Plant and crop simulation models: powerful tools to link physiology, genetics, and phenomics

Plant and crop simulation models are powerful tools for predicting the impact of climate change, innovative crop management practices, and new trait- or gene-based breeding technologies on the production of crops and agricultural systems. In this special issue, we gather a collection of review, opinion, and primary research papers that represent the current status and future of crop and plant simulation models. The models presented in these papers have been informed and improved by new hypotheses and concepts from (eco)physiology, fueled by phenomics data, and used to bridge the gap between genes/genotypes and phenotypes. Process-based crop simulation models (CSMs) are dynamic computational tools that simulate the development and growth of a crop in relation to environmental conditions (e.g. air temperature, soil water, evaporative demand, and atmospheric CO2 concentration) and management practices (e.g. sowing date, N fertilizer application, and crop residue). The first computational models of crop and soil processes were developed more than 60 years ago (see Bouman et al., 1996; Jones et al., 2017 and for historical perspectives). Over time, the development of CSMs has essentially been motivated by both their capacity to aggregate our ever-growing knowledge of crop functioning, and by the need to provide informed support for decisions/policy. While early models focused on canopy architecture, light interception, and C assimilation, subsequent efforts have considered a wider breadth of processes over the crop life cycle with a focus on development and growth dynamics, C partitioning, and water/N balances in the crop and soil (Boote et al., 2013). At the same time, and without much interaction with the crop-modelling community, various groups have developed plant models with an explicit description of 3-dimensional (3D) morphology and form of plants, known as functional–structural plant models (FSPMs; Vos et al., 2010; Prusinkiewicz and Runions, 2012). In contrast with CSMs, which generally model processes at the canopy level, FSPMs consider processes at the level of individual plants and/or organs, from which complex properties at the plant community level emerge. Different FSPMs have a focus either on the above- or below-ground parts of the plant. Most of them do not consider the whole plant life cycle, and instead only consider a limited number of processes, depending on the research questions being addressed. However, the breadth of processes considered in FSPMs has greatly increased in recent years, which now allows them to address questions related to the interactions of crops with environment and management that could previously only be addressed by CSMs (Evers et al., 2018). CSMs and FSPMs are complementary and can inform each other on processes acting at different scales. Both types of models are considered in this special issue. The long history of crop modelling has produced a large number of models. For instance, there are more than 40 wheat CSMs currently in use (Asseng et al., 2019b). Most models have evolved from a few landmark models such as CERES and EPIC in the USA or SUCROS in Europe (Box 1). CSMs differ according to their choice of relationships and hypotheses regarding process functioning and feedback loops, and their combinations of mechanistic components (e.g. photosynthesis, soil water transport) and functional components (also termed phenomenological; e.g. radiation use efficiency, stomatal conductance) (van Ittersum et al., 2003; Parent and Tardieu, 2014). Since 2010, the AgMIP (Rosenzweig et al., 2013) and MACSUR (Rötter et al., 2013) networks of crop modelers have carried out several landmark CSM inter-comparison studies. These studies have clearly highlighted both the potentials and the limits of current CSMs (Asseng et al., 2013; Bassu et al., 2014; Li et al., 2015; Ruane et al., 2017) and have led to important model improvements (Maiorano et al., 2017; Wang et al., 2017). The limits of CSMs can be seen in particular when crops are exposed to extreme climatic events and/or to multiple (biotic and abiotic) stresses, which are characteristics of low-input agricultural systems exposed to climate change. Prediction power can be greatly increased by averaging outputs of multi-CSM ensembles (Martre et al., 2015; Wallach et al., 2018), as is done in the field of climate forecasts (Tebaldi and Knutti, 2007). However, the multi-model ensemble approach does not allow focusing down to the explanatory level, and thus it cannot reveal which processes/traits at the plant level are the most important ones to target, for instance in breeding programs. Models are arranged horizontally according to their publication data. Those in grey are key concepts/process models. The models in blue were developed in the Netherlands; those in orange were developed in France; those in red were developed in the US; and those in green were developed in other regions/schools. Solid arrows indicate an exchange of knowledge/processes between models. Dashed arrows indicate models implemented in modelling/simulation platforms (boxes). Individual-based and/or functional–structural plant models are in italics. For histories of crop modelling in the US see Jones et al. (2017) and Hoogenboom et al. (1992), and for histories of crop modelling in the Netherlands see Bouman et al. (1996) and van Ittersum et al. (2003). There is a growing consensus that CSMs must evolve, and there are several directions in which they could do so (Boote et al., 2013). Recent decades have seen a tremendous amount of knowledge generated by plant (eco)physiology in its broadest sense. What components of that knowledge could be formalized and incorporated into CSMs? In which cases would this be useful or necessary? Answers to these questions are likely to be different depending on the purposes of the CSM and the targeted problem. Forward genetic tools (notably genome-wide association analyses and whole-genome predictions) have also made considerable progress in terms of resolution, accountability of population structure, and prediction power. Are CSMs ready to incorporate such information, as was proposed for quantitative trait loci (Yin et al., 2005; Chenu et al., 2009)? Early attempts have shown promising results (Technow et al., 2015), but much work