Predicting high resolution total phosphorus concentrations for soils of the Upper Mississippi River Basin using machine learning

The spatial distribution of soil phosphorus (P) is important to both biogeochemical processes and the management of agricultural landscapes, where it is critical for both crop production and conservation planning. Recent advances in the availability of large environmental datasets together with big data analytical tools like machine learning have created opportunities for evaluating and predicting spatial patterns in complex environmental variables like soil P. Here, we apply a random forest machine learning model to publicly available soil P datasets together with nearly 300 geospatial attributes summarizing aspects of soil type, land cover, land use, topography, nutrient inputs, and climate to predict total soil P at a 100 m grid scale for the Upper Mississippi River Basin (UMRB), USA. The UMRB is one of the most intensively farmed regions in the world and is characterized by widespread water quality degradation arising from P-associated eutrophication. Although potentially complex interacting drivers determine total soil P, the predictive accuracy of our random forest model was relatively high (R2 = 0.58 and RMSE = 129.3 for an independent validation dataset). At the regional scale represented by our model, the variables with the greatest comparative importance for predicting soil P included a combination of soil sample depth, land use/land cover, underlying soil physical and geochemical properties, landscape features (such as slope, elevation and proximity to the stream network), nutrient inputs, and climate-related factors. An important product of this research is a fine-scale (100 m) raster data layer of predicted total soil P values for the UMRB for public use. This dataset can be used to improve conservation planning and modeling efforts to improve water quality in the region.