Suitability Assessment of Remotely Sensed Urban Air Quality Data

The application of remotely sensed PM2.5 concentration datasets has become increasingly widespread, but the spatial precision verification at local scales is lacking. This study aims to investigate the consistency of PM2.5 concentration between remotely sensed data and ground-based data and optimize the accuracy of remotely sensed PM2.5 concentration data at the urban scale. Specifically, taking Changsha city as a case, four evaluation indices—R2, RMSE, uncertainty, and high deviation rate (HDR)—were employed to evaluate the credibility of remotely sensed data at national and dense ground-based stations, then analyze spatial variations of credibility and develop a Recursive Feature Elimination–Cross-Validation Random Forest (RFECV-RF) model to improve local fitting accuracy. Results show that remotely sensed data exhibit high credibility at national stations, while credibility at dense stations varies spatially and tends to decline with increasing distance from national stations. After optimizing by the RFECV-RF model, the credibility of remotely sensed data can be significantly improved, with R2 increasing from 0.87 to 0.98, RMSE decreasing from 8.59 µg/m3 to 3.08 µg/m3, HDR reducing from 2.01% to 0.04%, and uncertainty declining from 18.93% to 8.27%. Nevertheless, certain regions still require additional monitoring to further expand the credible spatial extent. These findings provide valuable insights for improving PM2.5 concentration remote sensing monitoring methods and designing the integrated “air–space–ground” observational network scheme.