Reconstructing continental‐scale variation in soil δ15N: a machine learning approach in South America

• DOI: 10.1002/ecs2.3223
• Publication Date: 2020-08-31T14:47:11Z
• Authors: João Paulo Sena‐S...; Benjamin Z. Houlton; Luiz Antônio Mart...; Gabriela Bielefel...
• Keywords: Biome; Soil carbon

Abstract Soil nitrogen isotope composition (δ 15 N) is an essential tool for investigating ecosystem balances, plant–microbe interactions, ecological niches, animal migration, food origins, and forensics. The advancement of these applications limited by a lack robust geospatial models that are capable capturing variation in soil δ N (i.e., isotopic landscapes or isoscapes). Due to the complexity cycle general scarcity information, previous approaches have reconstructed regional global patterns via highly uncertain linear regression models. Here, we develop new machine learning approach ascertain finer‐scale understanding geographic differences N, using South American continent as test case. We use training set spanning 278 locations across continent, all major biomes. tested three different methods: cubist, random forest (RF), stochastic gradient boosting (GBM). 10‐fold cross‐validation revealed RF method outperformed both cubist GBM approaches. Variable importance analysis framework pointed biome type most crucial auxiliary variable, followed organic carbon content, determining model performance. thereby created biogeographic boundary map, which predicted expected multiscale spatial pattern with high degree confidence, performing substantially better than America. Therefore, showed be great opportunity explore broad array ecological, biogeochemical, forensic issues through lens N.