Hybrid Machine Learning for Integrating Pedological Knowledge into Digital Soil Mapping to Advance Next-Generation Earth System Models

Abstract

Environmental models often require soil maps to represent the spatial variability of soil properties. However, mapping soils using conventional in situ survey protocols is time-consuming and costly. As an alternative, Digital Soil Mapping (DSM) offers a fast-mapping approach that has the potential to estimate soil properties and their interrelationships over large areas. In this study, we address the currently outdated spatial information on soil properties across a tropical region (approx. 98,000 km2) with a ~700-km longitudinal gradient of contrasting topography, climate, and vegetation in Brazil by developing and applying statistical soil models for this region using a novel hybrid machine learning (HML) framework. This framework reduces prediction redundancies due to high multicollinearity by implementing a recursive feature selector algorithm for input selection. The hybrid framework’s core is composed of the Soil-Landscape Estimation and Evaluation Program (SLEEP) and a calibrated Gradient Boosting Model (GBM) capable of modeling the spatial distribution of soil properties at multiple soil depths. The use of SLEEP and GBM allowed us to explain the spatial distribution of various basic physical and chemical soil properties and their environmental modulators. The model training and testing approach used six topographical, ten meteorological and two vegetation properties, and data from 223 soil profiles across the study area. Our models demonstrated a consistent performance with spatial extrapolations exhibiting r2 values ranging from 0.79 to 0.98, and percent bias (PBIAS) from -1.39 to 1.14%. The properties related to topographic and climatic conditions were dominating when estimating the number of soil layers, percentage of silt and the sum of bases. Our framework features high flexibility, while reducing capital investments and increasing accuracy when compared to traditional mapping protocols that require extensive surveys.