Mapping phosphorus sorption and availability in California vineyard soils using an ensemble of machine learning models

Authors: WILSON, STEWART - California Polytechnic State University; Steenwerth, Kerri; O'GEEN, ANTHONY - University Of California, Davis

Submitted to: Soil Science Society of America Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/30/2022
Publication Date: 12/22/2022
Citation: Wilson, S., Steenwerth, K.L., O'Geen, A.T. 2022. Mapping phosphorus sorption and availability in California vineyard soils using an ensemble of machine learning models. Soil Science Society of America Journal. 87(1):119-139. https://doi.org/10.1002/saj2.20487.
DOI: https://doi.org/10.1002/saj2.20487

Interpretive Summary:

Technical Abstract: Spatial variability of soil phosphorus (P) is tied to pedogenic state factors, as well as to management practices in cultivated soils. The distribution of P availability and sorption was modeled in the Napa and Lodi American Viticulture Areas (AVA) in California, utilizing a predictive soil mapping framework. We tested three machine learning algorithms (MLA), Random Forest (RF), Extreme Gradient Boosting (XGB) and Cubist, as well as two super learner ensembles of base models, model stacking and model averaging. 141 pedons were analyzed for Olsen P and P sorption index (PSI) and aggregated by depth weighted average (0-30 cm and 30-100 cm). Point data was combined with rasters of environmental predictors, to model Olsen P and PSI. Base models (RF, XGB and Cubist) performed well for PSI prediction (R2=0.68-0.73), but less well for Olsen P (R2=0.46-0.56). For ensembles, model averaging was selected for PSI at 0-30 cm (R2=0.77) and model stacking was selected for PSI at 30-100 cm (R2=0.74). For Olsen P, model averaging was selected for 0-30 cm (R2=0.42), and model stacking for 30-100 cm (R2=0.52). Predictions highlighted regional differences in pedogenesis and P dynamics suggesting that vineyards may benefit from place-based P management. Predictions were strong for PSI, and less robust for Olsen P. Fe/Al-(hydr)oxides control P sorption in weathered soils, whereas management influences Olsen P. The spatial variability of Fe/Al-(hydr)oxides is tied to pedogenesis. Thus, P sorption can be readily predicted with environmental predictor variables and ensemble machine learning, owing to the pedological underpinnings of predictive soil mapping.