Urban Dark Sky Environmental Evolution Identification and Associated Explanatory Factors Analysis for Dark Sky Park Development

The expansion of artificial nighttime lighting has increased the risk of urban dark sky environmental degradation. Identifying long-term evolution patterns and their associated socioeconomic and environmental factors is essential for conserving dark sky resources and planning dark sky parks. Taking Wuhan City as the study area, the NPP-VIIRS annual nighttime light data from 2013–2023 were used. The Theil–Sen slope was used to quantify long-term nighttime light changes and was denoted as the dark sky change trend (DSCT), while the Mann–Kendall test was applied to evaluate trend significance. These results were further integrated to identify dark sky environmental evolution (DSEE) types. An interpretable machine learning framework combining random forest (RF), partial dependence plots (PDPs), accumulated local effects (ALEs), one-vs.-rest (OVR), and SHAP methods was then used to identify core explanatory variables, nonlinear response patterns, class-specific contributions, and local explanatory patterns. A spatial diagnostic zoning framework was further developed for dark sky park planning. The results revealed that (1) Wuhan exhibited a clear center–periphery differentiation pattern from 2013–2023. Significant degradation zones were concentrated in the central urban areas and expansion zones, while high-quality stable zones were distributed mainly in the peripheral areas, accounting for 40.43% and 45.78% of the study area, respectively. Improvement zones accounted for only 1.65%, indicating limited systematic recovery. (2) PM2.5 concentration (PM), population intensity (PI), normalized difference vegetation index (NDVI), cloudless days (CLD), and digital elevation model (DEM) were identified as the core explanatory variables, with PM contributing the most, followed by NDVI and PI, while CLD and DEM mainly provided supplementary explanatory information. (3) Higher PM and PI were associated with higher predicted probabilities of dark sky degradation, whereas higher NDVI was associated with stable conditions and lower predicted degradation probabilities. (4) PM made the largest predictive contribution to the degradation and stable zones, whereas improvement zones exhibited more dispersed explanatory contributions among multiple variables. (5) Seven spatial diagnostic zones were identified by integrating various evolution types and local explanatory patterns. Potential development zones and development restriction zones accounted for 43.28% and 41.65%, respectively, forming the dominant spatial structure of the diagnostic zoning. This study extends urban dark sky assessment from single period brightness characterization toward long-term evolution identification and explanation-informed spatial diagnosis, thereby providing a transferable framework for dark sky resource management and preliminary planning assessment.