Capturing spatiotemporal variation in salt marsh belowground biomass, a key resilience metric, through geoinformatics

Abstract

The Belowground Ecosystem Resiliency Model (BERM) is a geoinformatics tool that was developed to predict belowground biomass (BGB) of Spartina alterniflora in salt marshes based on remote sensing of aboveground characteristics and other readily available hydrologic, climatic, and physical data. We sought to characterize variation in S. alterniflora BGB over both temporal and spatial gradients through extensive marsh field observations in coastal Georgia, USA, to quantify their relationship with a suite of predictor variables, and to use these results to improve performance and expand the parameter space of BERM. We conducted pairwise comparisons of S. alterniflora growth metrics measured at nine sites over three to eight years and found that BGB grouped by site differed in 69% of comparisons, while only in 21% when grouped by year. This suggests that BGB varies more spatially than temporally. We used the BERM machine learning algorithms to evaluate how variables relating to biological, climatic, hydrologic, and physical attributes covaried with these BGB observations. Flooding frequency and intensity were most influential in predicting BGB, with predictor variables related to hydrology composing 61% of the total feature importance in the BERM framework. When we used this expanded calibration dataset and associated predictors to advance BERM, model error was reduced from a normalized root mean square error of 13.0% to 9.4% in comparison to the original BERM formulation. This reflects both an improvement in predictive performance and an expansion in conditions for potential model application. Finally, we used regression commonality analysis to show that model estimates reflected the spatiotemporal structure of BGB variation observed in field measurements. These results can help guide future data collection efforts to describe landscape-scale BGB trends. The advanced BERM is a robust tool that can characterize S. alterniflora productivity and resilience over broad spatial and temporal scales.