MODIS satellite temperature data lack full validation across China. Using 18 years of balloon data from 74 stations, we evaluated its accuracy, lower-atmosphere stability, and inversion detection. Nighttime accuracy exceeds daytime; upper layers outperform lower layers . A bias exists at 700 hPa. Stability retrieval is more reliable by day. Inversion detection rates are low: 6.6 % day, 13.1 % night. This guides appropriate use of MOD07_L2 in atmospheric science, hydrology, and ecology.

1. Developed a Bi-clustered Recursive Bayesian Forest model, improving wind speed forecast accuracy by 48.2 % and reducing error metrics by more than 60 %. 2. Incorporated Sea-land breeze, weather stability, and atmospheric circulation index as features, using bi-clustering modal classification to mitigate wind speed magnitude interactions. 3. Proposed a machine learning-based correction technique that outperforms traditional numerical models for more reliable wind speed forecasting.

Reliable rainfall forecasts several weeks ahead are essential for managing weather risks in the Yangtze River Delta, where floods and winter icing cause major impacts. This study evaluates whether reducing systematic forecast biases can improve rainfall predictions. An adaptive correction applied to existing forecast models leads to higher accuracy, better spatial agreement, and improved representation of extreme events, supporting more effective early warning and risk management.