ACSs (Dupraz et al., 2005). The aim of a mechanistic model was established to enable studying the processes at work in ACSs, and to increase the scope for using the model outside the area of its initial development, which was the Mediterranean region (Dupraz et al., 2019). The focus on European ACSs is what makes this model particularly interesting to this study.

The HS model was initially developed by a team at the French INRA facility during the SAFE project of the European Union running from 2001 till 2005 (Dupraz et al., 2005). However, the model is still under development, and the current model version may be perceived as a foundation to build on.

One key trait of HS, setting it apart from other models, is its capacity for examining three-dimensional traits of the growing system. The model traces the trajectory of sunbeams, accurately simulating their descent upon the canopies of trees and understory. It allows multiple trees of any combination of species to be placed in any given pattern on the scene of simulation, and it allows the ground cover in the tree rows to be planted to another crop than the main crop. Finally, the STICS software principally allows the simulation of crop mixtures (intercropping) (Brisson et al., 2008), albeit this functionality has not been included for use in the HS main crop yet. The model can technically run simulations of indefinitely long periods, and any arbitrary crop rotation. Thus, HS is technically an extremely flexible modelling program.

Yield-safe

The yield-safe (YS) model was originally developed as a spin-off model of the HS project (Dupraz et al., 2005; van der Werf et al., 2007). Because the development of the HS model had taken longer than initially anticipated, there was no final product to present at the end of the EU funded AGFORWARD program 2002-2006 that funded the development project. Partly for this reason, a simpler, minimum viable product-style formulation of the model was developed, and eventually named yield-safe (van der Werf et al., 2007). YS shares several traits with HS. They both run with a resolution of daily time steps, and the basic framework is set up for the simulation of AFSs. However, the simplicity of yield-safe lies in the fact that it to a much greater extent is empirically based; it does not consider a ”scene” in the sense of a space with dimensions, as opposed to the three-dimensional structure of HS; and it is not intended to enable the user to make mechanistic simulations of ACSs.

YS is also a lot simpler than HS in the parameters it uses. YS was built as a parameter-sparse model (van der Werf et al., 2007), and thus it only requires the setting of 24 parameter values to be able to run (Palma et al., 2010).

The YS model has been used quite a lot since its development, in a wide variety of countries, including Northern European ones such as the UK (Burgess et al., 2012), Denmark (Xu et al., 2019) and Germany (Seserman et al., 2019).

YS is interesting to study because it, like HS, is oriented towards modelling of AFS and developed in a European context, and yet is easier to parameterize and use.

1.3 Research question:

The objective of the study is to compare the performance of the HS and YS models when simulating productivity in Northern European alley cropping systems.

2 Materials and method

In this chapter, the criteria for the selection of models and modelled growing systems are listed, and the growing systems are presented. Then, the general procedure used in the modelling effort is outlined, and the statistics used for assessing the results are described. Finally, the creation and parameterisation of models of all growing systems with HS and YS is described.

2.1 Selection of the models

The models chosen for the study were chosen for the following reasons:

- Because they were specifically designed to model agroforestry systems
- Because they were both relatively recently developed, and are both still undergoing development
- Because they both support simulation of the impacts imposed by differences in latitude.

2.2 The modelled growing systems

Two sets of growing systems were selected for the study. The MAFF-sponsored trials comprised three English alley cropping trials, and the CFEC is located in Denmark. The two sets, and the types of data used from each system, are presented below.

2.2.1 Selection of the growing systems

In the search for useable agroforestry trials that could be simulated, the following criteria were observed in order to evaluate their appropriateness:

- Geography. The studies had to have been done in coastal Northern Europe.
- Data availability. Proper data on yield and tree growth had to be attainable, as well as data on weather, soil and management practices.
- Continuity. Studies that had been running sufficiently long for the tree element to mature or have been harvested so that the interaction of the tree and crop elements would be evident.
- Credibility. The trials had to have been described and written about in comprehensive scientific texts and articles.
- Diversity. It was attempted to find growing systems that were in some ways structurally different so that the tested modelling programs could be tested in different circumstances.

2.2.2 MAFF-sponsored alley cropping trials

In the UK, three agroforestry trials were established in 1992 in locations scattered across England with sponsorship from the British Ministry of Agriculture, Food and Fisheries (MAFF). The trials were placed near Leeds in the north of England (“Leeds trial”), near Gloucester in south-west England (“Cirencester trial”), and near Bedford in central England (“Silsoe trial”). The trees were grown to be tall trees for timber and other non-energy purposes. Figure 2.1 presents the general layout of the trials. Tree rows were planted 10 m apart, with 1 m uncropped strips between the trees and the crops. The ground was mulched with plastic mulch in 0.5 m wide bands to each side of the tree trunks. In the next 0.5 m of the uncropped strip, weeds were regularly mown. In the tree row, the distance between the trees was 6.4 m (Burgess et al., 2005).

Abbildung in dieser Leseprobe nicht enthalten

Figure 2.2: Schematic of the CFEC trial design. Modified from Ghaley and Porter (2014).

Hojbakkegaard, will henceforth be referred to with the abbreviation CFEC (“CFE, Copenhagen”). It was established in 1995 as a short rotation coppice (SRC) system combined with the growing of a grain and grass rotation. The system has been managed organically for almost twenty years, the only significant source of nitrogen since 2012 being atmospheric N fixation in crops and trees. Specifically, one in five trees is a nitrogen-fixing alder tree (Alnus rubra), which sheds its nitrogen-containing leaves every autumn, and much of the crop rotation is clover-ley that is cut thrice per year, probably resulting in large amounts of root turnover and root exudates stimulating the soil ecosystem. The crop rotation includes winter and spring grains as well as grass-legume mixtures. The layout of the whole experiment is illustrated in figure 2.2.

The system is designed to test the significance of alley width, as it includes four different “alleys” or fields of the same lengths but different widths, bordered on the east and west sides by equal-sized biofuel hedges. The four alley widths are 50 m, 100 m, 150 m and 200 m, making up four subsystems with different distances between hedges.

Information on the CFE system on Hqjbakkegârd in Denmark was gathered by direct inspection, by conversations with the researchers studying the growing system and the technical personnel responsible for the direct management of the system, by examining and using figures from the many recent studies done on the system, and by drawing from the archive data of the university farm. An estimate of the standing biomass in the CFE system was made based on the method described in Ghaley and Porter (2014).

Data for all years from 2000 till 2019, including both, were available, so this period was simulated. Data on yields of agricultural crops ([t/ha]) and of wood ([t dry matter harvested]) were available. The latter figures were converted into figures of harvested dry matter per tree ([kg DM/tree]), and predictions of these two variables were produced with the models and compared to the measurements. The figures of harvested dry matter per tree were produced in this way:

Abbildung in dieser Leseprobe nicht enthalten

where mtree is the mass of dry matter per tree, mtotai is the mass of dry matter from all biofuel hedges, and n is the total number of trees in all hedges. As there were six hedges with 2,000 trees in each, n was assigned the value 12,000 trees.

2.3 Modelling principles and procedure

In this section, the general procedure of parameterisation and calibration followed is outlined, the sources and processing of the weather data used are mentioned, and the way statistics were calculated is described.

2.3.1 Parameterisation, calibration and validation of the models

The overall process of model creation, parameterisation, calibration and validation is illustrated in figure 2.3 . The MAFF-sponsored trials encompassed three distinct growing systems, and the CFEC consisted of 4 different subsystems with differences in structure. Therefore, seven distinct models were created with each of HS and YS for a total of 14 different growing system models (GSMs). Each of these reflected a distinct growing system.

Abbildung in dieser Leseprobe nicht enthalten

Since with both models, at least one of the studied growing systems had never been modelled before, new input parameters for crops and trees had to be made.

1. In the model creation process, first parameters defining the given growing system were fixed. This included parameters such as distance between tree rows and latitude.
2. Then, during scene calibration, unknown values that could be set by expert opinion were fixed (soil pH and management parameters).
3. and 4. In the process of plant calibration, lower and upper bounds were set by expert opinion for crop and tree parameters, such as frost tolerance and number of grains per spike, and these parameters were then calibrated. For details about the phenological parameterisation, see Appendix B.
5. The GSMs underwent final calibration. During this process, critical plant growth parameters, such as N demand and light use efficiency, underwent calibration using subsets of the data available. These subsets were not used in the later validation process. The calibration data subsets were:

- For the MAFF-sponsored trials: The entire Cirencester trial dataset, years 1992-2002. This was appropriate because data was available for three structurally identical systems.
- For the CFEC: Data from the years 2012-2020 in all four growing system parts, 50 m, 100 m, 150 m and 200 m biofuel hedge distance. Yield measurements of the hedges were used for 2012 and 2016. Splitting the dataset for the total period in two parts resembles the procedure used in other studies (Seserman et al., 2018; Fahad et al., 2019).

The calibration process ended with a comparison between the output of the simulation in question with regards to crop yields and tree growth, and corresponding reported figures from the data in the corresponding years. Three points were observed to decide when the calibration was complete:

- Whether the simulated and measured figures concurred.
- Whether the statistical figures calculated to compare simulated and measured values indicated a good correlation and low bias.
- Whether other significant unsatisfactory model behaviour (unrealistic hydrological conditions or the like) occurred during the simulation runs.

When it appeared that no further improvements could be made to improve on the three points mentioned, model calibration was considered complete. 6. Finally, during model validation, the validity of the models was evaluated by running simulations of the growing systems or years not used for calibration. For the MAFF- sponsored systems, these were the datasets from the Leeds and Silsoe trials, and for the CFEC, it was the dataset for the period 2000-2011. The output figures of crop yields and tree growth were examined and compared to the reported corresponding figures both graphically and by calculating a variety of statistical figures, see section 2.3.3 (Smith et al., 1996).

2.3.2 Acquisition of weather data

Weather data for the appropriate locations for the period from 1992-01-01 until 2002-12-31 was obtained from the Met Office weather service of the UK for the MAFF-sponsored trials. Likewise, appropriate weather data for the period 2000-01-01 till 2019-12-31 was obtained from Copenhagen University’s weather station and the Danish state weather service, DMI for the CFEC growing systems. In the few cases of lacking entries in the data, simulated weather data was supplemented for completion using the CliPick web climate picker service (Palma, 2017). For further details, see Appendix D.

2.3.3 Evaluating the simulation outputs

Output from the validation runs was evaluated both by graphical evaluation, and by statistical analysis. Specifically, it was assessed whether the predicted and measured figured concurred, whether there was good correlation between predicted and measured figures, whether the predictions were biased, and whether the variances of the predicted and measured datasets were similar. Furthermore, the sizes of the errors were assessed. The statistical tests used are described below.

Statistical analysis

The quality of the simulation outputs was evaluated in terms of agreement with measured data using four different statistics: The coefficient of determination (r2), the Root Mean Square Error (RMSE), a paired F-test for significance of association between datasets, and a paired t-test for bias in the predictions. r2 of the relation between predicted and measured datasets was calculated as the square of Pearson’s correlation coefficient (p) using the ‘stats’ -package in the R software for statistical programming (R Core Team, 2019).

Non-normalised RMSE was calculated as:

Abbildung in dieser Leseprobe nicht enthalten

To evaluate whether there was any statistically significant association between measured and predicted values, an F-test was performed. The F value was calculated as:

Abbildung in dieser Leseprobe nicht enthalten

This formula follows the method set forth in Smith et al. (1996). This figure was compared to the F - distribution, which was identical to the t2 -distribution since the comparison was bivariate. The inverse t2 -distribution was used to determine the number of standard deviations away from the mean corresponding to 95% probability. F values exceeding the 95% probability quantile of this distribution were taken to indicate a significant association between predicted and measured values (Demidenko, 2019).

