Abstract
Global, spatially interpolated climate datasets such as WorldClim and CHELSA, widely used in research, are based on station data, which are rare in tropical mountains. However, such biodiversity hotspots are of high ecological interest and require accurate data. Therefore, the quality of such gridded datasets needs to be assessed. This poses a kind of dilemma, as proving the reliability of these potentially weakly modelled data is usually not possible due to the lack of stations. Using a unique climate dataset with 170 stations, mainly from the montane and alpine zones of sixteen mountains in Tanzania including Kilimanjaro, we show that the accuracy of such datasets is very poor. Not only is the maximum amount of mean annual precipitation drastically underestimated (partly more than 50%), but also the elevation of the precipitation maximum deviates up to 850m. Our results show that, at least in tropical regions, they should be used with greater caution than before.