BioEarth: Envisioning and developing a new regional earth system model to inform natural and agricultural resource management

The overarching goal of BioEarth is to improve understanding of the interactions between coupled C:N:H₂O dynamics and human actions at regional and decadal scales under global change to 1) better understand the role that resource management actions have in impacting earth system dynamics, and 2) inform resource managers about the consequences of their decisions on the earth system, with a particular focus on quantifying uncertainties, environmental feedbacks, and economic and environmental tradeoffs. BioEarth enables users to quantify the impacts of human decisions on greenhouse gases and other atmospheric pollutants, water quantity and quality, terrestrial ecosystem health, and economics through simulating the management of cropping systems (e.g., crop selection, irrigation, fertilization, and residue management), forested ecosystems (e.g., thinning and restoration), rangeland ecosystems (e.g., grazing and restoration), water supply management (e.g., reservoirs, water rights curtailment, and water transfers), and air quality (e.g., regulation of emissions of pollutant and pollutant precursors from cars, power plants, and industrial facilities). This information, when combined with other more issue-specific decision-support tools and the constraints (e.g., economic, regulatory, or physical) and priorities of a decision-maker, allows for improved decision-making.

Varying levels of integration from one-way (a.k.a., “offline”) to two-way (a.k.a., “online” or “full”) coupling exist in this framework. The atmospheric and terrestrial models can either be linked offline or fully-coupled at 30- to 60-min time-steps. Many of the other components are coupled at longer time-steps (e.g., water rights curtailment decisions occur weekly). The Kepler scientific workflow toolkit (kepler-project.org) is utilized for implementing and automating BioEarth workflows.

BioEarth has a modular/flexible framework and is designed such that only some of the components may be utilized for the specific research question under investigation; e.g., for a question involving quantification of the impacts of water scarcity on irrigated agriculture without considering atmospheric feedbacks, the interactions between land surface processes, reservoir operations, and interruption of water rights can be simulated without tight coupling to atmospheric models. In determining which components to include, our goal is to match the level of model integration to the interconnectivity of the decision-process under investigation. The level of integration needed for informing specific management activities is partially determined through feedback from our stakeholders (see Section 4); e.g., agricultural producers are asked about the extent to which various environmental factors play a role in their decision-making. This approach minimizes model complexity and computational resources while still capturing key interactions. Furthermore, incorporating too much complexity (for a specific question) into an integrated model can have negative consequences. As an example, coupling a land surface model to an atmospheric model can result in large biases and reduced performance in informing management decisions, as compared to driving a land surface model with observed meteorology or with bias-corrected atmospheric model output (Liu et al. 2013). This loss of accuracy may prohibit a coupled model’s usefulness for specific management decisions. Regardless of the complexity of the model chosen for each application, when a coupled model is used for informing decision-making, rigorous testing is required to evaluate model results and enhance the usability of model outputs.

We take advantage of the CESM flux coupler that facilitates coupling of earth system components to allow data exchange between, e.g., atmospheric and land surface models. This flux coupler is modular and allows for several options for each of the land and atmosphere components. For BioEarth, we include two additional land surface options (see Section 3.4): 1) VIC-CropSyst, and 2) Upscaled-RHESSys. The flux coupler provides flexibility for the atmospheric and land surface models to be applied on their own grids and spatial resolutions. This is particularly important for coupling WRF to Upscaled-RHESSys, which runs on variably-sized watershed units rather than grid-cell units. For coupling with an atmospheric model, VIC requires code restructuring to adopt a space-before-time structure so that state variables for all grid-cells are updated before advancing to the next time step, which has been accomplished in other projects (Leung et al. 2011), but requires updating for BioEarth developments. The BioEarth project utilizes this modified version of VIC but with modifications and updates at the University of Washington through the Regional Arctic System Model (RASM) project (Maslowski et al. 2012).

For applications involving atmospheric chemistry, the CMAQ model can either be run offline from or fully-coupled to WRF. The fully coupled version accounts for radiative feedbacks between chemistry and physics by considering ozone effects on long-wave radiation and aerosol effects on both short- and long-wave radiation (Wong et al. 2012). For BioEarth, the CESM framework will be adapted to pass hydrologic information to MEGAN to determine biogenic emissions needed by CMAQ.

