COS 108-1 - Is salt marsh ecosystem resilience driven more by changes in space or time?

Abstract: Coastal biophysical models are critical tools to assess resilience and inform management of ecosystem service preservation. The Belowground Ecosystem Resiliency Model version 1.0 (BERM1.0) is a recently developed geospatial informatics tool to characterize salt marsh resilience and productivity, calibrated for South Atlantic salt marshes dominated by the marsh grass Spartina alterniflora. BERM1.0 estimates above- and belowground biomass (BGB) metrics at a monthly timestep and 30 m resolution using remote sensing, climate, and tide information through a machine learning approach. However, ecosystem models can only be confidently applied to novel areas when conditions are within the range of model calibration data, and ground truth data collection in salt marshes of high spatiotemporal variability is costly. Due to a limited calibration dataset, BERM1.0 had narrow applicability. Our goals were to advance BERM from version 1.0 to 2.0 for wider application across coastal areas to characterize spatiotemporal salt marsh productivity patterns. To do this, we expanded the ground truth calibration dataset, recalibrated models to version 2.0, and analyzed the impact of spatial and temporal variation on model performance. The BERM1.0 calibration dataset included ground truth data from four S. alterniflora salt marsh sites during the growing season. To calibrate BERM2.0, we expanded this dataset to nine S. alterniflora salt marsh sites sampled throughout the year. This represents a wider range of hydrodynamic and biophysical conditions as sites experienced different elevation gradients, tides, and watershed inputs. It also accounts for seasonal variation in reflectance-vegetation relationships. BERM performance was improved with the advancement to version 2.0. Root mean square errors (RMSEs) of preliminary BERM2.0 BGB estimates were lower than those of 1.0 when tested against a split of the calibration data (1.0: 389 g/m²; 2.0: 277 g/m²) and at a novel site (1.0: 378 g/m²; 2.0: 364 g/m²). Model goodness-of-fit metrics indicated that spatial variation impacted model performance more than temporal variation. The range of RMSEs among sites was 442 g/m², while that among years was 84 g/m². Average correlation between predicted and measured BGB in testing data among sites was 0.63, and among years was 0.89. By advancing BERM as a geospatial informatics tool, we can generate more accurate estimates across novel areas. Further, we found evidence that salt marsh vegetation metrics vary more across space than over time. These findings can help us better evaluate salt marsh resilience to sea level rise and other stressors in near-real time.