Exploring complementarities between modelling approaches that enable upscaling from plant community functioning to ecosystem services as a way to support agroecological transition

Promoting plant diversity through crop mixtures is a mainstay of the agroecological transition. Modelling this transition requires considering both plant-plant interactions and plants’ interactions with abiotic and biotic environments. Modelling crop mixtures enables designing ways to use plant diversity to provide ecosystem services, as long as they include crop management as input. A single modelling approach is not sufficient, however, and complementarities between models may be critical to consider the multiple processes and system components involved at different and relevant spatial and temporal scales. In this article, we present different modelling solutions implemented in a variety of examples to upscale models from local interactions to ecosystem services. We highlight that modelling solutions (i.e. coupling, metamodelling, inverse or hybrid modelling) are built according to modelling objectives (e.g. understand the relative contributions of primary ecological processes to crop mixtures, quantify impacts of the environment and agricultural practices, assess the resulting ecosystem services) rather than to the scales of integration. Many outcomes of multispecies agroecosystems remain to be explored, both experimentally and through the heuristic use of modelling. Combining models to address plant diversity and predict ecosystem services at different scales remains rare but is critical to support the spatial and temporal prediction of the many systems that could be designed. compared light partitioning simulated detailed 3D representations of with a radiation model followed the turbid-medium approach, with the plant canopy represented by one, two or ten more detailed of the in the of light partitioning in mixtures only slightly, the turbid-medium approach for estimating light competition at the canopy scale in crop models.


Introduction
New models are frequently developed for specialists in a field to answer specific scientific questions, without much interaction with other disciplines in the initial stages. During the past decade, however, modellers have integrated knowledge from multiple disciplines (e.g. micro-meteorology, environmental physics, ecophysiology, ecology, soil science) to better represent interactions between processes within plants, between plants, and between plants and their environment (e.g. Gauthier et al. 2020). The result is a distinct diversity of modelling approaches that can be used to The need to combine several modelling approaches, each with trade-offs in accuracy and generality, is crucial in all scientific disciplines and assumes that each model may improve understanding and predictions of ecosystem functioning. For instance, Evans et al. (2016) highlighted that the global models used to predict the geographic distribution of plant species throughout the world have low predictive power if they are not improved with process-based range models that predict impacts of environmental changes. Therefore, the need exists for accurate predictions of processes and more global and qualitative modelling approaches to understand an ecosystem, while also considering the feedback between different approaches, especially as the factors involved in ecosystem functioning are not necessarily the same for the spatial scales considered (Pearson and Dawson 2003;Xu et al., 2021).
Building connections between modelling approaches is particularly crucial in the context of the current agroecological transition, which involves in-depth changes to agricultural practices, with more complex and diversified agroecosystems and a multifunctional view of agriculture (Caron et al. 2014; Duru et al. 2015; Gaba et al. 2015). Increasing plant diversity is a mainstay of the agroecological transition and the cornerstone for "biodiversity-based agriculture" (Duru et al. 2015), which depends on agrobiodiversity at field, farm and landscape scales (Kremen and Miles 2012; A c c e p t e d M a n u s c r i p t 4 Prieto et al. 2015; Tscharntke et al. 2021). In these types of agriculture, ecological processes are fundamental to agricultural production, which requires particular focus on production-ecology trade-offs (Sabatier and Mouysset 2018). More than ever, modelling synergies must be identified to enable upscaling from plant functioning (i.e. ecophysiological processes and plant-plant interactions) to ecosystem services to support the agroecological transition (Tixier et al. 2013).
To illustrate how these modelling synergies and complementarities are essential to better characterize biodiversity-based agriculture, we focus on modelling species and cultivar crop mixtures instantaneous, daily, crop-cycle, rotation or long-term) and spatial resolution (e.g. plant, field or landscape, along with its multiple cultivated and uncultivated components) to use to represent the multiple interactions of interest. These considerations suggest that a single modelling approach is not sufficient to meet these legitimate expectations. Moreover, how these issues are addressed depends on which stakeholders use the models.
In this opinion article, we advocate that complementarities and coupling of different modelling approaches are critical to consider the complex and diversified agroecosystems involved in the agroecological transition, as well as to upscale from the plant and/or field scales to the ecosystem services targeted in diversified agroecosystems. Using several examples, we demonstrate that the complementarity between individual-based models (including functional-structural plant models (FSPMs)), crop models and physical or more qualitative or statistical models, improves understanding and facilitates simulating the functioning of crop mixtures and the ecosystem services for which they are designed. These modelling complementarities are discussed through the lens of crop mixtures or are integrated at larger scales to address three important modelling challenges: to i) quantify and understand plant-plant interactions and their underlying processes, ii) represent ownloaded from https://academic.oup.com/insilicoplants/advance-article/doi/10.1093/insilicoplants/diab037/6449487 by INRAE Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement user on 08 December 2021 A c c e p t e d M a n u s c r i p t 5 impacts of the environment and agricultural practices on the functioning of crop mixtures and iii) assess the ecosystem services provided by these heterogeneous covers (Figure 1).

