Climate comfort is a key indicator for measuring urban livability and the tourism experience. This study focuses on Chengdu as the research area and integrates multi-source data—including ground-based meteorological observations

satellite remote sensing data

and radar data—to construct a comprehensive climate comfort assessment model based on the Temperature-Humidity Index (THI)

Wind Effect Index (WEI)

and Clothing Index (ICL)

enabling detailed evaluations on a monthly and seasonal scale. For daily mean temperature

relative humidity

and daily mean wind speed

a combined prediction model (LSTM-FNN) was proposed

consisting of a four-layer Long Short-Term Memory (LSTM) network and a single-hidden-layer feedforward neural network (FNN). The model was trained and tested using daily meteorological data from 2015 to 2023

with Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) as evaluation metrics

and compared with LSTM

BiLSTM

and BRNN models. The results indicate that in Chengdu

spring and autumn are "comfortable" or "fairly comfortable

" summer is "somewhat uncomfortable

" and winter is "uncomfortable"; the RMSE values of the LSTM-FNN for temperature

humidity

and wind speed forecasts were 1.82°C

5.37%

and 0.41 m/s

respectively

significantly outperforming the comparison models

demonstrating a distinct advantage in multivariate nonlinear fitting. This study provides a quantifiable technical solution for urban climate assessment and forecasting.