Assessing land suitability for leguminous crops in the okavango river basin: A multicriteria and machine learning approach

• Multi-criteria decision-making and data-driven multivariate models were employed to investigate suitability of agricultural land to grow legume crops, primarily relying on remotely sensed data. • Machine learning models require well-balanced datasets and may exhibit a higher degree of uncertainty in predicting suitability in areas where crops are not typically grown or underrepresented in the data. • Over 70% of the suitable class coverage is observed within a 2 km radius from the river or its tributaries due to higher soil moisture and potentially higher nutrient levels found in soil types near rivers. This study aimed to create a model to identify land suitable for growing sunn hemp ( Crotalaria juncea ) and pigeon pea ( Cajanus cajan ) in the Okavango River basin of the Kavango East region of Namibia. Advanced tree-based ensemble learning models, including Random Forest, Extra Trees, Gradient Boosting, XGBoost and multivariate regression analysis , were employed to enhance analytical accuracy. The Random Forest and XGboost models exhibited outstanding performance, as evidenced by their respective accuracy values of 0.97 and 0.96. In addition, this study proposed an innovative approach through the integration of subjective and objective analytical methods, which are independent of one another. The subjective component of the analysis employed a Multi-Criteria Decision Making-Analytic Hierarchy Process (MCDM-AHP). On the other hand, the objective component used a data-driven multivariate approach supported by tree-based learning algorithms. Twenty-two variables were considered, encompassing climatic conditions, hydro-geomorphologic features, soil characteristics, vegetation patterns, and socio-economic factors. These variables played a crucial role to identify the most suitable areas for growing the selected leguminous crops. The MCDM-AHP method utilised expert evaluations to rank the importance of variables, identifying water sources, slope, and soil properties as key factors. A suitability mapping analysis revealed that 17.63% of the area was highly suitable and 62.77% moderately suitable, while 10% was less suitable and 9.59% unsuitable for growing these two legumes. According to the data driven methodology, soil fertility and nitrogen content emerged as key determinants for land suitability. This is particularly relevant for nitrogen-fixing leguminous crops such as sunn hemp and pigeon pea, which play a central role in improving soil quality and ensuring food security.