Özet:
Soil fertility evaluation is crucial for long-term agricultural management and land-use planning, particularly in dry and semi-arid countries. The fertility condition of agricultural soils in Bardarash and Semel districts, Duhok Province, northern Iraq, is assessed using an integrated method that includes field sampling, laboratory analysis, geostatistics, and machine learning. A total of 105 composite soil samples (52 from Bardarash and 53 from Semel) were analyzed for 19 key physicochemical attributes, including soil texture, bulk density, pH, electrical conductivity, organic matter (OM), calcium carbonate (CaCO₃), cation exchange capacity (CEC), and available macro and micronutrients. To create the Soil Fertility Index (SFI), two different weighting schemes were used: Analytical Hierarchy Process (AHP) and Principal Component Analysis (PCA) for Bardarash and integrated with fuzzy logic for Semel. The normalized soil characteristics were weighted and aggregated to provide SFI values, which were then modeled using several prediction algorithms. SFI prediction in Bardarash was achieved using regression-based machine learning models with field spectroradiometer data as input factors. Semel, on the other hand, used machine learning techniques using remote sensing-derived indicators like as NDVI and soil indices as predictors to perform SFI classification. Gradient boosting regression (GBR) had the strongest predictive reliability in Bardarash, whereas classification-based techniques produced robust findings in Semel. Model performance was tested using R², MAE, MSE, and RMSE metrics. Soil fertility is mostly limited by OM depletion, poor phosphate availability, excessive CaCO₃ concentration, and insufficient CEC, according to the investigation. Additionally, spatial mapping revealed that approximately 58% of the soils in Semel and about 70% in Bardarash were classified as having low to very low fertility levels. In order to promote precision agriculture and sustainable land management practices in northern Iraq, these findings emphasize the significance of combining multivariate weighting approaches with geospatial machine learning and remote sensing to create precise, site-specific soil fertility maps.
