High resolution 4D soil organic carbon stock mapping at farm scale

Soil organic carbon (SOC) is intrinsically linked to global carbon balance, climate change mitigation, soil health, and agricultural productivity. Therefore, obtaining accurate information on the spatial-temporal of the soil organic carbon stock (SOCS) is essential. We proposed a four-dimensional (4D) SOCS mapping approach, encompassing space (two dimensions), depth and time. A 100 × 100 m grid sampling was conducted in 1997 and 2022 at two depths (0–20 cm and 80–100 cm) in a sugarcane farm. Dynamic and static covariates representing the soil formation processes were used to fit three Cubist models. We tested three strategies for SOCS spatial-temporal mapping: 1) a model for each year, 2) a model fitted using only the data of the last year (2022) and 3) a multitemporal model using data of both periods. Based on external validation, strategy 1 and 3 produced more accurate and less biased maps, with coefficient of determination (R 2 ) of 0.74, root mean square error (RMSE) of 7.69 ton ha −1 and bias of 3.43 ton ha −1 and R 2 of 0.76, RMSE of 7.19 ton ha −1 and Bias of 3.25 ton ha −1 for strategy 3, respectively. Strategy 2 was less efficient (R 2 = 0.71; RMSE = 11.06 ton ha −1 ; Bias = 7.93 ton ha −1 ). Strategy 3 is particularly useful for SOCS mapping when only limited temporal observations are available. Soil attributes (static) were most important covariates for modeling, followed by a bare soil image (dynamic) and vegetation information (dynamic). An increase in SOCS was observed in most sampling sites and predicted maps. The SOCS dynamic was related to soil type, geology and showed an inverse relationship with bare soil frequency. Finally, the SOCS saturation deficit was assessed by spatio-temporal mapping.