Data-driven Livestock Manure Allocation: Integrating Soil Analysis and Nutrient Requirements with Machine Learning

Efficient livestock manure management has become increasingly critical as farm intensification generates large nutrient surpluses that may threaten soil and water quality if not properly allocated. Addressing and implementing existing animal waste through appropriate planning is one of the major environmental problems in our country. In this study, a hybrid model integrating machine learning algorithms (MLA) and the multi-criteria decision-making (MCDM) approach was developed. The study was conducted across 18 fields at Aşkın Dairy Farm, located in Dümrek Village, central Çanakkale province. Gradient Boosting (GB), Random Forest (RF), Support Vector Regression (SVR), and K-Nearest Neighbors (KNN) algorithms were used in the study. Furthermore, the MCDM model enabled prioritization of fields by considering criteria such as distance from barns to fields, plant nitrogen requirements, and analysis results. The results revealed that the GB algorithm was the most successful, with the lowest Root Mean Square Error (RMSE= 0.0057) and the highest R² (0.99). The findings demonstrate that tree-based methods, in particular, are more effective for manure distribution. Furthermore, it was determined that the 1,454 tons of annual manure produced on the farm meets the needs of six fields, while the remaining fields require chemical fertilizer support.