<title>Abstract</title>Predicting rainfall-induced landslides hinges on the quality of the rainfall product. Satellite rainfall estimates or rainfall reanalyses aid in studying landslide occurrences especially in ungauged areas, or in the absence of ground-based rainfall radars. Quality of these rainfall estimates is critical; hence, they are commonly crosschecked with their ground-based counterparts. Beyond their temporal precision compared to ground-based observations, we investigate whether these rainfall estimates are adequate for hindcasting landslides, which particularly requires accurate representation of spatial variability of rainfall. We developed a logistic regression model to hindcast rainfall-induced landslides in two sites in Japan. The model contains only a few topographic and geologic predictors to leave room for different rainfall products to improve the model as additional predictors. By changing the input rainfall product, we compared GPM IMERG and ERA5 rainfall estimates with ground radar–based rainfall data. Our findings emphasize that there is a lot of room for improvement of spatiotemporal prediction of landslides, as shown by a strong performance increase of the models with the benchmark radar data attaining 95% diagnostic performance accuracy. Yet, this improvement is not met by global rainfall products which still face challenges in reliably capturing spatiotemporal patterns of precipitation events.