The modular nature of the BioEarth framework allows users to apply the land surface model most appropriate for each research question. Two unique land surface models are in development for BioEarth: 1) VIC-CropSyst primarily for applications involving intensely-managed cropping systems, and 2) Upscaled-RHESSys primarily for applications involving less intensely-managed forest and rangeland ecosystems, but with eventual capabilities to handle cropping systems. RHESSys also includes capabilities to simulate urban environments, allowing for investigation around questions that involve these landscapes, such as urban expansion.

VIC and RHESSys were originally developed for different scales and applications. VIC was developed as a land surface model that can be used with atmospheric models and includes detailed process-based parameterizations for water and energy fluxes with the flexibility to run at sub-daily time-steps, both of which are necessary for atmospheric coupling (Liang et al. 1994). Like all macro-scale land surface models, VIC is coarse in its representation of lateral hydrologic processes. To reduce computation time and avoid breaking model assumptions, grid-cells are necessarily large (~10⁷-10¹⁰ m²), and sub-grid heterogeneity (e.g., in land cover, elevation, saturated extent, and snow cover) is represented implicitly in space through empirical relationships (Fig. 3). Lateral fluxes are not specifically simulated within VIC while a post-process is available for streamflow routing. Conversely, while RHESSys is less detailed than VIC in representing vertical fluxes, it is more detailed in space, capturing surface/subsurface lateral flow between neighboring patches (~10¹–10⁴ m²). Below, we describe the development of the two BioEarth land surface models which are based on VIC and RHESSys, with elements of CropSyst. VIC-CropSyst’s strength lies in capturing land-atmosphere interactions, particularly as related to agricultural activities; while Upscaled-RHESSys’ strength lies in capturing hydro-ecological processes that are governed by relatively fine-scale spatial heterogeneities and lateral hydrologic connectivity, e.g., the role of soil moisture redistribution in impacting ecological hot-spots in terms of C:N:H₂O dynamics.

Open image in new window

The water resources of the CRB are heavily managed and this human alteration of the hydrologic cycle needs to be considered to better estimate seasonal available water supply. We take a two-pronged approach involving 1) an offline reservoir model (ColSim) that has detailed information on reservoir operating rules and, 2) a generic store-and-release algorithm that can be fully-coupled with the hydrologic model. The generic algorithm will estimate storage and releases either based on time-series results from the offline reservoir model or on empirical calculations similar to Voisin et al. (2013) or Biemans et al. (2011). This two-pronged approach gives the flexibility of choosing the framework best suited to a specific question.

NEWS sub-models have been successfully applied at sub-basin spatial scales (e.g., Harrison et al. 2010) for various forms of C and nutrients. In BioEarth, NEWS is utilized and evaluated for sub-basins within the CRB to predict seasonal fluxes, sources, and sinks for various forms (inorganic/organic, particulate/dissolved) of N and C. NEWS takes as input simulated hydrologic and N fluxes from the terrestrial models, meteorological information from WRF, and water storage from the reservoir models.

The purpose of integrating economic models into mechanistic biophysical models is to base human action and their influences on biogeochemical cycles (e.g., from land use, nutrient loading, and irrigation) on microeconomic foundations rather than on arbitrary rules (Harou et al. 2009). A two-pronged approach will be taken to model economic behavior by utilizing 1) a CGE model of the regional economy and 2) an ABM. While both work from a foundation of economically-optimizing decision-makers, they differ significantly in terms of temporal/spatial disaggregation and model structure. The CGE and ABM models will be run separately but integrated on annual time-steps. Multiple groups have applied CGE modeling of regional economies to represent agricultural production (for a review, see Harou et al. 2009). CGE modeling is suited for capturing the complex array of possible economic responses to changing conditions, including input/output substitution and investment. Hydrologic models inform water availability, and crop growth models are used to parameterize crop production functions for varying levels of deficit irrigation under a range of atmospheric conditions. Input use determined by the CGE model defines the timing of water use and other inputs, including fertilizers that can be fed back into other models. Sub-regions are differentiated by their parameterization and constraints, which is usually important for representing production in irrigated agriculture where water is a limiting resource.

