A framework for incorporating ecology into Earth System Models is urgently needed

This commentary considers the benefits of a new framework to incorporate ecological processes in Earth System Models (ESMs) to both Earth system science and to ecology. Adding ecological processes to ESMs skillfully will likely improve the long-term performance of these models. The rigor required to achieve this will prompt ecologists to test complex ecological hypotheses on regional and global scales. Some candidate processes are suggested. A few years ago, I heard Colin Prentice speak at a conference—he quipped that one sure fire way to write a splashy paper was to identify a new Earth System Model (ESM) process, demonstrate its profound importance to global climate, and claim that all models should henceforth include it. This semi-serious critique begs an important question; “How do we decide which processes should be included in ESMs?” ESMs are designed to simulate global carbon, water, and energy cycles and so must include ecological processes that influence them. In this issue of Global Change Biology, Kyker-Snowman et al. lay out a framework for how we should decide which ecological processes should be included. Interactions, connections, feedbacks, and complexity are hallmarks of ecology, but it would be unwise to add endless complexity to ESMs. The criteria of Kyker-Snowman's framework are, first that new ecological processes should influence Earth's climate on a large scale or that the process must result in changes to the carbon, water, or energy balance of ecosystems. Second, any new process cannot require more of the model than the model can currently provide. For example, leaching of nutrients cannot be added to a model without a nutrient cycle. Third, there should be sufficient understanding of the process and data to test the process globally; adding poorly established theory or theory that cannot be independently verified will cause potentially serious and unquantifiable bias. Fourth, the new processes must be governed by mathematics that are within reach of our current computational capacity and fifth, there must be a dedicated community of researchers to develop, test, and maintain the process in the model. These last four criteria may seem strange to ecologists; surely an ecological process that is important to climate must be included in ESMs? Perhaps, but these criteria put focus on what is tractable first. Moreover, Kyker-Snowman et al. call for a new research paradigm that challenges ecologists to collect the data required to meet the demands of global testing. Their call for tighter collaboration between ESM science teams and empirical and theoretical scientists should reduce the practical limitations that currently stifle progress. If we can add ecological processes to ESMs skillfully, we would improve model performance over the decadal to centennial time horizon. If you analyzed the code of the most used set of ESMs, the processes and calculations of energy balance would be many times more complex and realistic than the processes and calculations of vegetation dynamics, carbon allocation or soil nutrient cycling. Short-term measures of carbon, water, and energy balance are often poor indicators of interannual or long-term productivity (Richardson et al., 2010; Montané et al., 2017). The amplifying and stabilizing feedbacks that control what plants can grow in a place and how large a leaf area a location can support are governed by ecological processes that are inadequately represented in ESMs. Some processes are missing because there are competing ecological hypotheses, others are based on well-founded and widely accepted mechanisms. In terrestrial systems, tipping points are often observed as shifts in the type or nature of vegetation and they can be caused by some ecological resource passing a threshold or by ecological disturbances. The large-scale climate effects of widespread changes in vegetation cover are well established. However, ESMs have trouble predicting these ecological tipping points because they do not include the ecological processes that govern long-term ecosystem function. On the decadal to centennial timescales relevant to ESM projections, there are many candidate processes that are worth investigating rigorously (Figure 1). The framework proposed recommends that candidate ecological processes should be tested in isolation using simple models before testing in ESMs. Simple models typically ignore or parameterize core Earth System processes, for example the Simplified Photosynthesis and Evapotranspiration model lacks meaningful energy balance or nutrient cycling (Zobitz et al., 2008). However, simple models can be useful in studying the ecological controls of ecosystem fluxes (Richardson et al., 2010; Roby et al., 2019; Zobitz et al., 2008). Their simplicity facilitates evaluating different model structures, or ecological hypotheses, using information criteria techniques that are widely used in ecology (Burnham & Anderson, 2004). ESMs can also be deployed in point mode to test different model structures or hypotheses (e.g., Montané et al., 2017). It would be very useful to modularize processes in ESMs so that different representations of new processes could be more easily tested (Fisher & Koven, 2020). Even with careful study of ecological processes in simple models, implementation within ESMs can cause complex and unexpected patterns to emerge as the new process interacts with the existing set of Earth system processes. Infusing ecology into ESMs would benefit the study of ecology. The framework outlined by Kyker-Snowman et al would compel ecologists to rigorously quantify ecological processes that control long-term function of ecosystems. We know many of the processes that dictate ecosystem function. However, many are represented by ecologists as relational diagrams rather than mathematical rules (Chapin et al., 2011). Ecosystem processes are controlled in part by state factors: climate, geological parent material, topography, the species that could exist in an ecosystem, and time since a substantive ecological change. These factors are locally insensitive to the ecosystems themselves. The complexity builds, however, because the effects of state factors on ecosystems are modulated by interactions with vegetative characteristics, or traits, which may include features of leaf thickness, rooting depth, deciduousity, mycorrhizal association, and so on (Chapin et al., 2011). Ecological function has long been known to oscillate in cycles regulated by sometimes complex feedback mechanisms (Hutchinson, 1948), and ecosystems have self-stabilizing or amplifying mechanisms, that complicate how short-term responses to meteorological drivers translate into annual, decadal, or centennial responses (Chapin et al., 2011; Briggs & Walters, 2016). Combining models and data helps us make more useful measurements and by probing the spaces where models fail, we can infer which processes might better explain observations (Roby et al., 2019; Zobitz et al., 2008). Ecologists and Earth system scientists have greater understanding and more relevant data available to them than at any point in the past. Advances in biometeorology over the last two decades have improved our ability to measure carbon, water, and energy exchange in ecosystems (Novick et al., 2018) and, as Kyker-Snowman et al. explain, this has been mirrored by advances in large-scale modeling. Coordinated environmental observation networks like AmeriFlux, the Integrated Carbon Observation System, the National Ecological Observatory Network, the Long-Term Ecological Research network, and ‘network of networks’ like FLUXNET, provide an enormous amount of globally distributed ecological information. To achieve all of this, we need a shift in how ecologists and Earth system scientists collaborate. The various specializations in ecology and Earth system modeling are challenging and require hard work, but specialization can lead to the isolation of the two communities to the detriment of both. Ecologists are highly specialized scientists who are rightly trained in not just how ecology works the way it does but also why. Modeling the Earth system requires special skills also: detailed knowledge of mathematics, programming, physics, atmospheric science, and ecology or ocean science. Kyker-Snowman et al. provide a useful set of skills and resources for ecologists who wish to cross-train in model development and suggest that modeling experts revise their own practices and training. The paper recommends resources to learn efficient programming languages, software design, model structural design, and parameterization as well as robust means of evaluating model outcomes. As a start, I would recommend that modelers who wish to add more ecological processes to their work should keep a copy of the following texts in close reach—Principles of Ecosystem Ecology (Chapin et al., 2011), Community Ecology (Morin, 2009), Plant Variation and Evolution (Briggs & Walters, 2016). Cross-training across empirical and modeling approaches is important. While not every ecologist can become an expert on modeling and not every modeler can become an ecologist, having enough background to communicate effectively would be excellent footing to start adding the appropriate level of ecological detail to ESMs. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.