To assess the bias in the predictions, a two-sided paired Student's t-test was performed with the ‘stats' -package in the R software for statistical programming. This was compared to a two-tailed Student t-distribution with n - 2 degrees of freedom. A value of t greater than the probability quantiles corresponding to 97.5% probability in the two-tailed t -distribution was assumed to indicate a statistically significant bias in the predictions (Smith et al., 1997).

2.4 Use of Hi-sAFe

In this section, only the most significant assumptions of the HS parameterisations are described, as well as the aspects of growing systems that could not be represented in the GSMs in a direct manner. The section discusses the MAFF-sponsored trials first, and the CFEC systems second.

Abbildung in dieser Leseprobe nicht enthalten

Table 2.1 and table 2.2 show descriptions of the inputs to HS of the parameters that express the key structural features of the MAFF-sponsored trials and the CFEC, respectively. For the specific parameterisations of crops and trees, see Appendix A. For details on the parameterisation of crop growth phenology, see Appendix B. For a presentation of the crop rotations grown in the real systems, and rotations simulated, see Appendix C. For an in-depth description of parameterisations of soil, management, and other system traits, see Appendix D.

For the MAFF-sponsored trials, the crop types were split into winter wheat, winter barley and spring grain. The spring grain parameterisation was created as a compromise between spring barley and - wheat because there were too few data points of each of the individual spring grain species to calibrate.

HS predicts tree biomass values in terms of kg C in the tissues in question. Therefore, predictions of CFEC tree mass had to be converted into kg dry matter per tree. This was done by multiplying values of kg C per tree by a factor of 2 because the ratio of dry matter biomass to mass of C was assumed to be 2 in the woody tissues (Zhang et al., 2009).

Initially, it was planned to model all four subsystems of the CFEC (50 m to 200 m between hedges), but the simulation of the 200 m system demanded too much memory from the computer utilized, so running the simulation proved to be technically impossible.

HS did not support the simulation of coppice, so the CFEC growing systems were simulated in multiple simulations lasting four years each, corresponding with the periods between two harvests in the real systems (e.g., March 2000 till March 2004).

Table 2.1: Parameterisation . for HS of the shared, system-defining parameters of the MAFF-sponsored trials.

Abbildung in dieser Leseprobe nicht enthalten

Table 2.2: Parameterisation . for HS of the system-defining parameters of the CFEC.

Abbildung in dieser Leseprobe nicht enthalten

2.5 Use of yield-safe

In this section, only the most significant assumptions of the YS parameterisations are described, as well as the aspects of growing systems that could not be represented in the GSMs in a direct manner. The presentation discusses the MAFF-sponsored trials first, and the CFEC systems second.

Abbildung in dieser Leseprobe nicht enthalten

Table 2.3 and table 2.4 present descriptions of the inputs to YS of the defining input parameters that express the key structural features of the MAFF-sponsored trials and the CFEC, respectively. For the specific parameterisations of crops and trees, see Appendix A. For details on the parameterisation of crop growth phenology, see Appendix B. For a presentation of the crop rotations grown in the real systems, and rotations simulated, see Appendix C. For an in-depth description of parameterisations of soil, management, and other system traits, see Appendix D.

Several input parameters of YS are generalisations of some detailed features. For instance, soil texture can be parameterized using one of several premade compositions ranging from very coarse to very fine soil. Where appropriate and possible, one of the premade parameterisations where chosen, rather than going through the process of creating a new parameterisation from scratch. This was done to test an alleged strength of yield-safe, namely its ease of use.

YS runs on daily inputs of not only minimum and maximum temperature, average wind speed, total insolation and precipitation, but also on daily figures on evapotranspiration and average relative humidity. For the MAFF-sponsored trials, where data on average humidity was not available, this figure was produced as the average of daily minimum and maximum relative humidity. Evapotranspiration was calculated using the modified Makkink method (Hansen, 1984). All the needed values for climatic constants (such as the psychrometric constant) were calculated following the procedures outlined in Allen et al. (1998).

For the MAFF-sponsored trials, the crop types were split into winter grain and spring grain. The spring grain parameterisation was created as a compromise between spring barley and -wheat, while the winter grain parameterisation was created as a compromise between winter barley and -wheat.

YS did not support the simulation of coppice, so the CFEC growing systems were simulated in multiple simulations lasting four years each, corresponding with the periods between two harvests in the real systems (e.g., March 2000 till March 2004).

Table 2.3: Parameterisation for YS of the shared, system-defining parameters of the MAFF-sponsored trials.

Abbildung in dieser Leseprobe nicht enthalten

Table 2.4: Parameterisation for YS of the system-defining parameters of the CFEC.

Abbildung in dieser Leseprobe nicht enthalten

3 Results

This chapter presents the results of the simulations. First, the outputs from the final calibrations are shown for the MAFF-sponsored trials and the CFEC, respectively. Then outputs are shown from the validation runs in the GSMs of the MAFF-sponsored trials and the CFEC, respectively. Values of r2 and RMSE are given in the figures. Furthermore, every simulation run is accompanied by a brief graphical evaluation of outputs as well as results of the statistical tests for association between datasets and bias.

F value comparisons are denoted with the syntax F = X / Y Here, X is the F value calculated from the p of the comparison between the predicted and measured values in question, and Y is the value corresponding to the 95% probability quantile in the F-distribution for the comparison. As noted in section 2.3.3, F values higher than the value corresponding to the 95% probability interval indicate a statistically significant association between predicted and measured data. Likewise, for the bias-tests, a t-value less than the value corresponding two-tailed 97.5% (effectively 95%) probability indicates no statistically significant bias. F and t comparisons are given only in cases where compared values are relatively close, indicating a narrowly significant or not significant result.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.1: Graphs comparing real measurements and simulated figures from the HS calibration trial dataset. a: crop yield. b: tree height. c: tree DBH. For tree height and DBH, measurements were done in late autumn. Black line: measurements, red line: simulated figures of heights and DBH. - = measurements, O = simulated yield figures.

3.1 Final calibrations

Figure 3.1 presents a comparison of the outputs from the final calibration run in the Cirencester GSM and the measured values from the same growing system. The crop element was challenging to calibrate satisfactorily, due in part to the range of crop species and types involved (including winter- and spring types of both wheat and barley). As is evident in figure 3.1a , even though some substantial discrepancies still existed after the final calibration run, results were somewhat satisfactory. The crop yields failed the association test, albeit narrowly (F = 5.27/5.99) and were not biased. As shown in figure 3.1b and c , it was not difficult to calibrate the tree element satisfactorily, and the association and bias tests indicated significant association and no bias.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.2 shows a comparison between outputs from the final calibration run for the YS MAFF Cirencester GSM and the corresponding measured values. Calibration of both crops and trees in the MAFF systems with YS was generally unproblematic, and a satisfactory result was attained. It is evident from the graphs that there were good correlations, and that the shapes of the measured datasets were correctly reflected. Furthermore, all three comparisons indicated significant associations and an indication of no bias was given for both crop yields and tree diameter. The values did indicate a significant bias for the tree heights. Figure 3.2b shows that it is a negative bias. However, this bias was very small.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.2: Graphs comparing real measurements and simulated figures from the YS calibration Cirencester dataset. a: crop yield. b: tree height. c: tree DBH. For tree height and DBH, measurements were done in late autumn. Black line: measurements, red line: simulated figures of heights and DBH. - = measurements, O = simulated yield figures.

Figure 3.3 shows the results of the calibration of the HS CFEC GSMs. As noted in section 2.4, the 200 m distance system could not be simulated. Crop calibration in the model was generally successful, but it was quite difficult to attain crop files that reacted realistically to varying growth conditions, such as drought or warm weather. Statistics indicated significant association and no significant bias.

Calibration of the tree element, on the other hand, proved to be difficult, since, in the 100 m GSM, as opposed to the two other GSMs (50 m and 150 m), significantly higher wood yields were predicted by the model in one of the calibration dataset periods (2012-2016), which seems unrealistic. Statistics indicated no significant association (F = 6.28 / 7.71), an no significant bias.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.3: a: Graph comparing real measurements and simulated figures from HS of crop yields across the 50 m, 100 m, and 150 m hedge distance growing systems, during the calibration dataset years of 2012-2019. A varying number of observations were available for each year, which explains the varying number of measured and simulated values in each year. - = measurements, O = simulated yield figures. b: Comparison of measured woodyields and predictions of wood yields for all three HS CFEC GSMs. The numbers for ratios of simulated:measured values are inset so that their height along the y-axis correspond with the ranking of the wood yield sizes in each GSM.

Figure 3.4 presents comparisons of predicted and measured values for the YS CFEC GSMs. The predicted crop yields showed no bias, but no significant association (F = 1.52 / 4.75). From a graphical evaluation, the predicted values seem to reflect the patterns in the measurements well, so calibration of crops in the CFEC system with yield-safe went reasonably well overall. The predicted tree growth values were significantly associated with the measurements and had no bias. They were thus successfully calibrated.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.4: a: Graph comparing real measurements and simulated figures from YS of crop yields across the 50 m, 100 m, and 150 m hedge distance growing systems, during the calibration dataset years of 2012-2019. A varying number of observations were available for each year, which explains the varying number of measured and simulated values in each year. - = measurements, O = simulated yield figures. b: Comparison of measured wood yields and predictions of wood yields for all four YS CFEC GSMs. The numbers for ratios of simulated:measured values are inset so that their height along the y-axis correspond with the ranking of the wood yield sizes in each GSM.

3.2 Validation runs

3.2.1 Validation of the Hi-sAFe MAFF-sponsored growing system models

Figure 3.5 shows a comparison between predictions and observations for the Silsoe and Leeds trials. Prediction of the crop yields was somewhat successful. In Silsoe, the correlation between simulated and measured values was significant and had no significant bias. In the Leeds trial, the figures indicated no significant correlation and significant bias. The simulated values did, to some extent, share the shape of the measured values dataset (figure 3.5a ), but in some years, the differences between simulated and measured values were considerable. Simulation of tree growth in the Leeds trial was overall successful, and the statistics indicated significant association and no bias for both height and DBH. In the Silsoe trial, large discrepancy existed between simulated and measured values. No association was found, but bias was indicated. This was probably a general failure of simulation, because it was not easily traceable to any of the inputs to the model.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.5: Graphs comparing real measurements and HS simulated figures from the validation Leeds (a, b, c) and Silsoe (d, e, f) datasets. a, d: crop yield. b, e: tree height. c, f: tree DBH. For tree height and DBH, measurements were done in late autumn. Black line: measurements, red line: simulated figures of heights and DBH. - = measurements, O = simulated yield figures. Numbers close to lines and points: ratio of simulated value to measured value in the given year.

Figure 3.6 shows comparisons of YS simulated and measured values for the Leeds and Silsoe GSMs. The simulation of the crop yields was somewhat successful. In Leeds, the F-test showed that there was no significant association between predicted and measured data, and the t-test indicated that the predictions were biased. However, judging from a graphical comparison, the shapes of the simulated and measured datasets agreed reasonably well (figure 3.6a ). Association between values were not significant in Silsoe but evaluated graphically, the correlation seemed somewhat reasonable (figure 3.6d ). Albeit, in both growing systems, the predicted values displayed negative bias. The cases of crop failure or near-zero yield in either predictions or measurements probably had a considerable negative effect on RMSE as well as the other statistical figures, making these rather useless. For instance, the t-test of the Silsoe crop yields indicated no bias, which conflicted with the impression obtained from the graphical evaluation.