Modelling plant-plant interactions in crop mixtures to assess ecosystem services at fine scales (plant and field)
Plant-plant interactions in crop mixtures are the foundation for the ecosystem services provided by diversified agroecosystems. Although these interactions can provide large-scale ecosystem services, for most processes they usually occur at a fine scale due to interactions between neighbouring plants or between plants and microorganisms. To illustrate this, we focus on complementarities between existing modelling approaches to simulate production and regulating services quantified at the plant and/or field scales.
2.1. Modelling plant-plant interactions to quantify underlying processes for production services One widely known advantage of crop mixtures is their potential to achieve higher yields due to more  Compared to these ecophysiological approaches, ecological approaches are particularly relevant for studying and understanding the functioning of systems in which multiple heterogeneous populations, such as crop mixtures, interact. However, representing mechanistically the processes that interact within these systems and quantifying the resulting ecosystem services requires knowledge and conceptual frameworks that are well theorized in ecophysiology and agronomy.
Thus, dialogue between these different disciplines -environmental physics, ecophysiology, agronomy and ecology -is crucial for modelling these agroecological systems (Evans et al. 2016;Brooker et al. 2021). In particular, the concept of "functional trait" commonly used in ecology and While FlorSys simulates aboveground crop-weed canopies, the ArchiSimple metamodel represents the trophic connection between above-and belowground growth, and the STICS soil submodels represent soil structure and climate and their effects on root growth (Figure 3). Illustrating the issues faced by coupling models or modules with different time or space scales, conversions had to be done to make optimal days of RSCone compatible with thermal time in FlorSys. This smart solution thus generates outputs and proxies that can be used to assess contrasting ecosystem services such as crop grain yield; weed-caused yield loss; weed seed production (as a proxy for future yield loss) and weed-based trophic resources for domestic bees (as one example of weed benefits), resource uptake or striga risk (Pointurier et al. 2021). It was necessary to develop working and modelling assumptions as the aim of this multi-faceted model was to cover a wide range of flora (including This research model was then used to identify agroecological mechanisms and provide decision support for farmers. This mechanistic and individual-based approach induces considerable algorithmic complexity and slow simulations; thus, using it in decision-support systems is time consuming, as it requires assigning many input variables and calibrating many parameters, particularly when simulating many diverse crops simultaneously. This is solved by aspects of metamodelling (Colas et al. 2020) that can identify potential changes to cropping systems that might improve their performance. However, a biophysical parent model is still required to provide biophysical explanations that farmers will accept ).
The authors reconstructed the functioning of a diversified agroecosystem by coupling models that could represent systems (the plant and its aerial and root structure, seeds, soil layers and their structure and microclimate) and mechanisms at similar scales. In particular, they integrated two aspects that are essential to understand and manage these types of agroecological systems: consideration of long-term processes (e.g. evolution of a seed bank) and impacts of management decisions on these processes and the targeted ecosystem services, with consequences that could occur over several years.

Assessment of a bundle of ecosystem services at the farm scale
Assessing the ecosystem services provided by diverse crop mixtures is challenging due to the many ecosystem services targeted by farmers and the diversity of crops to be investigated (Verret et al. 2020). Coupling models may be a promising solution to understand this complexity and diversity because it benefits from the strengths of diverse modelling approaches. However, predicting how management activities and changing future conditions will alter ecosystem services is rendered more complex by interactions (e.g. trade-offs, synergies) among multiple ecosystem services (Agudelo et al. 2020). More widespread use of process-based models to estimate ecosystem services could identify physiological processes, or even the traits, that influence interactions between ecosystem services. However, simulating the ecosystem services provided by crop mixtures requires representing their inclusion in crop rotations and long-term effects of the environment. This could ownloaded from https://academic.oup.com/insilicoplants/advance-article/doi/10.1093/insilicoplants/diab037/6449487 by INRAE Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement user on 08 December 2021 A c c e p t e d M a n u s c r i p t 12 be achieved by combining the knowledge provided by process-based models and using more qualitative models based on farmers' expertise.
In agreement with this idea, Meunier et al. (2022) designed a serious game to help users (farmers or students) explore and assess a bundle of ecosystem services (i.e. cereal and legume grain yield, cereal protein content, potential nitrogen supply to the next crop, maintenance of soil structure and pest regulation) provided by a wide range of binary cereal-grain legume intercrops (Figure 4). The serious game encapsulates a modelling chain that they constructed from three modelling approaches: (iii) A knowledge-based multi-attribute model built using DEXi software (Bohanec 2008) was used to turn attainable biomass into actual biomass considering pest damage and assessing pest-regulation services. Other multi-attribute models also enabled assessment of five more ecosystem services that result from the actual biomass of the cereal-legume intercrop at harvest and/or cropping-system features.
The serious game was designed to explore the ecosystem services provided by both common and less-common intercropping scenarios, and to promote debate and knowledge sharing among users.  From the examples listed above, different strategies can be identified to combine models at different 5 scales and predict consequences of plant-plant interactions, from local responses up to ecosystem 6 services at the cropping system and farm scales ( Figure 5). Besides direct coupling of models, which is 7 rarely feasible across all scales, we identified three particularly promising approaches to address this 8 issue: 9 ownloaded from https://academic.oup.com/insilicoplants/advance-article/doi/10.1093/insilicoplants/diab037/6449487 by INRAE Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement user on 08 December 2021 A c c e p t e d M a n u s c r i p t 13 • Inverse modelling, which connects models by identifying input parameters from simulated 10 data. This approach is common to many scientific disciplines (Evans et al. 2016) and uses 11 simulated datasets to determine parameter values from other models to supplement the 12 observed data available. Using simulated datasets to improve exchanges between models 13 and modellers is particularly valuable to facilitate parametrization of existing models, as 14 Although not yet developed for agroecosystems, hybrid modelling could also consider 40 Bayesian approaches that have been effective at aggregating different types of models and 41 data, including those that concern consequences of plant-plant interactions in natural 42 systems (Pagel and Schurr 2012). 43 These three broad categories are not mutually exclusive and can be combined to build original 45 models across scales. Each can help illustrate the potential of process-based models to assess certain 46 key ecosystem services (related mainly to crop productivity, biogeochemical cycling and weed 47 control). Moreover, the temporal scale at which processes and interactions occur and ecosystem 48 services are built may require long-term simulations ( Figure 5). Generally, when the spatial scale 49 increases (from plant-plant to the landscape), the temporal scale increases. However, modelling 50 could target more ambitious applications than those documented to date, such as more 51 comprehensive representation of environmental drivers (e.g. pests and pathogens, soil phosphorus 52 content, climate change) and greater detail in the relationships between plant diversity (crop, service 53 and weed plants) and biodiversity at other trophic levels in agroecosystems (pests and diseases). 54

