Multi-source precipitation products represented by satellite remote sensing is a necessary supplement to the measured data of meteorological stations, they play an important role in the accurate identification of the precipitation distribution in data-deficient areas. In this research, precipitation products of CFSv2, ERA5 and TPP (constructed by an improved climatic teleconnection model based on model-X knockoffs and random forest) were compared with the measured precipitation of meteorological stations from the spatial and temporal dimensions to evaluate their applicability in the upper Paran River Basin, Brazil. The determination coefficient (R2), Nash-Sutcliffe efficiency coefficient(NSE) and relative error (RE) were adopted as evaluation indices. The results showed that the fitting accuracy between the predicted value and measured value of the TPP and ERA5 products are higher than that of CFSv2. In the cross periods, the fitted R2 and NSE of the area rainfall expressed as ERA5>TPP>CFSv2, among which, the area rainfall calculated by CFSv2 was larger than that of the measured one, and the fitted RE reached 28.2%; on the contrary, the calculated value of ERA5 was smaller than the measured one, with a fitted RE of -10.3%. The RE between the calculated area rainfall of TPP products and the measured value was the smallest, which was 0.33%. Spatially, the fitted R2 and NSE of the three products with the measured precipitation at the selected typical meteorological stations were ERA5>TPP>CFSv2, while the |RE| expressed as TPP