The simulation of the growth of the trees was generally successful, considering shapes of curves, as well as figures for correlation and bias. Statistics indicated significant associations between values for tree heights and DBH in both the Leeds and Silsoe GSMs. The model was not able to predict the acceleration of growth in tree DBH in the Leeds trial, which produced the slightly s-shaped curve of the measured tree DBH figures (figure 3.6c). For this reason, the latter half of the years in the period in question showed too low predictions of tree DBH. Moreover, there did seem to be some negative bias in the model predictions of tree growth in the Silsoe trial (figures 3.6e and f). However, only the discrepancy in the DBH values in the Leeds trial amounted to statistically significant bias.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.6: Graphs comparing real measurements and YS simulated figures from the validation Leeds (a, b, c) and Silsoe (d, e, f) datasets. a, d: crop yield. b, e: tree height. c, f: tree DBH. For tree height and DBH, measurements were done in late autumn. Black line: measurements, red line: simulated figures of heights and DBH. - = measurements, O = simulated yield figures. Numbers close to lines and points: Ratio of simulated value to measured value in the given year.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.7 shows comparisons of the outputs from HS and real measurements from the growing systems. It looked like the crop yield predictions of the model missed some critical source of variability in the real growing system. The pattern in the dataset of measurements was not reflected at all in the predicted values. Statistics indicated no significant bias and no significant association between predicted and measured datasets. As mentioned in section 2.4, the system with 200 m distance between biofuel hedges could not be simulated.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.7: Graphs comparing real measurements and simulated crop yield figures from HS from the CFEC growing systems of a: 50 m between hedges, b: 100 m between hedges, and c: 150 m between hedges. - = measurements, O = simulated yield figures. Topmost numbers in each graph: ratio of simulated value to measured value in the given year.

Figure 3.8 presents the wood yield predictions of the model, along with the measured values. It was evident that the simulation of the tree element of CFEC in HS only saw marginal success. This may be due to the potential defect mentioned above in the discussion on calibration of the CFEC model (section 3.1.3). The predictions for wood yield in 2004 were very satisfactory. However, when compared across all the different periods, the predicted yields were generally too stable. Statistics indicated bias and no significant association. There must have been some element in the real system that was not adequately represented in the model since this was not sensitive enough to the weather differences between the periods.

Furthermore, the predicted values did not reflect the pattern of the dataset of measured values. This was evident when considering the ranking of the values between the different harvest years. In the measured values, that ranking would be 2004 = lowest, 2008 = highest and 2020 = middle, while when considering the mean of the predictions for each separate harvest year, this would be 2004 = highest, 2008 = middle and 2020 = lowest.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.8: Comparison of measured wood yields and predictions of wood yields for all three HS CFEC GSMs. The numbers for ratios of simulated:measured values are inset so that their height along the y-axis correspond with the ranking of the wood yield sizes in each GSM.

3.2.4 Validation of the yield-safe CFEC growing system models

Figure3.9 shows the predicted yield values from the YS CFEC GSMs and the measured yields. Despite the relatively high RMSE-value and the fact that the statistics indicated no significant association, it could be argued that YS predicted crop yields with some success in the CFEC system. There was no significant bias in the predictions and viewed at a glance, it did not look like there were any overall tendencies in the predictions that set them apart from the measured values. Albeit, particular discrepancies could be emphasized, such as the prediction of much too high yields in all GSMs in 2009.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.10 compares predictions from the YS CFEC GSMs and measurements. Simulation of tree growth was generally unsuccessful. Statistics indicated significant bias and no significant association between datasets. The yield predictions were too stable, so there must have been some elements in the real system that the model did not adequately consider. Furthermore, the pattern in the predicted and measured datasets were not alike; the ranking of the measurements of the three years respectively (2004 = lowest, 2008 = highest, 2020 = middle) was not the same as the ranking of the means of predictions in those three years (2004 = middle, 2008 = lowest, 2020 = highest).

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.9: Graphs comparing real measurements and simulated crop yield figures from YS from the CFEC growing systems of a: 50 m between hedges, b: 100 m between hedges, c: 150 m between hedges, and d: 200 m between hedges. - = measurements, O = simulated yield figures. Topmost numbers in each graph: ratio of simulated value to measured value in the given year.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.10: Comparison of measured wood yields and predictions of wood yields for all four YS CFEC GSMs. The numbers for ratios of simulated:measured values are inset so that their height along the y-axis correspond with the ranking of the wood yield sizes in each GSM.

4 Discussion

In this chapter, the performance and other aspects of the models are compared. After this, the advantages and disadvantages of different elements of model design are discussed, and elements which can be perceived as breakthroughs are highlighted. The model comparison concludes with a summary of strengths, weaknesses, and suggestions for further development of the models. Then, the limitations of this study and sources of error are emphasized. Finally, the role of models in the development of alley cropping in Northern Europe is discussed and suggestions for this development are given, emphasizing the need for more studies of the critical interactions and traits of ACSs.

4.1 Performance of the models

In this section, first, the performance of the models is discussed. The association between predicted and measured values and prediction accuracy in the simulations run in the study are evaluated. This is followed by a presentation of results with the models in other studies. Other model traits, such as flexibility, ease of use, and appropriateness for transdisciplinary studies, are evaluated. Finally, an overall comparison of the usefulness of the models is made.

4.1.1 Correlation and accuracy in comparison with measured data

Table 4.1 presents a summary of how successful predictions of different system elements were with the models. The successfulness of predictions was determined with the method described in section 2.3.3.

Abbildung in dieser Leseprobe nicht enthalten

Table 4.1: Summary of evaluations of the results attained. Values in parentheses are RMSE values for the element in question. Colour indicates simulation type (yellow = calibration, blue = validation). Colour saturation (light, moderate or dense) of the cell indicates an overall rating of the results.

Abbildung in dieser Leseprobe nicht enthalten

Accuracy of Hi-sAFe predictions

With the HS model, it was generally possible to simulate the MAFF growing systems reasonably well and the CFEC growing systems less well. In the MAFF systems, the predictions of the HS models showed some correlation with the observed values. Several of the patterns in the data were reflected, albeit perhaps not all. One problem encountered during calibration runs in the Cirencester GSM was that the crops in tree shade did not experience lowered temperatures during flowering. The temperature during flowering is a key determinant of yield size in the crop simulation module of HS (STICS). Therefore, HS would predict too high yields in the late years of the period, when the trees had grown tall and shaded the crop a lot. It was necessary to lower the growth efficiency ([g biomass / MJ radiation]) of the crops, so the predicted and measured data at the end of the period would agree better. This removed a positive bias in yield predictions when assessed over the whole period (1992­2002); however, it instated a negative bias in the start of the period and is likely also part of the reason for the negative bias in the validation runs. The team of developers has confirmed that the model has this problem (C. Dupraz, 2020, personal communication, April 30th).

The HS based models of the CFEC generally failed to reflect the variability evident in the real growing system. The yield predictions were evidently too stable during the simulated period. When the crop yields across the entire period were plotted on a graph, it was evident that the variation in the yields were lower in the calibration period (2012-2019) than in the validation period (2000-2011), see figure 4.1 . The reason for this may be the changes in management carried through around 2012, including the cessation of all inputs of N. Albeit, these inputs were small in the former period. However, after the unsatisfactory results of low variance was acknowledged, attempts were made to parameterise crop files in ways that would create a better reflection of the variance found in the real measurements. Despite substantial effort, this was not achieved, and the variance was continually very low. It could be that the large variance found in the crop yields of the CFEC validation dataset was very uncommon, and thus generally unpredictable. However, this is refuted by the success of YS to produce crop yield predictions with the appropriate level of variance. Summing up, it seems that there is no good reason for the failure to reflect the variance evident in the yields.

Abbildung in dieser Leseprobe nicht enthalten

Figure 4.1: Measured crop yields for wheat, barley and oat in all four CFEC subsystems in the period 2000-2019. Varying numbers of crops and subsystems were measured in each year, which explains the differing number ofpoints for each harvest year.

Neither was there much success in the predictions of tree yields in the CFEC growing systems with HS. Part of the error appeared already in the calibrations, where a considerable difference between wood yields in the 100 m system and wood yields in the other two simulated systems was predicted. Despite efforts, this problem was not amended by calibration. It cannot be verified if it indeed was the case in the real growing system that the hedges bordering the 100 m system yielded significantly more than the others because only data of the total yield from all the biofuel hedges are available. However, there does not seem to be any apparent reason why these hedges would be more productive, and the differences in predicted yield look like a defect in the model.

The appearance of the problems mentioned above is perhaps not surprising, considering that the HS model was not developed to simulate this kind of SRC systems. The coppicing of the trees every fourth year, as well as the high planting density of the trees in the biofuel hedges, makes for very different growing conditions and plant interactions than the tree rows that the model was initially designed to simulate. The model does not have appropriate concepts to represent the coppiced trees, and these were simulated as single, slender, low stems with a large proportion of branches, in order to create a model that structurally resembled the real system.

The predictions of the tree yields in the CFEC systems also suffered from the difficulty to calibrate the files for multiple different GSMs simultaneously, which is also revealed by the relatively low accuracy in the calibrations of the wood yields in the HS CFEC models.

Also, in the HS MAFF GSMs, a severe error was evident in the predictions. In the Silsoe trial, predictions of tree heights and DBH showed a marked decrease in growth rate from the fourth year onwards, which was not found in the real measurements. No explanation can be given for this, and it looks like a model defect. Of course, the smaller trees simulated probably also impacted on the crops, which could partly explain the excessively high yields predicted for 1999 and 2002.

To sum up, it seems that the GSMs produced with HS in this study have suffered from several technical bugs. These may have emerged due to bugs in HS, or because of parameterisation errors made by the modeller. Either case could be used as an argument to criticise HS. The argument is self- explanatory in the former case, and in the latter, it could be argued that the risk of erroneous parameterisation is higher when there are many parameters, and their relations are not transparent.

Accuracy of yield-safe predictions

YS crop yield predictions in the MAFF-sponsored trials showed correlation with measurements and appropriate shapes of datasets. For the MAFF-sponsored trials, YS had a negative bias in crop yield predictions as well as small negative biases in the tree heights and DBH. Generally, however, the YS models simulated tree growth in both Leeds and Silsoe trials adequately.

For the CFEC, even though there was no general agreement between predicted and measured values of crop yields, the general shape and the variance of the measured dataset was reflected in the predicted dataset to some degree. Predictions of CFEC wood yields generally failed. The reasons for the failure are unknown. As YS is a generic model, designed for creating models of a wide variety of growing systems, SRC systems should not present any specific problems, and other authors have had success with simulating this kind of systems (Seserman et al., 2019; Seserman et al., 2018).

Compared to the HS GSMs, the YS GSMs were much easier to calibrate, which translated into better calibration results in several instances.

Magnitude of errors

For HS, the magnitude of the errors in crop yields in the MAFF GSMs were in the area of 2-3 t/ha, which is considerable. In an English, conventional farming context, a yield of 8 t/ha could be considered a high yield, a yield of 5 t/ha a low-medium yield, and a yield of 2 t/ha a low yield (Pantazi et al., 2016), so the errors in the predictions were close to what amounts to a difference of one whole yield class. This is not satisfying. The same is the case in the CFEC GSMs created with HS. Here, the RMSE value of the yield predictions was lower than in the MAFF GSMs, but so were the values of the observed and predicted yields, so the errors arguably corresponded to a yield class in this organic, low-input system also. In the MAFF Leeds trial, the magnitude of the errors on the predictions of tree growth was satisfactorily low. This was not the case in the MAFF Silsoe trial or in any of the CFEC systems.

