An Ensemble Machine Learning Framework for Predicting Environmental Quality using Multisource Remote Sensing Data

Abstract

Accurate, spatially continuous assessment of environmental quality is a prerequisite for effective environmental governance, pollution monitoring, and sustainable urban planning. Conventional ground-based monitoring networks provide precise point measurements but suffer from sparse spatial coverage, high operational costs, and inconsistent temporal density. Satellite remote sensing offers a transformative complementary modality providing synoptic, multi-temporal, and globally consistent environmental observations yet existing single-source machine learning prediction approaches fail to exploit the rich complementarity across multiple satellite platforms and spectral domains. This paper proposes EnsembleEQ, a novel ensemble machine learning framework for predicting a composite Environmental Quality Index (EQI) integrating air quality, water quality, vegetation health, and urban thermal stress dimensions from multisource remote sensing data. EnsembleEQ fuses radiometrically calibrated imagery from four satellite sources Sentinel-2 MSI, Landsat-8 OLI/TIRS, MODIS MOD11A1/MOD13A1, and Sentinel-5P TROPOMI to derive a comprehensive 47-feature representation per spatial grid cell, including spectral indices, thermal bands, atmospheric composition products, and spatial texture descriptors. three-tier ensemble architecture trains base learners Random Forest, XGBoost, LightGBM, and Support Vector Regression independently on this multisource feature set, with a Ridge Regression meta-learner combining their predictions through cross-validated stacking. EnsembleEQ is evaluated on a multi-site dataset spanning 18 environmentally diverse study sites across South Asia, including industrial zones, agricultural belts, coastal areas, and urban cores, validated against 2,847 ground-truth measurements from national monitoring networks.