Challenges and difficulties linked to modelling solutions 55
Reusing and coupling existing models faces several methodological and technical challenges. To be 56 effective, direct coupling and hybrid modelling often require developing specific adapters or new 57 model code. Too many inconsistencies between models, such as differences in temporal and spatial 58 resolutions, concepts and coupling variables, can make it more difficult to couple the models. The 59 coupling time step must be defined and be consistent with the time step of the interactions between 60 the simulated systems. This indicates that it may be necessary to increase (temporal upscaling) or 61 decrease (temporal downscaling) the time step of one of the coupled models; the latter assumes 62 knowing how to describe processes at a finer temporal resolution. Furthermore, the processes 63 considered can occur at different spatial scales (e.g. from field to watershed) depending on the type 64 of organisms and the factors involved, and can be influenced by multiple interactions. Modelling 65 platforms do not always have sufficient technical development to combine these contrasting 66 resolutions to describe systems and their functioning. Moreover, coupling models promotes dialogue 67 between disciplines (e.g. agronomy and hydrology) and thus requires agreeing on a common lexicon 68 or an ontology. 69

Modelling perspectives and opportunities 70
Many consequences of multispecies systems remain to be explored, both experimentally and 71 through modelling and theoretical studies. We advocate practicing both during the transition 72 towards more agroecological systems. Models cannot be developed without supporting data, and a 73 lack of reliable models hinders data analysis. This is particularly true regarding consequences of 74 plant-plant interactions, for which the magnitude and hierarchy of the major processes involved 75 ownloaded from https://academic.oup.com/insilicoplants/advance-article/doi/10.1093/insilicoplants/diab037/6449487 by INRAE Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement user on 08 December 2021 A c c e p t e d M a n u s c r i p t 15 remain hotly debated despite over 80 years of manipulative and observational studies (Brooker 76 2006; Weisser et al. 2017). This lack of understanding prevents identification of a consensual, much 77 less optimal, model structure. However, it also promotes the development of a variety of models to 78 test and benchmark interactions between mechanisms that act simultaneously (e.g. competition and 79 complementarity for resources, different forms of facilitation, physical and chemical signalling). In 80 this context, combining specific model developments with effective strategies to aggregate them 81 encourages parallel progress in key disciplinary issues (related to biophysical aspects and social 82 sciences in managed ecosystems), while still enabling integration of outputs relevant for predicting 83 ecosystem services at different scales. 84 Connecting data with models to develop diversified cropping systems provides an opportunity to 85 address issues involved in quantifying biodiversity-based services. As a part of managed ecosystems, 86 these services are scrutinized more closely than those in natural systems and benefit from 87 To this end, the ability to predict consequences of within-field diversity at different spatial and 98 temporal scales is required in order to assess the overall interest of various diversification scenarios. 99 A general belief about natural ecosystems is that plant diversity alone provides the ecosystem 100 services targeted, and that increasing species and genetic diversity in cropping systems should be a 101 goal to provide multiple services. However, how and why a particular arrangement of practices, or a 102 given range of diversity, should be chosen largely remain to be solved. By definition, managed 103 agroecosystems have an economic purpose and often target particular marketable products. From a 104 farmer's perspective, diversification thus has advantages (resilience) and disadvantages (not all 105 species are equal from an economic viewpoint; more complex management). Our opinion is that 106 combining models that can represent plant diversity and predict ecosystem services at multiple 107