For the YS based GSMs, crop yields had errors of similar magnitude as for the HS based GSMs. The tree growth in both Leeds and Silsoe was simulated with low error, but the error on the tree growth predictions in the CFEC system was larger.

Generally, RMSE values of the HS simulations were a bit lower than the corresponding values for YS simulations, implying a generally lower level of error in the HS predictions. The difference was small, however.

Biases

For unknown reasons, both YS and HS models showed a negative bias in their crop yield predictions in the MAFF systems. If the complete failure of the HS model to simulate Silsoe tree growth is not considered, the YS model could be said to generally have a more negative bias in the predictions of tree growth, which is detectable both in the MAFF and the CFEC systems. For the MAFF systems, this may partly have arisen in the calibration process, where slightly low tree heights were simulated in the final calibration runs, and a significant bias was indicated by the statistics. However, all these differences were rather small and cannot be used to infer anything general about the relationship between biases of YS or HS, respectively.

4.1.2 Model validity assessed in other studies

YS has been used in multiple other studies, including five studies of Northern European ACS. Seserman et al. (2019) attained remarkable agreement between predictions and a validation dataset. In the other studies assessed, generally, the tree growth was simulated well (Seserman et al., 2018), while the crop yields showed deviations from the measurements of about a yield class (Graves et al., 2010; van der Werf et al., 2007). De Jalon et al. (2018) successfully calibrated YS and reported a high level of agreement between predictions and the calibration dataset. However, crops were calibrated using a mean yield value, so the high level of concurrence will include an allowance for some variation around this mean.

To sum up, only one study attained great concurrence with validation values. This study was the only one to use the model improvements suggested by Palma et al. (2016). When these improvements are implemented, the model considers the effects of moderation of microclimate by the trees, such as the creation of wind shelter and the like, which are critical interactions in ACSs. As these improvements were not implemented in the models created in my study, maybe they could have improved the results.

HS has not been used in many published studies, and especially not in a Northern European context, even though specific elements in HS have been advantageously used. For examples, see section 4.1.4.

4.1.3 Model purpose and other model traits

This section commences with an argument for keeping the purpose of any given model clear and concise and goes on to mention other model traits that are widely recognised as important, including simplicity and flexibility. Then, YS and HS are evaluated regarding these properties.

The importance of a concise model purpose

During the ‘golden era of crop modelling' (Affholder et al., 2012) in the 1980'ies, a universal, or generic, crop model was envisioned. In the 1990'ies, however, doubts were voiced regarding the efficiency of the dominant strategy of development of crop models. For the following reasons, the pursuit of a universal crop model was proclaimed to be futile:

- Cropping systems or crops are not deterministic systems. Randomness is involved, and the outcome of a given initial state is not always the same. (Sinclair and Seligman, 1996).
- Cropping systems are complex. There is a large amount of possible variation in all different aspects of a crop during the entirety of its lifespan, and over the whole of the space it occupies (van Gardingen et al., 1997).
- The aim of each modelling project will be different. Building a universal model necessitates having many redundant factors in the model. Models must be kept simple enough to remain usable and relatively transparent (van Gardingen et al., 1997). This conflicts with the considerable redundancy in parameters necessitated by the objective of making a model universal (de Oliveira et al., 2017; Spitters, 1990).

Passioura (1996) emphasized the distinction between scientific models and engineering models and defined the two types as roughly synonymous with mechanistic and empirical models, respectively. He argued that the complexity of the simulated situations made it unfeasible to develop mechanistic (scientific) crop models with predictive power. Therefore, the distinction mentioned above must remain clear in order to avoid futile attempts to produce mechanistic (scientific) models for use in practical problem-solving.

Furthermore, Passioura (1996) suggested specific general characteristics for the error of a predictive model: namely, that the amount of error in the predictions from a given model correlated with model complexity, defined as the number of parameters used by the model. As more parameters were added to the model, the error stemming from an incomplete representation of the real modelled system would decline. However, during parameterisation, every parameter would likely be given a value that deviates a little from the value that would be a perfect representation of the corresponding real-world phenomenon. This was a source of error in the model predictions. Therefore, as more parameters were added to the model, the error stemming from error in the parameter values would rise. Thus, the paper claimed that there would be some optimum complexity of the given model and that the accuracy of the model predictions could not be infinitely improved by adding more parameters (see figure 4.2 ).

Abbildung in dieser Leseprobe nicht enthalten

Figure 4.2: Relationship between the complexity of a given model and the error of that model ’s predictions. Complexity is defined as the number of parameters used by the model. The 'Arising from structure' curve shows the amount of error stemming from incomplete representation of the modelled system, while the 'Arising from parameters' curve shows the amount of error stemming from the error of the parameter values used. The 'Total' curve is the combined amount of error from both sources. Adapted from Passioura (1996).

Several authors have argued that crop models intended for 'engineering' purposes, in the words of Passioura (1996), should be built with their purpose acutely in mind (Passioura, 1996; Sinclair and Seligman, 1996). Amongst other benefits, this would ensure that a given model was relevant to and designed in accordance with the context in which it would be applied.

This study tests the HS and YS models perceived as engineering-type models since the objective is to reveal if they can be applied to real-world problem solving and processes of system design. Therefore, it is relevant to evaluate these models concerning the above properties.

If judged by how well-defined and concise their purpose is, both models could be given a low grade. The HS model was built explicitly to model European AFSs and was built around simulating ACSs. However, there are so many more options in the model that it can be characterised as a relatively generic model. HS seems to have many of the negative traits associated with a generic model: Large redundancy, (too) high complexity and so on. Also, the focus on simulating everything in great detail is likely part of the explanation for the long run times.

YS could also be called a generic model, because it has a multitude of different options and systems it is capable of simulating. In a certain sense, for YS, this may be said to be less grave, since the many options arise from the many sub-modules that make it up. These sub-modules can be easily activated or deactivated and are easily tweaked. Furthermore, the model does not seem to contain the same level of redundancy. In the present study, this translated into better calibration results. The shorter run times of YS meant that it was feasible to run more calibrations in order to achieve greater accuracy. Still, one of the underlying assumptions of YS is that concerning competition for light and water, any combination and layout of trees and crops can be satisfactorily reduced to trees planted on a field in a regular planting pattern, with even spacing between all trees in both directions. This has implications for all interactions between trees and understorey vegetation, since rows of trees with higher densities will compete only with the fraction of crop plants near the row, while the same number of trees evenly distributed across the same area will compete with a higher proportion of crop plants. This likely gives rise to some error in the model, and a model developed for a more defined, narrow purpose (for example modelling ACSs with tree rows of only individual, lone standing trees in one row, or one developed explicitly for simulating SRC ACSs) probably would avoid some of this error.

Simplicity, flexibility and other model traits

Luedeling et al. (2016), published the most recent review focussed on the modelling of AFSs. Concluding from the entire history of AFS modelling, the authors stated that in general, the model software produced by earlier projects tended to go out of use when the researcher or developer team responsible for developing the model terminated their involvement. The paper deduced that model developers must ensure model longevity and long-term relevance during the process of model design and development. The authors suggested, in concurrence with several other review authors (Affholder et al., 2012; Jones et al., 2016; Holzworth et al., 2018) that this could be promoted by applying the following measures:

- Ensuring model simplicity.
- Designing and planning model maintenance as part of the development process.
- Building models in a modular fashion, leading to the creation of more flexible models. Specifically, the authors suggested that any model should be built in a way so that the given model could act as a sub-model of other models. Furthermore, the architecture should facilitate using outputs of other models as inputs, so that those other models could become sub-models of the model in question. This required that the models were built in standard, generally available software environments, with no use of special modelling software, so that any model could be combined with other models without the need of extensive model modification.

As these traits are widely recognised as important, it is prudent to evaluate HS and YS regarding them. HS is a much more complex and parameter-rich model than YS. In order to run the final simulations in this study, it was necessary to set values for 121 parameters and 7 state variables for every individual simulation, including every parameterisation and calibration simulation on the way to building the final models. In contrast to this, YS requires the setting of 24 parameters and initial conditions and 22 state variables (Palma et al., 2010). HS also has much longer simulation run times than YS. This encumbers the processes of parameterisation and calibration, as well as the process of learning to use the model. Furthermore, running simulations with the model demands much more memory from the computer running the simulation. So even though in principle, most system designs and layouts (even with multiple tree species, which is impossible in YS) can be simulated with HS, for most users, the layout of the systems that can be simulated with HS is limited in size and complexity.

As the HS model itself is built upon multiple different complex sub-models, it is dependent upon the sustained maintenance of these. A large part of model maintenance is the responsibility of people other than the model developers of HS, which may make maintenance a complicated task. Furthermore, HS has been developed by a relatively small team, building on some pieces of software that are not widely used (STICS and Capsis), so the amount of people who know how it is built is small. YS is built in Microsoft Excel, so here also, the model developers and users cannot maintain the base software. However, Excel is widely used software, and can probably be relied on as a relatively stable foundation for a long time to come.

HS is modular in the sense that the user can choose from a wide variety of different simulation outputs. For each simulation, the model can be set to not record specific elements in the simulation run (e.g., all values related to individual soil voxels, to trees, to the agricultural crops and so on). This influences the simulation run times a lot since more or fewer files must be written. This way, HS can be said to consist of several sub-models. Of course, being built upon two existing modelling programs is also a modular trait. However, this kind of modularity is not necessarily the kind that leads to the interoperability and flexibility called for by several authors (Jones et al., 2016; Luedeling et al., 2016), since it most likely makes it more complicated to build new models with HS as a submodel. YS has a modular set-up in the sense that it consists of multiple modules that can be activated or deactivated for any given simulation. In the version used in this study, the different modules include a crop module, a tree module, a fruit module, a livestock module, a SOC module, and a cork module. Several of these modules have been developed after the release of the original model because they were needed for some concrete task. This indicates the flexibility of YS. Also, the fact that it is built in Microsoft Excel could be said to be a modular trait, since this software is widely used for building models, and it will generally be easy to use the outputs from YS models in other Excel-based models, and vice versa. This is an example of modularity leading to an interoperable model. The development of the internet­based version, Eco Yield-SAFE (AGFORWARD, n.d.), can be said to be a step that complicates this modular use because in that version, the algorithms of the model cannot be modified, and the output of the simulations cannot directly be linked to other Excel workbooks, but have to be downloaded first.

Jones et al. (2016) emphasized the demand for models that can be used in transdisciplinary studies. Judging from their design, both YS and HS seem to be well suited for this. HS can simulate virtually any element in the growing system, facilitating the application of the model in the context of spatial studies, of studies of the fate of nutrients, such as N, in the landscape, of landscape architecture, among many other possibilities. HS has also been used for a variety of tasks in transdisciplinary studies (Artru et al., 2017; Dupraz et al., 2018). YS, in a certain sense is a more empirical, and less mechanistic model, which necessitates a little bit more processing of the outputs from the model before they can be used in other contexts. It does not have the spatial concepts of HS, which produces a three-dimensional representation of the growing system across the period simulated. However, the development of further models in other disciplines based on YS, such as the Farm-SAFE model (Graves et al., 2007) that allows economic assessment of AFS, shows that YS to a high degree also is designed for transdisciplinary studies.

4.1.4 Breakthroughs in model design

The HS model does include many new interesting features that have much potential. The ray-tracing method of determining shading on the understorey crop, the way trees root in three dimensions, and root in interaction with the crop and with management, as well as the implementation of the toric symmetry principle all seem to represent important aspects of ACSs that could not be represented before. For a thorough explanation of these concepts, see Dupraz et al. (2019). For instance, without the ray-tracing method of simulating sunlight interception, the effects of different pruning strategies would likely be difficult to determine, and the same goes for the method of simulating root growth and different cultivation or root pruning strategies. The toric symmetry principle removes artificial edge effects that could otherwise have arisen in simulations. All these innovations seem like useful and in some ways, quite critical elements, that could make the difference between success and failure in finding ways of getting ACSs to work well. However, when all implemented together, they may be what makes the model incomprehensible, and what makes the run times so long.

