Combing remote sensing information entropy and machine learning for ecological environment assessment of Hefei-Nanjing-Hangzhou region, China

Urban ecological environment is the basis of citizens’ survival and development. A rapid and objective urban ecological environment assessment (UEEA) plays an important role in the urban sustainable development and environment protection. This study established an improved urban ecological comfort index (UECIIMP), which is based on our previous UECI and fully composed of four remote sensing indicators: normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), land surface temperature (LST), and aerosol optical depth (AOD), representing the greenness, dryness, heat, and atmospheric turbidity, respectively. Combining the entropy method and random forest (RF) algorithm, the weights of four indicators were calculated. To improve the accuracy of UECIIMP, the gap-filled quarterly mean results of each indicator with 30m resolution were obtained using the harmonic analysis of time series (HANTS) method and spatial-temporal information fusion based on non-local means filter (STNLFFM). UECIIMP was applied to the Hefei-Nanjing-Hangzhou Region to explore its spatiotemporal changes and response characteristics. Results show that the weights of UECIIMP fluctuate slightly (within 10%) before and after sensitivity analysis, with good stability and reliability. UECIIMP in Hangzhou > Hefei ≈ Nanjing, spring ≈ autumn > summer ≫ winter. From 2009 to 2019, UECIIMP has improved in all 33 districts of the Hefei-Nanjing-Hangzhou Region. The significant improvement of UECIIMP in 2014–2019 is 4.3 times than that in 2009–2014. The correlation between UECIIMP and economic index indicates that economic development has a positive impact on the urban ecological environment. The significant degradation of UECIIMP in the urban expansion area demonstrates a negative impact on the local environment from urban expansion. •UECIIMP is an index based on UECI and completely derived from remote sensing images.•Combining entropy method and random forest method to calculate the weights of indicators of UECIIMP.•Time series reconstruction and spatiotemporal fusion algorithms were used to improve the data quality of UECIIMP.•UECIIMP was applied to Hefei-Nanjing-Hangzhou region from 2009 to 2019 for its ecological environment assessment.•The application of UECIIMP shows that it has the potential to be applied in different cities or regions.