Land Surface Temperature (LST) is a key parameter in climate and environmental studies, influencing ecosystem processes and biological dynamics at various scales. While meteorological stations provide high-temporal-resolution LST measurements, their spatial coverage is limited due to the point-based nature of the observations and the high costs and maintenance requirements. In contrast, remote sensing data enables continuous and extensive monitoring of LST. However, a major challenge in thermal remote sensing is the sensitivity to cloud cover, which results in data gaps. This study introduces a novel hybrid approach to reconstruct LST in cloudy regions using MODIS satellite data and ground-based meteorological observations. The methodology consists of a Random Forest(RF) regression model trained on MODIS brightness temperature bands (31 and 32) and ground station data. For cloud-covered regions, two reconstruction scenarios were implemented: (1) interpolation-based estimation using Inverse Distance Weighting (IDW) refined by RF, and (2) index-based estimation using MODIS-derived indices such as NDVI, NDBI, and NDMI. A genetic algorithm (GA) algorithm was employed to combine the outputs of both scenarios by optimizing their weights to minimize the estimation error. The hybrid approach resulted in an RMSE of 0.78°C, demonstrating an improvement of approximately 0.8°C over individual scenarios. This method enables the generation of continuous and accurate LST maps under cloudy conditions and holds promise for enhancing thermal remote sensing applications in challenging environments.