Accounting for the spatial variation of phosphorus available explained by environmental covariates

• Published in: Geoderma Regional Vol. 32; p. e00594
• Main Authors: Nalin, Renan Storno, Dalmolin, Ricardo Simão Diniz, de Araújo Pedron, Fabrício, Moura-Bueno, Jean Michel, Horst, Taciara Zborowski, Schenato, Ricardo Bergamo, Soligo, Matheus Flesch
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01-03-2023
• Subjects: Agriculture precision; Digital soil mapping; Magnetic susceptibility; Multiple soil classes; Random forest; Random forest; Multiple soil classes; Digital soil mapping; Agriculture precision; Magnetic susceptibility
• ISSN: 2352-0094; 2352-0094
• DOI: 10.1016/j.geodrs.2022.e00594

The demand for maps of soil attributes in agriculture has increased during the past years, especially aiming the rational use of phosphorus fertilizer which can lead to serious environmental damage, like eutrophication of water bodies. The development of strategies and protocols for the accurate spatial modeling of P is necessary to a sustainable soil fertility management and food production. The objective of this study was to account for the spatial variation of available phosphorus explained by environmental covariates in southern Brazilian farm. The study was conducted in an agricultural area of 162 ha, located in the county of Tupanciretã, Rio Grande do Sul State. A set of 162 soil samples was collected in the 0–10 cm layer. Based on magnetic susceptibility and chemical and physical analysis, 9 soil maps were produced by Ordinary Kriging to be used as covariates. A Digital Elevation Model with 12 m of resolution was used to derive 13 topographic covariates. To predict the spatial distribution of the available soil phosphorus content, a set of six models was formulated based on different combination of the covariates. The models were fitted using a Random forest algorithm. Independent probability samples (n = 50) were used to evaluate the maps. Available phosphorus content from samples varied from 4.79 to 220.45 mg dm−3, with an average of 48.80 mg dm−3. Among predictive models, the one fitted using only topographic covariates (model 1) presented the highest predictive error (RMSE = 30.65 mg dm−3). When all available covariates were included in the model formulation, the predictive error decreased (RMSE = 28.05 mg dm−3). In general, including soil covariates result in better prediction than using only topographic covariates. The lack of soil covariates related to clay and iron fractions could be replaced for magnetic susceptibility data.
•Soil covariates produce more accurate available P maps compared to topographical.•The use of topographic and soil covariates improved the accuracy of available P maps.•Soil Fe has the greatest contribution to explaining the variability of P.•Magnetic susceptibility has the potential to explain the variability of available P.•Magnetic susceptibility could replace the clay and iron data as soil covariates.