Albeit, they can be perceived as breakthroughs in the field of ACS modelling if viewed in isolation. If these could be run as standalone models, or be used as sub-modules of other models, they might contribute a considerable deal more to the field of AF modelling. However, preparing them for this may be a difficult task, since they are intricately integrated into HS, and perhaps do not work on their own. Excellent examples of successful, creative use of the model software, which makes use of the advantages of these model elements, do exist. Talbot (2011) used HS together with data from an ACS in Southern France to answer the question: “How do the components of the growing system, and the system as a whole, respond to changes in tree density and placement?”. Holst et al. (2012) used HS to assess the mitigation of nitrogen pollution attainable by ACSs in northern China. However, challenges were allegedly encountered while adapting the model to the Chinese context (Luedeling et al., 2016). Artru et al. (2017) measured wheat yields under shade cloth in Belgium and combined this with predictions from the ray-tracing simulator and the tree growth simulator in HS. With this method, it was determined how many years from tree planting the observed yield reductions in the wheat would materialise. Dupraz et al. (2018) used HS to determine irradiation of crops in ACSs at different latitudes, including at 65° North or South. However, the studies did not simulate crop yields, so they cannot be used to infer anything about the reliability of yield predictions. These examples indicate the real strengths and potentials of the model to aid in the design of novel ACSs.

The main achievement of the YS model is to make it possible to simulate a wide variety of growing systems in a relatively easy way. The modular structure of the model also is a significant asset, and could be said to be a breakthrough, although other models have achieved the same before (e.g., WaNuLCAS (van Noordwijk and Lusiana, 1998)). The fact that the model has proven its worth in some previous studies also gives credibility to the statement that in YS, an easy to use model that produces reliable outputs has been achieved. However, in the present study, not all the results are equally impressive.

If viewed in combination, HS and YS seem to complement each other well. YS can produce useful predictions of wood yields, and, in some cases, crop yields across a long period, and at coarse levels of detail (e.g., Seserman et al. (2018)). HS can help to answer questions regarding concrete structural details in the design of ACSs (Talbot, 2011), and be used to gain a comprehensive understanding of the conditions of insolation at a given latitude and in a given situation (Dupraz et al., 2018).

4.1.5 Further model development

Some work obviously still needs to be done on basic technicalities in HS. The team of developers are currently working to correct the problem of the tree shade not influencing crop temperatures (C. Dupraz, 2020, personal communication, April 30th). The experience from the present study indicates that several similar bugs likely must be fixed before the software can make reliable predictions. Regarding YS, other studies have shown that it can produce reliable predictions in some situations, when the improvements suggested by Palma et al. (2016) are used. There is a need to examine if YS can consistently produce reliable predictions in various situations. This can be examined through continually using the model. Therefore, evaluating the use of the model in a variety of situations would be an appropriate way of developing the model further.

4.1.6 Overall comparison between models

Both the two models in this study produced results of varying levels of accuracy. Both models achieved a certain level of success in simulating the MAFF systems, even though the crop yields generally showed a negative bias. Moreover, both the models had difficulties making useful predictions of tree growth in the CFEC systems. HS failed at predicting crop yields in CFEC, while YS was more successful in this task. It was much easier to calibrate and use the YS model than the HS model, which also made it possible to make better calibrations. Some of these differences might have been less obvious if the hardware used to run the model was better, which might have cut down HS run times. Likewise, the differences might have been reduced with a more experienced modeller, who might have been able to make better judgements as to how best to make use of the HS and YS models, respectively. However, even these differences, that the YS model demands less experience to use, and have smaller requirements to the computer it is run on, may be perceived as essential advantages of the YS model. Table 4.2 summarises the strengths and weaknesses of HS and YS, respectively, and gives suggestions for further development.

Abbildung in dieser Leseprobe nicht enthalten

4.2 Sources of error

This section describes elements of the modelling procedure followed that may have decreased model performance. Following other procedures on these points might have produced better model outputs or have led to clearer study results.

The MAFF and CFEC systems were calibrated in different ways. In the MAFF systems, data for three structurally identical growing systems existed, and one of these were chosen so that calibration was done in a system that resembled the other systems very much, except for differences in soil and weather. Hence, calibration and validation simulations ran across the same period.

In the CFEC systems, all physical growing systems were located at the same place, but they were structurally different, due to varying distances between biofuel hedges. Therefore, one growing system could not be used for calibration in the entire period like in the MAFF systems, because in that case, the weather input would have been the same in calibration and validation runs, which would have constituted too similar conditions. Furthermore, for the CFEC, only figures for the total wood yield across all systems were available, not the wood yield for each individual system. These challenges obviated the possibility of choosing one of the CFEC sub-systems as a calibration dataset. Therefore, the total period for which data were available was split into a calibration and a validation section, respectively, which is a significant difference to the MAFF simulations. Both methods are valid methods of calibration, and it is not easy to tell whether one method is superior to the other, or whether one is more likely to give reliable predictions. This, combined with the difference in variance of measured yields between the periods 2000-2011 and 2012-2019, respectively (see section 4.1.1), may explain the difference in the level of success between the simulations of MAFF growing systems, producing some reasonable results, and the simulations of CFEC systems, which proved more difficult. However, the higher level of success attained in the YS CFEC crop yield predictions decreases the credibility of this explanation. Furthermore, the fact that the calibration procedure used for the CFEC seems to be more popular among other authors (see section 2.3.2) also undermines the argument that the procedure used for the MAFF-sponsored systems should be superior.

Summing up, the potential for achieving clearer results by finding trials that can be calibrated following identical procedures is likely only slight.

Initially, the HS GSMs often predicted that the soil water stock would not be replenished over winter, which seems unrealistic in a Northern European context. To correct this, the precipitation values in the weather data obtained were multiplied by 2 in all HS simulations, while the transpiration water use of the plants was correspondingly increased. This does seem like a risky way to obtain a more realistic representation of the growing system because it might create unrealistic reactions in other parts of the GSMs in question. Such unrealistic reactions were not observed, and the solution did solve the problem of extreme early-season drought. Even so, it is appropriate to note that a potential improvement of the models created in this study may be possible by making the correction in another way.

Palma et al. (2016) developed improvements of YS implementing tree effects on microclimate into the model. In the version of YS used in this study, these were disabled per default. For this reason, the improvements were not implemented in the simulations run in this study. The improvements might have been able to improve the predictions of crop yields and should be implemented in a future study with similar objectives.

Some of the crops simulated in the MAFF systems were calibrated with rather few data points. Having more data points might have improved the calibration of the crop element in the MAFF-sponsored trials and should be a priority for a future study with similar objectives.

4.3 The role of models in the development of alley cropping systems

The models studied here hold potential for taking their respective places in the work of developing ACSs in Northern Europe. YS can be used to make broadscale assessments of resource efficiency and environmental impacts (such as N pollution and soil C sequestration), as well as economic analysis when combined with the FARM-safe model (Graves et al., 2011). Therefore, the model can be very useful on a landscape scale and perhaps also on a national scale and beyond. HS can be used to optimise individual ACS designs (Talbot, 2011), and to explore in which regions the climate sets barriers to implementing ACSs (Dupraz et al., 2018). Furthermore, HS can be used to examine the importance of a range of interactions in ACSs for the general field of AF to progress.

The results from the present study indicate that a more complex model does not produce more accurate predictions of yields of trees and crops in an existing ACS. The easy-to-use, simpler model did the job about as well as the complex model, and with a much lesser expense of time used for learning to use the model, sensitivity analysis, parameterisation, calibration, and running the models. This reflects the distinction made by Passioura (1996) of scientific and engineering models. Albeit, in this study, even the simpler model did not produce very accurate or reliable crop yield predictions. As quite a lot of data was available for the work carried out in this study, the availability of more data probably would only be a small improvement. As stated above, implementing the model improvements suggested by Palma et al. (2016) might have improved the results. A third, critical factor is the skill and experience of the operating modeller. Operation of the models by more experienced modellers might also have improved results, as exemplified by Seserman et al. (2019).

Work is still underway on the HS model. As mentioned in section 4.1.5, the team of developers are currently working to correct the problem of the tree shade not influencing crop temperatures (C. Dupraz, 2020, personal communication, April 30th). This would likely have meant a substantial improvement of the HS crop yield predictions in this study and could possibly have eliminated the negative bias in the predictions. So, it is possible that with time and further development of HS, the model will be able to produce reliable crop yield predictions in ACS with large, lone standing trees.

The interactions in ACSs are many, and perhaps still not studied enough to be adequately simulated. Having a broader diversity of growing systems to study and model could improve this, because it would become clearer what the critical interactions in the systems are. With a clearer picture of this, it might be possible to make simpler models that still capture the critical interactions in the modelled systems. As noted above, HS could have a significant role in this effort, and ACS simulation should be integrated field studies of these systems. However, another thing that is urgently needed is the establishment of more different AF systems, and that comprehensive studying of these is carried out so that confidence can be attained in the knowledge of their traits and interactions.

5. Conclusion

Both the models studied can be used to successfully produce tree growth predictions in alley cropping systems with large, lone standing trees. None of the models made reliable predictions of crop yields in alley cropping systems in this study. For yield-safe, more experienced modellers might have achieved reliable predictions. Predictions of wood yields in short rotation coppice systems were not very successful in this study with any of the models. Yield-safe was found to be substantially easier to calibrate and use than Hi-sAFe, leading to better simulation results when measured on some parameters. However, the current work on Hi-sAFe to improve crop yield predictions may eventually give this model an essential advantage over yield-safe, in the form of lower error on crop yield predictions under tree canopies.

Evaluating other aspects of the design of the models, the profoundly mechanistic Hi-sAFe is flexible and well suited for transdisciplinary studies, while at the same time being complex and unwieldy. Yield-safe stands out because of its balance between being relatively simple and including most elements that are important in European agroforestry systems, and because of its modular build-up, that also lends it flexibility.

The models studied here cam currently only be a supplement when innovating novel kinds of agroforestry systems for the Northern European region. Simulations made with the software must be complemented by other kinds of studies, such as field trials. To improve the usefulness of the models, critical interactions in agroforestry systems should be studied in the field so that model design could be improved based on knowledge of critical interactions in alley cropping systems.

6. Literature

AFFHOLDER, F., TRITTONELL, P., CORBEELS, M., ROUX, S., MOTISI, N., TIXIER, P & WERY, J. 2012. Ad Hoc Modeling in Agronomy: What Have We Learned in the Last 15 Years? Agronomy Journal, 104.

AGFORWARD. n.d. EcoYield-SAFE - The web version to ease your assessments [Online]. Available: http://www.isa.ulisboa.pt/proj/ecoyieldsafe/#/dashboard [Accessed 08-05-2020 2020].

ALLEN, R. G., PEREIRA, L. S., RAES, D. & SMITH, M. 1998. Crop evaporation - Guidelines for computing crop water requirements. FAO Irrigation and drainage paper. FAO.

ANEL, B., OLIVIER, A. & RIVEST, D. 2017. Une agroforesterie pour le Québec. Document de réflexion d'orientation. Québec: Centre de référence en agriculture et agroalimentaire du Québec.

