<p>A new 10-type urban Local Climate Zone (LCZ) classification with 100-m resolution was developed, following the guidelines of the World Urban Database and Access Portal Tools (WUDAPT) over the Greater Bay Area (GBA). This LCZ dataset was incorporated into the Building Environment Parameterization (BEP)-Building Energy Model (BEM) multi-layer urban canopy scheme used by the Weather Research and Forecasting (WRF) model, with key parameters (such as fraction of impervious surface, building height/width, road width, air conditioning usage) determined from local building morphology and energy consumption patterns. The impacts of using such detailed 10-type LCZ, as compared to using remapped 3-type LCZ and using default WRF 1-type urban land cover were assessed, based on parallel integrations of the WRF system at 1-km resolution for a historical hot-and-polluted event over the GBA. It was found that the model surface temperature, air temperature, humidity and wind speed in the 10-type LCZ run were in closer agreement with in-situ observations, demonstrating the value of detailed urban LCZ data in improving the model performance. Smaller diurnal temperature range and higher nighttime temperature were found in the 10-type LCZ run compared to the 3-type LCZ and 1-type runs. Increased building height in the 10-type LCZ setting also reduces positive bias of wind speed in the lower planetary boundary layer at urban locations. The cold and dry biases over the non-urban areas in the 10-type LCZ run could be further reduced through considering updated land cover, soil type, soil hydraulic/thermal parameters, soil moisture/temperature. Owing to the improvement in capturing the urban meteorology, incorporating more detailed LCZ classification might also improve air-quality simulations. These findings should be relevant to the development of comprehensive, high-resolution earth system models, which are an indispensable tool for mitigation of and adaption to regional environmental and climate changes.</p>
Feature Importance Analysis for Local Climate Zone Classification Using a Residual Convolutional Neural Network with Multi-Source Datasets