Explainable, uncertainty-aware machine learning for maize land suitability and climate resilience in Lorestan Province, western Iran
Description
This dataset provides the 520 georeferenced reference points used to calibrate and validate an explainable, uncertainty-aware machine-learning model of maize land suitability for Lorestan Province, western Iran. Each point (fields Point_ID, Latitude, Longitude) is characterised by 37 environmental predictors spanning four domains: topography (Elevation, Slope, Aspect, TWI, TRI, TPI), soil properties (pH, Clay, Sand, Silt, SOC, Bulk_Density, CEC, Coarse_Fragments, Nitrogen, Soil_Depth, Available_Water_Capacity), climate (Annual and Growing-Season Precipitation, Precipitation Seasonality, Mean/Max/Min Temperature, Temperature Seasonality, PET, AET, Water_Deficit, Soil_Moisture, Aridity_Index), and remote sensing/accessibility (NDVI_Mean/Max/Min, EVI_Mean/Max, Land_Cover, Distance_River, Distance_Road). Each point also carries a four-class Suitability_Class label (S1 = highly suitable, S2 = moderately suitable, S3 = marginally suitable, N = not suitable), assigned using the FAO maximum-limitation land-evaluation framework. All 520 points fall within cropland, spanning approximately 33.20–34.00°N and 48.00–48.90°E. The file is provided as a single-sheet Excel workbook (41 columns × 520 rows) so that other researchers can reproduce the suitability classification, retrain or benchmark alternative machine-learning classifiers, or reuse the predictor stack for related land-evaluation studies in similar semi-arid agricultural settings. Variable names, units and derivation are documented in the accompanying manuscript's Methods and Table 1.