In the recent years, the FengYun-3B passive microwave observations (FY3B) have been used to study global and regional soil moisture estimates, and land surface temperature (LST) has been an important input in the retrieval model. Previous studies proposed a linear model to derive LST from Ka-band brightness temperatures (H09 approach) which were routinely masked based on open water data. This study aimed to further understand microwave emission under such challenging open water conditions and developed the first data-driven approach to correct for such open water conditions. The H09 was first applied to the FY3B observations, and then inter-compared against MODIS (Moderate-resolution Imaging Spectro-radiometer), MERRA (Modern Era Retrospective analysis for Research and Applications) and ground observations over the Jiangsu province in the People’s Republic of China. Several freshwater bodies are located within this region, and the Pacific Ocean bounds at the east, this makes it an excellent region for the study of open water impact on the passive microwave observations. As in previous studies, the quality of the FY3B LST based on the H09 approach dropped over the regions with high open water. While the MERRA and MODIS LST products maintained an average correlation of about 0.95, the FY3B LST product had an average correlation of 0.85 with drops to about 0.65 over the open water regions. Additional comparisons also revealed biases over the open water regions within the passive microwave product. An open water metric, which is based on the percentage of a pixel and its surrounding pixels covered by water was developed to quantify the impact of open water. Based on a pixel-wise linear relationship between the MERRA LST and FY3B Ka-band brightness temperature, spatially variable slopes and intercepts were obtained as a function of open water to produce a correction for open water in the H09 which leads to the HG18 approach. Validations with both the MODIS LST product and the ground observations showed that there was a significant reduction in both the absolute bias (MODIS: 2 K, in situ: 10 K) and the standard errors (MODIS: 0.8 K, in situ: 1 K) using the HG18 approach. This opens up a new window towards further improving and understanding LST retrievals from passive microwave products.