needs to done to quantitatively link genotype to phenotype (Cooper et al., 2016). Reverse genetic approaches are also rapidly changing. For a number of years now, considerable efforts have been made to improve physiological processes, particularly with regards to photosynthesis and related functions (Kromdijk et al., 2016; Evans, 2013), and recently there has been spectacular progress using genome-editing technologies (e.g. South et al., 2019). How this progress might ultimately be translated into long-term yield increases has yet to be assessed, and modelling is certainly the best way to do it when improved crops will be faced with a variety of agro-climatic scenarios (Asseng et al., 2019a)—with the probable outcome that yield does not scale proportionally to the changes by which the functions or pathways are affected (Sinclair et al., 2004). Recent years have also seen considerable developments in phenomics. We are now able to acquire detailed information on plant or crop structure and functioning for hundreds of genotypes under control conditions or in the field. If data produced by phenotyping pipelines could also be used as (or as proxies for) state variables or parameters of CSMs, then the power and breadth of usage of CSMs could be greatly increased. This will be particularly the case if phenomics can demonstrate the capacity to account for the impact of the environment by parametrizing response curves/reaction norms (Tardieu et al., 2017). In turn, phenomics will also challenge current CSMs by providing an unprecedented amount of data to re-evaluate and improve the accuracy, robustness, and biological soundness of the models. Because CSMs are composed of modules that translate our knowledge of major processes at the plant and canopy levels (Box 2), there should be joint efforts to cross-evaluate these modules within common modelling frameworks. This can only be done if the community sets the tools and working principles to make modelling modules inter-operable, reusable, and open to anyone who wishes to try a new module and see how it compares to another. This should also prevent re-inventions when common modules can be re-used. The Farquhar, von Caemmerer, and Berry (FvCB) photosynthesis model (Farquhar et al., 1980) is a good example. Recent efforts are made by the plant science and crop modelling communities in this direction, such as the Crop in silico project (Cis; Marshall-Colon et al., 2017) and the Agricultural Model Exchange Initiative (AMEI; https://github.com/AgriculturalModelExchangeInitiative), are promising and should be encouraged. The equation gives a synthetic vision of CSMs (F) seen as an organized series (workflow) of processes (f), driven by genotype-dependent parameters (ai) accounting for the responses to the environment (Ej, input variables) given particular crop management practices (Mk). CSMs evolve based on improved knowledge of the main drivers of plant processes (1); leading to the proposal of (simple) formalisms (f). (Eco)physiological studies can also provide values for parameters (2). Phenomics provide species- or genotype-specific vectors for parameters or state variables (3). In turn, phenomics can propose developments that allow access to these variables and parameters (4). Forward genetics approaches feed models with traits variants (5) and models in turn provide the integrated outcome of these variants under various agro-climatic scenarios (6). Similarly, functional genomic approaches can be translated into model parameters (7) and CSMs can be used to predict their long-term impacts. Crop management practices (Mk) can be seen as levers to alter the environment (Ej) of the crop (8). This special issue, kindly published and supported by JXB, has its origin at a session organized at the annual SEB meeting in Gothenburg, Sweden, in July 2017. At this meeting, we advocated for a renewed discussion between modelers, physiologists, and geneticists around the avenues mentioned above. We formulated very specific questions that we believe are worth reiterating here. Is thermal time still a valid concept? How should we account for the acclimation of photosynthesis to temperature or light fluctuations? Should we include the circadian clock in CSMs? Do carbon partitioning rules need to be modelled or are they derived properties of sink and source activities? Are fertility/grain-filling steps correctly formalized? Are data acquired under steady-state conditions relevant for modelling in the field? How can we make models able to predict intraspecific genetic variability and genotype by environment by management interactions (G × E × M)? How can we best make data available to the community for model improvement? Almost two years after this session, we are pleased to share with the plant science community a series of papers that address several of these questions and that hopefully help to further our vision of the central position occupied by crop modelling at the crossroads between physiology, genetics, and phenomics. At the core of CSMs, thermal time sets the clock as a universal concept that is able to account for fluctuations in the rate of plant development. However, this concept is subject to criticism, particularly with regards to its robustness, since questions have been raised about changes in phyllochron between phenological stages, day and night, and different plant genotypes. Parent et al. (2019) revisit the concept of thermal time and show that apparent differences in temperature responses may be due to the confounding effects of evaporative demand, the range of temperatures, and the time interval at which measurements are taken. Different from the dynamic, time-based models presented in other papers in this special issue, Zardilis et al. (2019) propose an agent-based modelling strategy to predict the fitness of the model plant Arabidopsis thaliana grown under fluctuating environmental conditions. Their approach is appealing and could be worth exploring for predicting yields in crops. CSMs are very diverse, depending on the schools of thought behind them, their history, or even the programming language in which they are written. Even for a single crop such as winter wheat, there exist dozens of models, based on converging or even diverging modelling principles (Box 1). A major structural feature of this diversity is the way carbon capture by the plant is formalized, either