In many instances the lack of spatial resolution in CGE models becomes problematic. While the development of sub-regions within the regional CGE model can account for spatial heterogeneity in growing conditions and resource constraints, they usually involve a significant level of spatial aggregation and are ill-suited towards modeling spatially-dependent decision-making. This is an important limitation because of the extensive spatial interactions related to water movement that result from farm-level adaptation, water transfers, or any other changes in water management. This weakness provides the motivation for the development of an ABM.

ABM allows for improved modeling of nutrient use and water quality impacts by more precisely locating each agent in space relative to water systems; and it is possible to separate the region of study into grid-cells that match those of the biophysical models. This allows for full coupling between human and environmental models. Adoption of approaches like ABM among economists remains small relative to more accepted economic modeling frameworks because they typically require a more simplistic representation of economic decision-making. However, they are becoming more common as recognition of the importance of spatial interactions has grown (Irwin 2010).

For stakeholders to be engaged in the model development process, opportunities for stakeholders to influence how research questions are prioritized must be created, and bi-directional communication between researchers and stakeholders must be facilitated. A series of stakeholder advisory workshops organized around specific resource management issues are occurring between the years of 2013 and 2016. Each workshop focuses on different management issues of critical concern to the region, including N and C management, water quantity/quality, forest and rangeland management, and air quality. Stakeholders participating in these workshops include a broad set of individuals who could benefit from using model results to inform their decision-making; these include representatives from industry, government agencies (federal, state, local and tribal), and environmental and other advocacy organizations. These workshops are designed to allow for in-depth discussion of stakeholders’ issue-specific information needs and preferences for how model outputs, complexity, and uncertainty might be best communicated. The four goals of these stakeholder advisory meetings are to 1) elicit stakeholders’ insights, perspectives and recommendations to inform prioritizations of model development; 2) communicate the potential value and utility of the model to stakeholders and provide them with an enhanced understanding of the model development process, complexity, and uncertainty of outputs; 3) establish positive, productive relationships and enhance mutual understanding between a broad range of stakeholders and the modeling team; and 4) increase researchers’ understanding of the factors driving resource management decisions and the information needs of decision-makers to help the modeling team understand how stakeholders prioritize various environmental and economic concerns.

Based on the interactions, feedback and surveys during and after the first two stakeholder workshops that occurred in February 2013, an innovative system for analyzing and acting upon stakeholders’ feedback was devised. The stakeholder input was categorized into discrete recommendations about model components, capabilities and scenarios that could be addressed, and these recommendations were arranged in a spreadsheet with details about frequency of recommendation, the number of individuals with expertise in that specific area, and the spatial and temporal scales at which each recommendation would need to be addressed. Once this database was created it was shared with all of the BioEarth modelers and scientists, and served as a focusing tool for a project-wide process of prioritizing the recommendations according to the BioEarth project scope and available resources.

The communication team is also analyzing the perceptions and understandings of stakeholders and scientists throughout the research process using a temporal series of surveys and interviews, with a focus on 1) the scientists’ perspectives on stakeholder engagement; and 2) the stakeholder’s and scientists’ perceptions of the utility and relevance of the integrated model.

As stakeholder engagement in environmental modeling is increasingly expected by funding agencies, understanding the range of perceptions on the value of stakeholder engagement can help facilitate productive interactions. Acknowledging this, the BioEarth communications research team began with analysis of the scientists’ perspectives on stakeholder engagement as well as model utility during the first year of the project (Allen et al. 2013). Results demonstrate a broad range of perceptions about the value of the stakeholder engagement process and varying expectations for the production of decision-relevant information within BioEarth. Researchers who anticipate the model to be relevant to policy decision-making rank communication with stakeholders as a central challenge in the project, while those researchers who are thinking about the model as something primarily relevant to academic audiences tend to focus on technical challenges associated with model integration rather than the model’s practical or social relevance. The communications team is continuing to conduct surveys and interviews with researchers and stakeholders to understand their perspectives on model utility and the stakeholder engagement process.