Urbanization diminishes net ecosystem productivity by changing the landscape pattern

Urban net ecosystem productivity (NEP_u) plays a pivotal role in enhancing urban ecological conditions and human comfort in living environments. However, the response mechanism of NEP_u to urbanization remains unclear due to the challenge of accurately estimating NEP_u. This study proposes a hybrid model architecture that combines Residual Networks and Carnegie–Ames–Stanford approach (CASA) model (ResNets-CASA) to estimate NEP_u. The ResNets-CASA model is trained and verified at four urban eddy covariance (EC) sites in Tianjin and Shenzhen, China. The model shows an average root-mean-square-error (RMSE) of 1.09 gC/m²/day and a coefficient of determination (R²) of 0.83 for daily NEP_u simulation across the four EC sites. The trained ResNets-CASA model at the site scale is further employed for regional-scale NEP_u mapping and the generation of long-term historical NEP_u datasets (1986–2022) for the two cities. The long-term trend analysis indicates that the NEP_u of the two cities shows a significant downward trend over the past 37 years, with an average decline rate of 2.1 gC/m²/year across the two cities. The main cause for the decline trend of NEP_u is the changed urban landscape pattern, including: 1) a decrease in urban vegetation coverage resulting from urbanization; 2) shifts in the composition of urban vegetation species due to the substitution of natural woodland and cultivated land with urban landscape grassland. These results emphasize the dominant role of changes in urban landscape patterns, rather than urban microclimate, in the long-term decline of NEP_u trends.