ARTRU, S., GARRÉ, S., DUPRAZ, C., HIEL, M.-P., BLITZ-FRAYRET, C. & LASSOIS, L. 2017. Impact of spatio-temporal shade dynamics on wheat growth and yield, perspectives for temperate agroforestry. European Journal of Agronomy, 82 , 60-70.

BARRIOS, E., SILESHI, G. W., SHEPHERD, K. D. & SINCLAIR, F. 2012. Agroforestry and Soil Health: Linking Trees, Soil Biota, and Ecosystem Services. In: WALL, D. H. (ed.) Soil Ecology and Ecosystem Services. 1. edition ed. London: Oxford University Press.

BEST, L. B., WHITMORE, R. C. & BOOTH, G. M. 1990. USE OF CORNFIELDS BY BIRDS DURING THE BREEDING-SEASON - THE IMPORTANCE OF EDGE HABITAT. American Midland Naturalist, 123 , 84-99.

BHARATI, L., LEE, K. H., ISENHART, T. M. & SCHULTZ, R. C. 2002. Soil-water infiltration under crops, pasture, and established riparian buffer in Midwestern USA. Agroforestry Systems, 56 , 249-257.

BLASER, W. J., OPPONG, J., HART, S. P., LANDOLT, J., YEBOAH, E. & SIX, J. 2018. Climate­smart sustainable agriculture in low-to-intermediate shade agroforests. Nature Sustainability, 1 , 234-239.

BRISSON, N., LAUNAY, M., MARY, B. & BEAUDOIN, N. 2008. Conceptual basis, formalisations and parameterization of the STICS crop model, Versailles, Éditions Qus.

BURGESS, P J. & ROSATI, A. 2018. Advances in European agroforestry: results from the AGFORWARD project. Agroforestry Systems, 92 , 801-810.

BURGESS, P J., CASADO, M. R., GAVU, J., MEAD, A., COCKERILL, T., LORD, R., VAN DER HORST, D. & HOWARD, D. C. 2012. A framework for reviewing the trade-offs between, renewable energy, food, feed and wood production at a local level. Renewable & Sustainable Energy Reviews, 16 , 129-142.

BURGESS, P. J., INCOLL, L., HART, B. J., BEATON, A., PIPER, R. W., SEYMOUR, I., REYNOLDS, F. H., WRIGHT, C., PILBEAM, D. & GRAVES, A. 2003. The Impact of Silvoarable Agroforestry with Poplar on Farm Profitability and Biological Diversity - Final Report to DEFRA. Institute of Water and Environment, Cranfield University.

BURGESS, P. J., INCOLL, L. D., CORRY, D. T., BEATON, A. & HART, B. J. 2005. Poplar (Populus spp) growth and crop yields in a silvoarable experiment at three lowland sites in England. Agroforestry Systems, 63 , 157-169.

BÖHM, C., KANZLER, M. & FREESE, D. 2014. Wind speed reductions as influenced by woody hedgerows grown for biomass in short rotation alley cropping systems in Germany. Agroforestry Systems, 88 , 579-591.

CARDOZO, E. G., MUCHAVISOY, H. M., SILVA, H. R., ZELARAYAN, M. L. C., LEITE, M. F. A., ROUSSEAU, G. X. & GEHRING, C. 2015. Species richness increases income in agroforestry systems of eastern Amazonia. Agroforestry Systems, 89 , 901-916.

CORREAL, E., ERENA, M., RIOS, S., ROBLEDO, A. & VICENTE, M. 2009. Agroforestry Systems in Southeastern Spain. In: RIGUEIRO-RODROGUEZ, A., MCADAM, J. & MOSQUERA-LOSADA, M. R. (eds.) Agroforestry in Europe: Current Status and Future Prospects. Dordrecht: Springer Netherlands.

DE JALON, S. G., GRAVES, A., PALMA, J. H. N., WILLIAMS, A., UPSON, M. & BURGESS, P J. 2018. Modelling and valuing the environmental impacts of arable, forestry and agroforestry systems: a case study. Agroforestry Systems, 92 , 1059-1073.

DEMIDENKO, E. 2019. Advanced Statistics with Applications in R, Hoboken, NJ, John Wiley & Sons Inc.

DE OLIVEIRA, C. P M., SIMOES, W. L., SILVA, J. A. B., LOPES, P. R. C. & ASSIS, J. S. 2017. QUALITY OF IRRIGATED APPLES IN THE SEMIARID REGION OF THE NORTHEAST OF BRAZIL. Revista Caatinga, 30 , 760-767.

DUPRAZ, C. 2020 april 30th 2020. RE: Discussion on the lack of influence of shade on crop temperature. Email to HANSEN, J.

DUPRAZ, C., BLITZ-FRAYRET, C., LECOMTE, I., MOLTO, Q., REYES, F. & GOSME, M. 2018. Influence of latitude on the light availability for intercrops in an agroforestry alley­cropping system. Agroforestry Systems.

DUPRAZ, C., WOLZ, K. J., LECOMTE, I., TALBOT, G., VINCENT, G., MULIA, R., BUSSIERE, F., OZIER-LAFONTAINE, H., ANDRIANARISOA, S., JACKSON, N., LAWSON, G., DONES, N., SINOQUET, H., LUSIANA, B., HARJA, D., DOMENICANO, S., REYES, F., GOSME, M. & VAN NOORDWIJK, M. 2019. Hi-sAFe: A 3D Agroforestry Model for Integrating Dynamic Tree-Crop Interactions. Sustainability, 11.

DUPRAZ, C., BURGESS, P., GAVALAND, A., GRAVES, A., HERZOG, A., INCOLL, L. D., JACKSON, N., KEESMAN, K., LAWSON, G., LECOMTE, I., LIAGRE , F., MANTZANAS, K., MAYUS, M., MORENO , G., PALMA, J., PAPANASTASIS, V., PARIS, P., PILBEAM, D. J., REISNER, Y, VAN NOORDWIJK, M., VINCENT, G. & VAN DER WERF, W. 2005. SAFE Final Report - Synthesis of the SAFE project. Montpellier.

FAHAD, M., AHMAD, I., REHMAN, M., WAQAS, M. M. & GUL FARHANA 2019: Regional wheat yield estimation by integration of remotely sensed soil moisture into a crop model. Canadian Journal of Remote Sensing, 45.

GARRETT, H. E. & MCGRAW, R. L. 2000. Alley Cropping Practices. In: GARRETT, H. E., RIETVELD, W. J. & FISHER, R. F. (eds.) North American Agroforestry: An Integrated Science and Practice. Madison, USA: American Society of Agronomy, Inc.

GHALEY, B. B. & PORTER, J. R. 2014. Determination of biomass accumulation in mixed belts of Salix, Corylus and Alnus species in combined food and energy production system. Biomass & Bioenergy, 63 , 86-91.

GIBBS, S., KOBLENTS, H., COLEMAN, B., GORDON, A., THEVATHASAN, N. & WILIAMS, P. 2016. Avian diversity in a temperate tree-based intercropping system from inception to now. Agroforestry Systems, 90 , 905-916.

GRAVES, A. R., BURGESS, P. J., PALMA, J. H. N., HERZOG, F., MORENO, G., BERTOMEU, M., DUPRAZ, C., LIAGRE, F., KEESMAN, K., VAN DER WERF, W., DE NOOY, A. K. & VAN DEN BRIEL, J. P. 2007. Development and application of bio-economic modelling to compare silvoarable, arable, and forestry systems in three European countries. Ecological Engineering, 29 , 434-449.

GRAVES, A. R., BURGESS, P. J., LIAGRE, F., TERREAUX, J. P., BORREL, T., DUPRAZ, C., PALMA, J. & HERZOG, F. 2011. Farm-SAFE: the process of developing a plot- and farm­scale model of arable, forestry, and silvoarable economics. Agroforestry Systems, 81 , 93­108.

GRAVES, A. R., BURGESS, P. J., PALMA, J., KEESMAN, K. J., VAN DER WERF, W., DUPRAZ, C., VAN KEULEN, H., HERZOG, F. & MAYUS, M. 2010. Implementation and calibration of the parameter-sparse Yield-SAFE model to predict production and land equivalent ratio in mixed tree and crop systems under two contrasting production situations in Europe. Ecological Modelling, 221 , 1744-1756.

HANSEN, S. 1984. ESTIMATION OF POTENTIAL AND ACTUAL EVAPOTRANSPIRATION. Nordic Hydrology, 15 , 205-212.

HARVEY, C. A. & VILLALOBOS, J. A. G. 2007. Agroforestry systems conserve species-rich but modified assemblages of tropical birds and bats. Biodiversity and Conservation, 16 , 2257­2292.

HOLST, J., LIU, W., XIAO, H., YU, Z. & DOLUSCHITZ, R. Agroforestry - exploring a mitigation option for nitrogen pollution in cropping systems of the North China Plain. 8th Asian Conference for Information Technology in Agriculture, 2012 Taipei, Taiwan. Taiwan Agricultural Research Institute.

HOLZWORTH, D., HUTH, N. I., FAINGES, J., BROWN, H., ZURCHER, E., CICHOTA, R., VERRALL, S., HERRMANN, N. I., ZHENG, B. & SNOW, V 2018. APSIM Next Generation: Overcoming challenges in modernising a farming systems model. Environmental Modelling & Software, 103 , 43-51.

HOOSBEEK, M. R., REMME, R. P & RUSCH, G. M. 2018. Trees enhance soil carbon sequestration and nutrient cycling in a silvopastoral system in south-western Nicaragua. Agroforestry Systems, 92 , 263-273.

IPCC 2014. Climate Change 2014: Synthesis Report. Contributions of Working Groups I, II and III to the Fifth Assesment Report of the Intergovernmental Panel on Climate Change. In: PACHAURI, R. K. & MEYER, L. A. (eds.). Geneva, Switzerland: IPCC.

JOHNSON, R. J. & BECK, M. M. 1988. INFLUENCES OF SHELTERBELTS ON WILDLIFE MANAGEMENT AND BIOLOGY. Agriculture Ecosystems & Environment, 22-3 , 301-335.

JONES, J., ANTLE, J., BASSO, B., BOOTE, K., CONANT, R., FOSTER, I., GODFRAY, C., HERRERO, M., HOWITT, R., JANSSEN, S., KEATING, B., MUNOZ-CARPENA, R., PORTER, C., ROSENZWEIG, C. & WHEELER, T. 2016. Brief history of agricultural systems modelling. Agricultural Systems, 155.

JOSE, S. 2009. Agroforestry for ecosystem services and environmental benefits: an overview. Agroforestry Systems, 76 , 1-10.

KAY, S., REGA, C., MORENO, G., DEN HERDER, M., PALMA, J. H. N., BOREK, R., CROUS- DURAN, J., FREESE, D., GIANNITSOPOULOS, M., GRAVES, A., JAGER, M., LAMERSDORF, N., MEMEDEMIN, D., MOSQUERA-LOSADA, R., PANTERA, A., PARACCHINI, M. L., PARI, P., ROCES-DIAZ, J. V., ROLO, V., ROSATI, A., SANDOR, M., SMITH, J., SZERENCSITS, E., VARGA, A., VIAUD, V., WAWER, R., BURGESS, P. J. & HERZOG, F. 2019. Agroforestry creates carbon sinks whilst enhancing the environment in agricultural landscapes in Europe. Land Use Policy, 83 , 581-593.

KRISHNAMURTHY, P. K. & REDDIAR, K. L. 2011. Agrobiodiversity for Livelihood Security: A Case Study of Agroforestry Technologies in Mexico. Journal of Life Sciences, 5 , 108-119.

LOAGUE, K. & GREEN, R. E. 1991. Statistical and graphical methods for evaluating solute transport models: Overview and application. Journal of Contaminant Hydrology , 51-73.

