Land suitability assessment for sisal production: A machine learning and Analytical Hierarchy Process integrated approach

Land suitability evaluation is critical in achieving efficient and sustainable use of land resources. This study applied multicriteria decision analysis by intergrading soil, climate, and topography parameters with the Analytical Hierarchy Process (AHP) and Random Forest (RF) model to analyze the land suitability for sisal production in Kilosa District, Tanzania. Mass preserving spline function was applied to harmonize topsoil (0–30 cm) properties from field survey and existing soil profile data. Classes of key soil properties for sisal suitability were predicted using the RF model with covariates derived from Sentinel 2 A and Digital Elevation Model. AHP based on 10 experts' opinion was used to derive the relative priorities of criteria in land suitability evaluation for sisal. The sisal suitability map was generated using the weighted overlay approach. K-fold cross-validation of the RF model revealed the accuracies in the prediction soil properties suitability class ranging from 0.45 to 0.85, with a Kappa index between 0.17 and 0.64. The AHP ranking determined soil as the most important criterion (0.67), followed by climate (0.22) and topography (0.11). The overall suitability analysis indicated that 20 % of the study area is highly suitable, 59 % moderately suitable, 8 % marginally suitable, and 13 % not suitable for sisal production. Soil drainage, slope, and nutrients were identified as the major constraints for sisal production. To improve suitability, surface water flow management and drainage, conservation agricultural practices in sloping land, and integrated soil fertility management are recommended.