Remote Sensing and Ensemble Machine Learning for Illegal Dumpsite Susceptibility Mapping: A Review
Purpose: This review asks whether, and under what conditions, combining satellite remote sensing with machine learning, and ensemble learning in particular, improves the mapping of illegal-dumpsite susceptibility, and where the evidence for that claim is still missing, with emphasis on sub-Saharan Africa and Nigeria. Methodology: A structured narrative review with a non-systematic, purposive selection of studies was carried out. Candidate studies published from 2000 to 2026 were identified through targeted keyword searches of scholarly databases and search engines (Google Scholar, Scopus-indexed journals and African Journals Online), combining dumping terms (illegal dump, illegal landfill, dumpsite) with remote-sensing, susceptibility and machine-learning terms, and through citation chasing from recent surveys. Empirical studies that detected dumpsites or mapped dumping susceptibility with remote-sensing or GIS predictors and reported a validation outcome were included, and each was appraised for source type and validation design. Twenty-one empirical studies (16 international detection and susceptibility studies and 5 Nigerian studies) were synthesised, supported by methodological literature. Findings: Susceptibility studies outside Africa report strong discrimination with comparatively simple models (logistic regression, overall accuracy 92–94%; discriminant analysis, AUC 0.97–0.98), and the largest head-to-head comparison (seven classifiers on the same data) found AUC values within a narrow range (0.86–0.89). The two African susceptibility studies (Dakar, Blantyre) used expert-weighted multicriteria analysis rather than machine learning. Only one study used spatial cross-validation, none used a stacking ensemble, and no Nigerian study has modelled susceptibility predictively. Unique Contribution to Theory, Policy and Practice: Unlike recent surveys that focus on detecting waste in imagery, this review compares susceptibility-mapping methods on their reported performance and validation practice, maps the regional evidence gap for Nigeria, and sets out minimum standards (simple-model baselines, spatial validation and treatment of pseudo-absences) that future ensemble studies must meet before ensembles can be recommended to waste-management authorities.
综述指出非洲仅达喀尔、布兰太尔两项易感性研究采用专家加权多准则分析而非机器学习，尼日利亚尚无预测性建模研究，并给出空间交叉验证等最低方法标准。