Systematic Review of Sentinel-5P Nitrogen Dioxide (NO₂) Validation and Surface Estimation Methods (2018–2025): Implications and Perspectives for Indonesia

Abstract

This review examines validation and conversion of Sentinel-5P TROPOMI column data to near-surface concentrations, focusing on NO₂ and contextualizing CO, SO₂, O₃, and HCHO from 2018–2025. Following PRISMA guidelines, a three-stage Scopus search identified 1,022 studies, narrowed to 150 validation-focused studies. Studies were included if they (i) converted TROPOMI vertical column density to surface concentration and (ii) reported at least one quantitative metric (R², RMSE, MAE, or correlation) against ground observations. After full-text assessment, 15 met both criteria. A single reviewer conducted screening and extraction, a limitation. Extracted fields included conversion method, algorithm, reference type, region, and performance metrics. Bibliometric mapping showed a shift toward machine learning in 2023–2024, alongside a Global North–Global South disparity and limited research in maritime continental regions. In this predominantly mid-latitude corpus, Random Forest, XGBoost, neural networks, and deep learning were most common and generally agreed more strongly with ground observations than direct column–surface correlations or chemical-transport-model scaling. However, no head-to-head comparisons on identical datasets existed, so this advantage is context-dependent. On held-out data, deep learning achieved validation R² of 0.914–0.919 across three annual UK test sets; Random Forest achieved R² = 0.875 (training R² = 0.949) in China’s Fenwei Plain; and XGBoost achieved pan-European test-set MAE of 7.77 µg/m³. Metrics represent validation/test performance unless noted otherwise. NO₂ dominates global validation, while evidence for CO, SO₂, and O₃ remains limited, particularly in tropical, cloudy regions with sparse monitoring. None of the 15 studies addressed the tropics or provides direct evidence for the Maritime Continent. In Indonesia, the critical bottleneck is the absence of Pandora or MAX-DOAS reference spectrometers. Installation in Jakarta and Surabaya should be prioritized, followed by staged NO₂, CO, and volcanic SO₂ estimation workflows adapted to tropical monsoon conditions.