Evaluating Model Transferability and Scaling in Adjacent Subtropical Wetland in Southern Florida, U.S.

Understanding the model transferability and scaling is necessary in development of global soil prediction models. However, research gaps exist on how a model developed in a specific area transfers to another area and how model predictor variables change. Our objectives were to (i) develop prediction models for soil total phosphorus (TP) utilizing remote sensing (RS) images, (ii) identify the environmental variables controlling the spatial distribution of soil TP, and (iii) investigate the model transferability among two similar freshwater ecosystems. We collected a total of 108 and 66 soil cores from the top 10 cm in Water Conservation Area-2A (WCA-2A) and Water Conservation Area (WCA)-3A North (3AN), respectively, the Florida Everglades, U.S. Random Forest (RF) models to predict soil TP were developed using topographic, hydrologic, geophysical properties, and spectral data derived from RS images including SPOT (10 m) and MODIS (250 m). We applied the RF models developed in WCA-2A to WCA-3AN (and vice versa) to test model transferability. In addition, prediction models developed in sub-regions (WCA-2A and WCA-3AN) were upscaled to the whole study region and RF soil prediction models were downscaled from the whole wetland region to sub-regions. The RF models developed in WCA-3AN showed moderate transferability with an R² between 0.38 and 0.40, when the models were applied to WCA-2A. In contrast, the RF models demonstrated weak transferability from WCA-2A to WCA-3AN. Our results showed that model transferability and scaling varies depending on attribute domain space (environmental covariates) and geographic domain space (i.e., the extent used to develop models).