Accurate tropical cyclone (TC) monitoring is limited by sparse direct wind observations. Geostationary infrared imagery provides frequent cloud top observations, but a single image does not fully describe recent TC motion or the link between TC center, intensity, and inner core size. This study presents a physics constrained TC monitoring model, named PCTCM, for global estimation of TC center, intensity, and radius of maximum wind (RMW). The model uses GridSat-B1 infrared images and historical TC information. It takes the current non centered infrared image and the historical TC information from 6 and 12 hours earlier as inputs. A point branch estimates center, intensity, and RMW. A map branch learns a spatial structure heatmap. To give this heatmap physical meaning, we build wind field targets from Holland radial profiles. The map output is also used to infer center, intensity, and RMW, so that the point output and map output are trained to be consistent. Tests on global TC cases from 2023 to 2024 show that PCTCM reaches a center location error of 30.19 km, an intensity RMSE of 2.35 m/s, and an RMW MAE of 8.72 km. Historical TC information provides the largest gain. Physics constrained wind field heatmaps further improve intensity and RMW estimation by linking cloud structure with radial wind structure. The results show that temporal TC history and physically meaningful heatmap supervision can improve satellite-based TC monitoring across global basins.

暂无