Predictive modelling benchmark of nitrate Vulnerable Zones at a regional scale based on Machine learning and remote sensing

Nitrate leaching losses from arable lands into groundwater were a main driver in designating Vulnerable Zones (NVZs) according to the Nitrates Directive, with view enhancing their water quality. Despite this, developing common strategies for effective quality control these areas remains challenge European Union. This paper evaluates performance of Random Forest (RF) machine learning algorithm combined Feature Selection (FS) techniques predicting nitrate pollution NVZs bodies different periods and using updated environmental features Andalusia, Spain. A set forty-four extrinsic used as predictors, an aim make this methodology exportable other regions. Phenological obtained through remote-sensing included measure dynamics agricultural activity. In addition, dynamic derived weather livestock effluents analyse seasonal interannual changes pollution. Three feature stacks two databases predictive modelling: Period 1 (2009), 321 samples training; 2 (2010), 282 validation initial spatial prediction; 3 (2017), assess probability content exceeding 50 mg/L. wrapper four sequential search methods was considered: backward selection (SBS), forward (SFS), floating (SFFS) (SBFS). From among all applied, SFS had best (overall accuracy = 0.891 six predictor features) linked highest three features: Normalized Difference Vegetation Index (NDVI) base level, NDVI value end growing season accumulated manure production farms; static slope, sediment depositional valley depth.