Uncovering smallholder patterns in agricultural-water-soil characteristics: an unsupervised machine learning method

Farmers, especially smallholders, are pivotal for global agricultural sustainability. Understanding their agricultural behavior can inform sustainable policy strategies. This study employs unsupervised machine learning (K-means) to unveil smallholder patterns, focusing on the interplay of agricultural, water, and soil resources. Three dominant factors driving pattern creation are identified. These factors encompass agricultural data (farming class, crop type, farming system), water data (water scarcity, ground water level, evapotranspiration), and soil data (nutrient availability, terrain slope, global soil organic carbon), along with climate data (temperature, precipitation). Transforming and scaling the data yield 30 features for comparability. K-means identifies five clusters: Europe and the east USA, central Asia, South America, western USA and Mexico, and India with central Africa. Dominant factors influencing pattern creation are water scarcity, ground water level, and Global Soil Organic Carbon (GSOC). These factors elucidate cluster differences. The discussion evaluates the algorithm's robustness regarding the number of clusters and random state. Consideration is given to alternative data for a more detailed pattern description. The study also scrutinizes the smallholder definition's suitability.