Spatio-temporal modelling of soil organic carbon stock change in relation to land use changes in selected sub-catchments of Lesotho

Résumé fourni par la source

Soil organic carbon (SOC) stock is the most pivotal indicator of soil fertility and monitoring its space-time changes is a prerequisite to establish strategies to reduce soil loss and preserve its quality. In this study, the topsoil (0–20 cm) SOC stock changes due to land use change in Maletsunyane and Makhalaneng sub-catchments between 2017 and 2022 were modelled using the quantile Regression forest (QRF) model. The QRF model was used to map the distribution of SOC, bulk density and coarse fragments which together with soil depth, helped determine SOC stock and estimate SOC stock changes from 2017 to 2022 in Makhalaneng and Maletsunyane sub-catchments. A total of 16 covariates for climate (mean annual temperature, mean annual rainfall), 34 covariates for organisms, 13 for terrain and 5 land use data were used. The performance and uncertainty of the QRF model were tested using cross-validation. The prediction R 2 of SOC in Makhalaneng was 0.668 in 2017 and 0.585 in 2022, and in Maletsunyane, it was 0.633 in 2017 and 0.570, respectively. The mean SOC stock decreased from 80.56 Mg C ha −1 to 75.30 Mg C ha −1 in Makhalaneng and from 109.27 Mg C ha −1 to 94.21 Mg C ha −1 in Maletsunyane, respectively, over a 5-year period. The QRF model showed a decreasing SOC Stock trend in Makhalaneng due to grassland conversion to cropland with a change of − 3.36 ± 4.96 Mg C ha −1 , and in Maletsunyane, the loss of natural grasses resulted in a change of − 15.24 ± 16.26 Mg C ha −1 . • We tested the use of machine learning for space–time mapping of soil organic carbon (SOC) stock. • Highest SOC stocks were found in grasslands. • There was a significant decrease in mean SOC stock over the 5-year period in both sub-catchments. • Land use changes, particularly the conversion of grasslands to croplands, were identified as a key factor affecting SOC stock changes. • Accurate machine learning SOC stock prediction requires dense soil sampling in space and time.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.