You are here:
Perspectives on improving the transferability of geospatial models of aquatic condition to watershed management
Citation:
Hill, Ryan A, S. Leibowitz, P. Ringold, E. Fergus, AND B. Rashleigh. Perspectives on improving the transferability of geospatial models of aquatic condition to watershed management. Virtual Annual Meeting of the Western Division of the American Fisheries Society, NA, Virtual, May 10 - 14, 2021.
Impact/Purpose:
Advances in the availability and geospatial data have allowed us to model and map several measures of instream ecological condition across the US. However, turning these maps and models into management scenarios is a major challenge. In this talk, we will describe the difficulties with associated with making management scenarios and decisions with large-scale geospatial modeling. We will also explore several possible solutions to this issue by describing recent work within the literature and research that is currently being conducted by EPA staff and affiliates. With this presentation, we hope to further discussion with the broader scientific community on improving the transferability of spatial models to resource management. This study supports the development of spatial interpolations of aquatic condition to improve economic research being conducted by the NCEE. This work presents progress on an FY21 deliverable under SSWR 1.2.2 (“Empirical models to interpolate benthic macroinvertebrate observed/expected ratios, or other biological indicator(s) of aquatic ecosystem health, from NARS stream and lake condition to HUC12 or HUC8 units over the conterminous US (CONUS) FY21”).
Description:
The influence of watershed land use on instream physical, chemical, and biological conditions is well documented. Recent advances in the availability of watershed data has allowed for modeling and spatial interpolation of several such features across the conterminous US. While these models and maps provide insights into the distribution of resources, their condition, and the land uses influencing them, translating these insights into management scenarios is challenging. This difficulty is due to several factors including informational coarseness of land use metrics, the irreversibility of some land uses on the landscape, or both. For example, percent watershed agriculture is often an important covariate in models, but agricultural intensity and impacts vary widely among regions and over time without changes in extent, and large-scale reversal of such land use is unlikely. Further, it is unknown whether popular modeling techniques (e.g., random forests) or models based on spatial differences among sites in land use are appropriate for predicting future outcomes of management. Addressing these challenges will require careful consideration of current limitations in both data and modeling. For example, better accounting of land use intensities (e.g., fertilizer application), the instream stressors they produce (e.g., excess instream nutrients), as well as mitigating factors (e.g., watershed wetlands) could improve the transfer of spatial models to management. Likewise, models that confer mechanistic understanding (e.g., process-based models) may also improve model transferability to management. Such refinements can be labor intensive and often offer little or no returns in model performance (e.g., r-squared), but their insights for management may outweigh these concerns. In this presentation, we review recent work and provide our perspective on additional research needed to improve the transferability of spatial models to the management of aquatic resources.