Mapping soil organic carbon and clay using remote sensing to predict soil workability for enhanced climate change adaptation

Abstract

Climate change is presenting sizeable challenges for agricultural production around the world. In some regions, shifting precipitation patterns in the spring and fall are negatively impacting farm operation by reducing the number of “workable days” or the days fields can be worked with heavy equipment without damaging soil structure. This can be particularly problematic for farms on clay soils and/or poor drainage. Approximating a water content threshold at which a soil is not workable due to soil structure destruction can be helpful for planning effective farm operations. In this study, we applied advanced remote sensing and machine learning tools to produce digital maps of soil organic carbon (SOC) and clay (CL) content and used them in existing pedotransfer functions (PTFs) to predict a workability threshold (WT) across a study area in Delta, British Columbia, Canada. We combined field data, soil and vegetation indices derived from multiple Landsat satellite images, topographic indices, and soil survey information to digitally map SOC and CL of the agricultural lands in Delta using random forest (RF) and generalized boosted regression model (GBM). When validated against an independent field dataset, the RF model outperformed GBM for all accuracy measures (coefficient of determination – R2, concordance correlation coefficient – CCC, and normalized root mean square error – nRMSE). We then spatially applied several PTFs using our digital maps to estimate the plasticity limits of the soil and produce WT map. The WT map was then tested against independent field samples of the soil water content at −10 kPa and we achieved R2 of 0.59, CCC of 0.70, and nRMSE of 0.15. Our analysis showed that 40% of the fields in the study area had WT < 30%, a threshold that is already being impacted by reduced workable days. This WT map could be used to improve spatial prioritizations of investmentsfor climatechangeadaptation at farm to regional scales.