LUEDELING, E., SMETHURST, P. J., BAUDRON, F., BAYALA, J., HUTH, N. I., VAN NOORDWIJK, K., ONG, C. K., MULIA, R., LUSIANA, B., MUTHURI, C. & SINCLAIR, F. L. 2016. Field-scale modeling of tree-crop interactions: Challenges and development needs. Agricultural Systems, 142 , 51-69.

MATOCHA, J., SCHROTH, G., HILLS, T. & HOLE, D. 2012. Integrating Climate Change Adaptation and Mitigation Through Agroforestry and Ecosystem Conservation. In: NAIR, P. K. R. & GARRITY, D. (eds.) Agroforestry - the future of global land use. Dordrecht: Springer.

MEAD, R. & WILLEY, R. W. 1980. THE CONCEPT OF A LAND EQUIVALENT RATIO AND ADVANTAGES IN YIELDS FROM INTERCROPPING. Experimental Agriculture, 16 , 217-228.

MORTIMER, P. E., GUI, H., XU, J., ZHANG, C., BARRIOS, E. & HYDE, K. D. 2015. Alder trees enhance crop productivity and soil microbial biomass in tea plantations. Applied Soil Ecology, 96 , 25-32.

MOSQUERA-LOSADA , M. R., MORENO , G., PARDINI , A., MCADAM, J. H., PAPANASTASIS, V, BURGESS , P. J., LAMERSDORF , N., CASTRO , M., LIAGRE , F. & RIGUEIRO-RODRIGUEZ, A. 2012. Past, Present and Future of Agroforestry Systems in Europe. In: NAIR, P. K. R. & GARRITY, D. (eds.) Agroforestry - the future of global land use. Dordrecht: Springer.

MUCHANE, M. N., SILESHI, G. W., GRIPENBERG, S., JONSSON, M., PUMARINO, L. & BARRIOS, E. 2020. Agroforestry boosts soil health in the humid and sub-humid tropics: A meta-analysis. Agriculture Ecosystems & Environment, 295.

NAHAYO, L., MUPENZI, C., HABIYAREMYE, G., KALISA, E., UDAHOGORA, M., NZABARINDA, V. & LI, L. H. 2019. Landslides Hazard Mapping in Rwanda using Bivariate Statistical Index Method. Environmental Engineering Science.

NAIR, P. K. R. 1993. An Introduction to Agroforestry, Dordrecht, the Netherlands, Kluwer Academic Publishers.

NAIR, P. K. R. 2012. Climate Change Mitigation: A Low-Hanging Fruit of Agroforestry. In: NAIR, P. K. R. & GARRITY, D. (eds.) Agroforestry - The Future of Global Land Use. Dordrecht: Springer.

NAIR, P. K. R. & GARRITY, D. 2012. Agroforestry - the Future of Global Land Use, Dordrecht, the Netherlands, Springer.

NAIR, P. K. R., KUMAR, B. M. & NAIR, V. D. 2009. Agroforestry as a strategy for carbon sequestration. Journal of Plant Nutrition and Soil Science, 172 , 10-23.

NEWMAN, S. M. & GORDON, A. M. 2018. Temperate Agroforestry: Key Elements, Current Limits and Opportunities for the Future. In: GORDON, A. M., NEWMAN, S. M. & COLEMAN, B. R. W. (eds.) Temperate Agroforestry Systems. 2nd ed. Boston, MA: CABI.

PALMA, J. H. N. 2017. CliPick - Climate change web picker. A tool bridging daily climate needs in process based modelling in forestry and agriculture. 2017, 26.

PALMA, J. H. N., BRUCE, J., ALMEIDA, A. & BATTAGLIA, M. 2010. Comparing YieldSAFE and CABALA, two process-based models with contrasting parameter requirements. International Union of Forest Research Organizations World Congress. Seoul.

PALMA, J., GRAVES, A., CROUS-DURAN, J., UPSON, M., PAULO, J. A., OLIVEIRA, T. S., DE JALON, S. & BURGESS , P. J. 2016. Yield-SAFE Model Improvements. Research Project: AGFORWARD 613520. Lisboa, Portugal.

PANTAZI, X. E., MOSHOU, D., ALEXANDRIDIS, T., WHETTON, R. L. & MOUAZEN, A. M. 2016. Wheat yield prediction using machine learning and advanced sensing techniques. Computers and Electronics in Agriculture, 121 , 57-65.

PASSIOURA, J. B. 1996. Simulation Models: Science, Snake Oil, Education, or Engineering? Agronomy Journal, 88 , 690-694.

PAUDEL, B. R., UDAWATTA, R. P., KREMER, R. J. & ANDERSON, S. H. 2012. Soil quality indicator responses to row crop, grazed pasture, and agroforestry buffer management. Agroforestry Systems, 84 , 311-323.

PRETTY, J. 2018. Intensification for redesigned and sustainable agricultural systems. Science, 362. R CORE TEAM 2019. R: A Language and Environment for Statistical Computing. Vienna, Austria. RAHMAN, P. M., VARMA, R. V. & SILESHI, G. W. 2012. Abundance and diversity of soil invertebrates in annual crops, agroforestry and forest ecosystems in the Nilgiri biosphere reserve of Western Ghats, India. Agroforestry Systems, 85 , 165-177.

ROCKSTROM, J., STEFFEN, W., NOONE, K., PERSSON, A., CHAPIN, F. S., LAMBIN, E., LENTON, T. M., SCHEFFER, M., FOLKE, C., SCHELLNHUBER, H. J., NYKVIST, B., DE WIT, C. A., HUGHES, T., VAN DER LEEUW, S., RODHE, H., SORLIN, S., SNYDER, P. K., COSTANZA, R., SVEDIN, U., FALKENMARK, M., KARLBERG, L., CORELL, R. W., FABRY, V J., HANSEN, J., WALKER, B., LIVERMAN, D., RICHARDSON, K., CRUTZEN, P & FOLEY, J. 2009. Planetary Boundaries: Exploring the Safe Operating Space for Humanity. Ecology and Society, 14.

ROSENBERG, K. V., LOWE, J. D. & DHONDT, A. A. 1999. Effects of forest fragmentation on breeding tanagers: A continental perspective. Conservation Biology, 13 , 568-583.

SANCHEZ, I. A. & MCCOLLIN, D. 2015. A comparison of microclimate and environmental modification produced by hedgerows and dehesa in the Mediterranean region: A study in the Guadarrama region, Spain. Landscape and Urban Planning, 143 , 230-237.

SCHULTZ, R. C., COLLETTI, J. P., ISENHART, T. M., MARQUEZ, C. O., SIMPKINS, W. W. & BALL, C. J. 2000. Riparian Forest Buffer Practices. In: GARRETT, H. E., RIETVELD, W. J. & FISHER, R. F. (eds.) North American Agroforestry: An Integrated Science and Practice. Madison, USA: American Society of Agronomy, Inc.

SEOBI, T., ANDERSON, S. H., UDAWATTA, R. P. & GANTZER, C. J. 2005. Influence of grass and agroforestry buffer strips on soil hydraulic properties for an Albaqualf. Soil Science Society of America Journal, 69 , 893-901.

SEITER, S., INGHAM, E. R., WILLIAM, R. D. & HIBBS, D. E. 1995. Increase in soil microbial biomass and transfer of nitrogen from alder to sweet corn in an alley cropping system. In: EHRENREICH JH, EHRENREICH DL & HW, L. (eds.) Growing a sustainable future. Boise: University of Idaho.

SEREKE, F., GRAVES, A. R., DUX, D., PALMA, J. H. N. & HERZOG, F. 2015. Innovative agroecosystem goods and services: key profitability drivers in Swiss agroforestry. Agronomy for Sustainable Development, 35 , 759-770.

SESERMAN, D. M., FREESE, D., SWIETER, A., LANGHOF, M. & VESTE, M. 2019. Trade-Off between Energy Wood and Grain Production in Temperate Alley-Cropping Systems: An Empirical and Simulation-Based Derivation of Land Equivalent Ratio. Agriculture-Basel, 9.

SESERMAN, D. M., POHLE, I., VESTE, M. & FREESE, D. 2018. Simulating Climate Change Impacts on Hybrid-Poplar and Black Locust Short Rotation Coppices. Forests, 9.

SINCLAIR, T. R. & SELIGMAN, N. G. 1996. Crop modelling: From infancy to maturity. Agronomy Journal, 88 , 698-704.

SMITH, J., SMITH, P & ADDISCOTT, T. 1996. Quantitative methods to evaluate and compare soil organic matter (SOM) models. In: POWLSON, D. S., SMITH, P. & SMITH, J. (eds.) Evaluation of soil organic matter models using existing, long-term datasets. Berlin: Spring­Verlag.

SMITH, P., SMITH, J. U., POWLSON, D. S., MCGILL, W. B., ARAH, J. R. M., CHERTOV, O. G., COLEMAN, K., FRANKO, U., FROLKING, S., JENKINSON, D. S., JENSEN, L. S., KELLY, R. H., KLEIN-GUNNEWIEK, H., KOMAROV, A. S., LI, C., MOLINA, J. A. E., MUELLER, T., PARTON, W. J., THORNLEY, J. H. M. & WHITMORE, A. P. 1997. A comparison of the performance of nine soil organic matter models using datasets from seven long-term experiments. Geoderma, 81 , 153-225.

SPITTERS, C. J. T. 1990. CROP GROWTH-MODELS - THEIR USEFULNESS AND LIMITATIONS.

SULTANA, N., HOSSAIN, M. & ISLAM, M. 2015. Income Diversification and Household Well­Being: A Case Study in Rural Areas of Bangladesh. International Journal of Business and Economics Research, 4 , 172-179.

TALBOT, G. 2011. L'intégration spatiale et temporelle du partage des ressources dans un système agroforestier noyers-céréales: une clef pour en comprendre la productivité? Doctor, Montpellier University II.

UDAWATTA, R. P., KREMER, R. J., ADAMSON, B. W. & ANDERSON, S. H. 2008. Variations in soil aggregate stability and enzyme activities in a temperate agroforestry practice. Applied Soil Ecology, 39 , 153-160.

VAN DER WERF, W., KEESMAN, K., BURGESS, P., GRAVES, A., PILBEAM, D., INCOLL, L. D., METSELAAR, K., MAYUS, M., STAPPERS, R., VAN KEULEN, H., PALMA, J. & DUPRAZ, C. 2007. Yield-SAFE: A parameter-sparse, process-based dynamic model for predicting resource capture, growth, and production in agroforestry systems. Ecological Engineering, 29 , 419-433.

VAN GARDINGEN, P R., CURRAN, P J. & FOODY, G. M. 1997. Scaling-up : from cell to landscape, Cambridge, Cambridge University Press.

VAN NOORDWIJK, M., LAWSON, G., SOUMARÉ, A., GROOT, J. J. R. & HEAIRIAH, K. 1996. Root distribution of trees and crops: competition and/or complementarity. In: ONG, C. K. & HUXLEY, P (eds.) Tree-Crop interactions: a physiological approach. Wallingford, UK: CABI.

VAN NOORDWIJK, M. & LUSIANA, B. 1998. WaNulCAS, a model of water, nutrient and light capture in agroforestry systems. Agroforestry Systems, 43 , 217-242.

WEERASEKARA, C., UDAWATTA, R. P., JOSE, S., KREMER, R. J. & WEERASEKARA, C. 2016. Soil quality differences in a row-crop watershed with agroforestry and grass buffers. Agroforestry Systems, 90 , 829-838.

WHALEN, J. K. & SAMPEDRO, L. 2010. Soil Ecology and Management, CABI.

