Smart City Construction and Urban Ecological Resilience Improvement — Empirical Evidence Based on China’s NDVI Data

Abstract

This study provides targeted verification for the ecological dimension of artificial‑intelligence‑enabled urban sustainable development. This paper evaluates the ecological externalities of China’s national smart city pilot program. Using a difference-in-differences (DID) design on a balanced panel of 287 prefecture-level cities from 2000 to 2024, we estimate the impact of smart city initiatives on urban vegetation density, measured by the satellite-derived Normalized Difference Vegetation Index (NDVI). To mitigate the influence of outliers, all continuous variables are winsorized at the 1st and 99th percentiles. Our baseline estimates indicate that smart city designation leads to a statistically significant and economically meaningful 6.33% increase in urban vegetation density (e^0.0614 - 1  6.33%). An event study analysis tracking up to ten pre-treatment periods supports the parallel trends assumption—confirmed by a joint Wald test of the pre-treatment coefficients (χ²(9) = 15.5292, p = 0.0774)—and shows that the ecological benefits grow over time.

To address self-selection and policy endogeneity, we employ propensity score matching (PSM-DID) to match treated and control cities on 2016 baseline characteristics, confirming a robust policy coefficient of 0.0634***. Furthermore, controlling for concurrent urban-level environmental pilots (specifically, the Low-Carbon City Pilot and the Sponge City Pilot) does not confound our results. Mechanism analyses explore two potential pathways: a direct green space planning channel and an indirect environmental monitoring channel (proxied by log-transformed industrial wastewater discharge). While the green space planning channel is robustly verified, the wastewater pathway is statistically unverified under the traditional Baron-Kenny stepwise framework due to an insignificant first-step regression (p = 0.2551), suggesting only a potential directional trend. Nonetheless, direct Bootstrap testing indicates a weak indirect mediation effect of 0.0135 (95% CI: [0.0011, 0.0265]) for the wastewater path, accounting for 21.92% of the total effect, whereas green space planning accounts for 14.60% (95% CI: [0.0045, 0.0142]). Our study offers crucial policy insights for integrating digital urban governance with ecological civilization goals. The research conclusions provide micro‑level empirical evidence that artificial‑intelligence pilot zones enhance urban ecological resilience and advance sustainable urban development.