through the concept of radiation-use efficiency or by accounting for photosynthesis processes through the FvCB biochemical model. Among the former, one of the most widely used CSMs is APSIM, which has been developed for more than 20 years in Australia (Holzworth et al., 2014). In this special issue, Wang et al. (2019b) review the current and future challenges facing APSIM, in particular the prediction of the phenotypic outcomes of G × E × M interactions and incorporating genetic information down to the QTL or gene level. Progress in our understanding of the major drivers of processes present in CSMs provides avenues for proposing improvements, with potentially less empiricism and more process-based formalisms. For instance, various experimental evidence has converged to indicate that thermal time is not the only driver of leaf appearance. While this is well acknowledged in the literature, it is classically corrected empirically in CSMs and often with abrupt changes in parameters at some developmental or time points (Jamieson et al., 2008, Abichou et al., 2018). Baumont et al. (2019) hypothesize and then demonstrate that the leaf appearance rate is reversibly and rapidly increased or decreased by the source–sink ratio of the plant, which is approximated through the photothermal quotient, i.e. the ratio of intercepted radiance (source) to temperature (sink). They plug a new formalism into a wheat CSM (SiriusQuality) and show the benefits both in terms of predictability and the physiological relevance of the formalism. Another example of improvement of processes concerns the acclimation of photosynthesis to temperature or light fluctuations which to date, are not accounted for in most CSMs. Yin et al. (2019) explore the hypothesis that acclimation of photosynthesis to temperature relates to the rapid partitioning of leaf N between different fractions, and they propose an update of the FvCB model on that basis. In the same vein, Pao et al. (2019) introduce a protein-turnover module to account for the acclimation of photosynthesis to light fluctuations that works well under contrasting N regimes. As noted above, genomic selection offers the possibility to decipher the genotype-to-phenotype link down to the level of genes and single-nucleotide polymorphisms. Echoing this, both Wang et al. (2019a) and Kadam et al. (2019) show how whole-genome predictions can be used to parameterize canopy growth of a process-based whole-plant model. CSMs are also useful tools for deciphering the influence of structural or physiological traits on crop performance and for proposing ideotypes that are adapted to targeted environmental scenarios. Two papers in this special issue show how CSMs can be used to help design wheat ideotypes after implementing new functions, related either to drought tolerance (Senapati et al., 2019) or to early vigour (Zhao et al., 2019). Illustrations of the link between phenomics and models are given by De Swaef et al. (2019), and the authors address the design of phenotyping experiments for accurate estimations of model parameters based on virtual experiments. Growing mixtures of crops and/or cultivars is a promising avenue for buffering fluctuations in production when faced with annual variations in resource availability or pest attacks. In recent years, several groups have started to design/adapt process-based simulation models to explore the outcomes of crop/genotype mixtures (Gaudio et al., 2019). FSPMs are particularly suited to study the heterogeneity in space and time of resource capture in heterogeneous canopies. In this special issue, Evers et al. (2019) discuss how and why FSPMs can further our knowledge of the functioning of crop mixtures, and how properties that emerge from FSPMs can be used to improve CSMs beyond monocultures. Faverjon et al. (2019) propose a FSPM for optimizing species composition in grassland, and its generic nature makes it applicable for other species mixtures. Chen et al. (2019) show that a model-based phenotyping approach can be used to predict the outcome of genotype mixtures. CSMs also need to evolve by incorporating new variables. Among these, wind is most often overlooked. Burgess et al. (2019) review modelling solutions to account for this important factor in FSPMs. Crop modelling is all about integrating knowledge of different processes by formalizing each of them and connecting input and output variables. Two papers in this special issue propose new whole-plant models that have each been built up from one single process-based core model: either photosynthesis in rice Chang et al. (2019), or berry growth in grapevine Zhu et al. (2019). In line with the idea of connecting modules with the purpose to scale-up temporally and spatially, Passot et al. (2019) propose a strategy for the modelling community to ‘connect the dots’, and make the link between small scales (e.g. root anatomy, soil micro-structure) and integrated functions of the plant (e.g. transpiration) over the long term. Finally, acknowledging that models are everywhere, Millar et al. (2019) advocate that the plant science community should be better organized to feed the models with formalisms, data, and parameters. Our own analysis is very much in line with this (Box 2). The wide variety of topics covered by this special issue thus illustrates the ever-increasing role of CSM. They are tools to make physiology and genetics more predictive by bridging the gap between genotype and phenotype. They can obtain their parameters from physiological studies (e.g. hydraulic architecture) and their hypotheses can be challenged by physiology/functional genomics approaches (e.g. using mutants or omics). Alternative modeling hypotheses should be tested using plug-and-play, shared, and re-usable modules, together with data, formalisms, and parameters available within an open and transparent repository. All this probably requires an updating of the agenda of the different communities within the fields of plant science and agronomy, the boundaries between which are tending to become more indistinct. Navigating a way through the many options offered to agriculture to face the current challenges of food security, sustainability, and climate change presents a complex task. As recently advocated by Asseng et al. (2019a) and Chenu et al. (2017), plant and crop simulation modelling is certainly a key component to consider.