WIESMEIER, M., MAYER, S., BURMEISTER, J., HUBNER, R. & KOGEL-KNABNER, I. 2020. Feasibility of the 4 per 1000 initiative in Bavaria: A reality check of agricultural soil management and carbon sequestration scenarios. Geoderma, 369.

WOLZ, K. J. & DELUCIA, E. H. 2018. Alley cropping: Global patterns of species composition and function. Agriculture Ecosystems & Environment, 252 , 61-68.

XU, Y., LEHMANN, L. M., DE JALON, S. & GHALEY, B. B. 2019. Assessment of Productivity and Economic Viability of Combined Food and Energy (CFE) Production System in Denmark. Energies, 12.

XU, Y., SMITH, J., SMITH, L., SANDOR, M., JITEA, M., GLIGA, A., BOREK, R., WAWER, R., PISANELLI, A., RUSSO, G., AUGUSTI, A., LAUTERI, M., CIOLFI, M., LEHMANN, L. M. & GHALEY, B. B. 2018. Manuscript on agronomic productivity and resource use efficiency for identification of knowledge gaps in INFS. In: GHALEY, B. B. (ed.) Optimization offood and non-food productivity in IFNS. SustainFarm.

ZHANG, Q. Z., WANG, C. K., WANG, X. C. & QUAN, X. K. 2009. Carbon concentration variability of 10 Chinese temperate tree species. Forest Ecology and Management, 258 , 722­727.

Appendix A - values of crop and tree parameters

This appendix features eight tables presenting values of the essential parameters for the crop and tree parameterisations used in the simulations. First, the parameters for the HS MAFF GSMs are presented, followed by the ones for the HS CFEC GSMs. Then, the parameters for the YS MAFF GSMs are presented, and finally the ones for the YS CFEC GSMs are given.

Hi-sAFe MAFF models

Abbildung in dieser Leseprobe nicht enthalten

Table A.1 shows values for essential crop parameters used in the HS MAFF GSMs. Table A.2 shows essential tree parameters used in these GSMs.

Table A.1: Values of the essential crop parameters except phenological parameters from the HS MAFF GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Table A.2: Values of the essential tree parameters from the HS MAFF GSMs. Parameters are explained in text only where the parameter names are insufficient as explanation.

Abbildung in dieser Leseprobe nicht enthalten

Table A.3 shows values for essential crop parameters used in the HS CFEC GSMs. Table A.4 shows essential tree parameters used in these GSMs.

Table A.3: Values of the essential crop parameters apart from phenological parameters from the HS CFEC GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Table A.4: Values of the essential tree parameters from the HS CFEC GSMs. Parameters are explained in text only where the parameter names are insufficient as explanation.

Abbildung in dieser Leseprobe nicht enthalten

Table A.5 shows values for essential crop parameters used in the YS MAFF GSMs. Table A.6 shows essential tree parameters used in these GSMs.

Table A.5: Values of the essential crop parameters from the YS MAFF GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Table A.6: Values of the essential tree parameters from the YS MAFF GSMs.

Abbildung in dieser Leseprobe nicht enthalten

TableA.7 shows values for essential crop parameters used in the HS MAFF GSMs. TableA.8 shows essential tree parameters used in these GSMs.

Table A.7: Values of the essential crop parameters from the YS CFEC GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Table A.8: Values of the essential tree parameters , from the YS CFEC GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Appendix B - parameterisation of crop growth phenology

This appendix describes how the growth phenology of crops was parameterised. First, it describes how a reference set of target times to a variety of stages for all crops was created. Then, it describes how this reference set was used to calibrate the crop phenology.

To create the reference set of target times, an extensive dataset was downloaded from the repositories of the central German weather service (DWD) (Deutscher Wetterdienst, 2020). The dataset included recorded times to different crop development stages for a large number of measurement localities, going back to the early 1950'ies. Measurement stations with a coastal climate influenced by the North Sea and connected waters were chosen for use. The stations were all located near the German-Danish border in Northern Schleswig. The data was transformed, so that averages across all years in twenty- or thirty-year-long periods of development times to the given developmental stages were produced, see table B.1 . The periods chosen were 1981-2010 winter wheat and -barley, and 1970-1990 for spring wheat, -barley and -oats. These were chosen because they represented the most recent data that was complete.

Table B.1: Table showing the dates used for the calibration of the stage phenology of the crops. Dates are in DD/MMformat.

Abbildung in dieser Leseprobe nicht enthalten

To calibrate the crop phenology, the DWD dataset was used in combination with the CFEC GSMs. The crops were simulated as a monoculture with no trees, and parameters of the crop files governing the temperature sums to different growing stages were varied. Eventually, average dates that the given crop entered the different stages in the model agreed well with the average dates calculated from the measured data. The comparison between model and measured data was challenging, since the development stages used for grain crops in HS does not resemble the development stages normally used for describing crop development, see table B.2 .

Table B.2: Comparison of stages from the data from the German weather service (DWD) and stages used by HS and YS, respectively, as well as an estimate of the corresponding stage on the Zadoks growth scale. Stages described in the same row are representations of the same real-world growth stage. Empty rows in any column signify that the scale in the column in question does not have a stage corresponding to the stage represented by the row. Topmost stages are earliest, and the crop development subsequently continues downwards. Tillage and sowing are omitted, because timings for these operations were determined using other sources of information than the DWD data. Beginning of grain filling and flowering were simulated as being concurrent.

Abbildung in dieser Leseprobe nicht enthalten

Appendix C - crop species simulated in different years

In this appendix, the crops grown in the real systems, and simulated in the GSMs, in the different years are presented. Table C.1 presents crops for the MAFF-sponsored trials, while table C.2 shows the crops in the CFEC systems.

Table C.1: Crop measurements used, and crops simulated, in the different years for the HS and YS MAFF- sponsored GSMs, respectively. If the crop simulated in HS and YS was the same for a given combination of trial and year, one crop is specified in the cell. If they varied, first the crop used in the HS model is given, then, after a slash (‘/’), the crop of the YS model is given.

Abbildung in dieser Leseprobe nicht enthalten

Table C.2: Crop measurements used, and crops simulated, in the different years for the HS and YS CFEC GSMs, respectively. If the crop simulated in HS and YS was the same for a given combination of trial and year, one crop is specified in the cell. If they varied, first the crop used in the HS model is given, then, after a slash (/), the crop of the YS model is given.

Abbildung in dieser Leseprobe nicht enthalten

Appendix D - values of parameters for growing system details

This appendix describes how parameters of lesser importance was set for all GSMs. First, some general points regarding the HS models are made. Then, values for the MAFF-sponsored GSMs are presented, and subsequently, values for the CFEC GSMs are given. Secondly, notes are mentioned regarding the YS models, leading into a presentation of values for the MAFF-sponsored and CFEC GSMs, respectively.

Details of Hi-sAFe parameterisation

HS has many variables, and most imaginable features of a real growing system can be accounted for in the model. It was generally attempted to use every feature and capability of HS appropriate to the context of the growing system in question. Furthermore, the model was parameterized with values closely reflecting the real systems wherever possible. However, the groundwater module was deemed unfeasible to use, since data on the level of the groundwater table in a specific location is not widely available from Northern European weather stations. Therefore, it was disabled during the simulation runs by setting the water table depth to -5 m for all days in every weather file.

The soil data acquired was transformed into the appropriate form. Many soil classification systems discern between the sediment fractions sand silt and clay. The soil simulation table in HS only considers the four elements sand, clay, limestone and organic matter, and the amount of silt in the soil cannot be specified. As the particle sizes of silt particles partially overlap with both those of clay and those of sand, it was decided that reported values of silt contents in data on the soils of the studied growing systems would be divided equally among the sand and the clay fractions. Thus, if a soil were reported to contain 60 % sand, 10 % silt and 30 % clay, the soil of the corresponding GSM would be parameterized as containing 65 % sand and 35 % clay. In the growing systems with shallow soil, the soil depth parameterisation reflected the actual soil depth. In systems with deep soil, soil depth was set to be deep enough to pose minimal restrictions to the growth of plants.

HS MAFF-sponsored GSMs

For the MAFF trials, the crops were reported to be “managed in the same way as commercial crops, receiving standard applications of fertiliser, herbicides, fungicides and insecticides as appropriate” (Burgess et al., 2005). I assumed that this meant standard applications of lime as well. For this reason, it was assumed that the pH of the soil was always 6.0. Furthermore, it was attempted to parameterise the management input files so that crop N-stress was minimised.

Table D.1 presents parameter values for minor, system-defining parameters used in the HS MAFF GSMs.

Abbildung in dieser Leseprobe nicht enthalten

HS CFEC GSMs

The CFEC is managed organically, and the soil in the systems receives organic matter that should be capable of buffering soil acidification from crop growing in the systems. Therefore, for all simulations, it was assumed that the soil had a healthy pH value of 6.0.

Because N-stress likely is a critical growth limitation in the real systems, in the simulations, it was not attempted to minimise N-stress. The essential input of N to the real systems is the growing and cutting of clover ley, which was simulated as one-year-long bare soil fallow where three applications of large amounts of plant residue were incorporated into the soil.

For simplicity, it was decided to build the HS CFEC GSMs as if every tree was a willow tree.

Table D.2 presents parameter values for minor, system-defining parameters used in the HS CFEC GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Details of yield-safe parameterisation

In the growing systems with shallow soil, the soil depth parameterisation reflected the actual soil depth. In systems with deep soil, soil depth was set to be deep enough to pose minimal restrictions to the growth of plants.

Challenges were encountered during the parameterisation of aboveground pruning. Sensitivity analysis showed that aboveground pruning only exerted minor influence on tree growth. Therefore, for the sake of simplicity, the pruning operations in the MAFF-sponsored trials were parameterized as lighter, annual pruning. YS does not support the simulation of root pruning, so this element was omitted in all GSMs.

YS MAFF-sponsored GSMs

Table D.3 presents parameter values for minor, system-defining parameters used in the YS MAFF- sponsored GSMs.

Abbildung in dieser Leseprobe nicht enthalten

Table D.4 presents parameter values for minor, system-defining parameters used in the YS CFEC GSMs.

Table D.4: Parameterisation of the minor, system-defining parameters of the of the CFEC used in the YS simulations.

Abbildung in dieser Leseprobe nicht enthalten

Appendix E - details on weather parameters

This appendix gives an overview of how the weather datasets used in the simulations were created.

In the simulation efforts, four different locations were considered: Cirencester, Leeds, Silsoe, and Taastrup (the location of CFEC). For the MAFF-sponsored systems, the weather data used originated from 5 different weather stations chosen for their proximity to the three growing systems in question, and for the availability of the data types needed. For the Cirencester trial, data from the Brize Norton weather station was used, for the Leeds trial, data from the Bradford and Bingley no. 2 stations was used, for the Silsoe trial, data from the Bedford and Rothamsted stations was used. For the CFEC, data from the weather station at Copenhagen University research farm, Hojbakkegaard, was used as well as data from the weather station at Holb^k Flyveplads and near Hoje Tâstrup.

The tables below indicate whether data for the varying parameters were real, measured data, or simulated values from CliPick. Table E.1 contains indications for the parameters used by HS, and table E.2 contains indications for the parameters used by YS.

Table E.1: Indications of the origin of the weather data used for HS simulations in the three MAFF-sponsored GSMs and the CFEC GSMs, respectively.

Abbildung in dieser Leseprobe nicht enthalten

Table E.2: Indications of the origin of the weather data used for YS simulations in the three MAFF-sponsored GSMs and the CFEC GSMs, respectively.

Abbildung in dieser Leseprobe nicht enthalten

[...]