Weather or not—Global climate databases: Reliable on tropical mountains?

Experimental design Precipitation data were collected with cumulative gauges and in parallel with automatic single tipping buckets connected to a data logger. In general, we used the data from the cumulative gauges, as this type has proven to be more reliable and has fewer failures than the automatic gauges. Although no objective quality correction could be applied, the use of accumulated rainfall should generally minimize errors and data inhomogeneity [13]. In case of missing data, the gaps were filled by data from the automatic gauges. Details of these stations (serial number of each station (if available), name of the station, start and end dates of the precipita- tion record) and elevation are given in S1 Table. On Kilimanjaro we have selected only stations with at least 10 years of data (without counting gaps in records), with the exception of 15 of our own stations in the montane and alpine zones with at least 5 years, where records are sparse. On the other mountains, where we started our measurements after 2010, most own sta- tions have at least 5 years of data. The period of records at the stations varied. In the lower areas records started mainly in the middle of the last century, our own records (mainly in the upper areas) started at the end of the last century (S1 Table). WC provides data from 1950– 2000, and C from 1979–2013 [10]. We have accepted these discrepancies in the recording peri- ods in order to have a sufficiently large data set for such a comparison, and because precipita- tion changes have been marginal at Kilimanjaro in recent decades [18].

For precipitation measurements, we used cumulative gauges consisting of a funnel, a pipe and a canister, sealed with silicon, that are read manually about every three months. In addi- tion, we measured precipitation with Pronamic single spoon tipping buckets at 0.2mm resolu- tion connected to a Driesen & Kern DK311 ruggedPlus MultiLog Datalogger. On plots outside the forest belt, these gauges are installed at least 50 cm above the canopy to prevent back- splash. Inside the montane forest zone of Kilimanjaro, we placed the gauges in clearings (see [13]). Statistical analysis Following the procedure of Ref. [2] we calculated the percent bias (pbias) that reflects the aver- age tendency of the modelled precipitation values psim to be larger or smaller than their observed values Pobs at the stations. The optimal value of pbias is 0, with low values indicating accurate model simulation. Since we are only interested in overall bias disregarding over- or underestimation, we use the absolute numbers. Pbias is defined as follows: pbias ¼ 100∗ Pn i¼0 jpsimi pobsi j Pn i¼0 pobsi ! Additionally, we also report the mean absolute error (mae) which is defined as: mae ¼ 1 n ðX n i¼0 jpsimi pobsi jÞ Validation between observed and modelled rainfall data was performed with Pearson’s cor- relation coefficient. Extraction of WC and C data was done with ArcMap 10.8.2 To validate the performance of WC and C, we compared it to our own data using the high- est resolution of the WC and C data (30 arc seconds), which has a better performance than the

courser resolution of 0.25˚ gridded data [2]. For this, we calculated the mean absolute error (mae) and the percentage bias (pbias). We performed this calculation separately for the dry north slope and the wet south slope, which have different precipitation regimes. This separa- tion of the ecoclimatic regions shown in Fig 3 was in accordance with [14]. PLOS ONE Global climate datasets on tropical mountains PLOS ONE | https://doi.org/10.1371/journal.pone.0299363 March 13, 2024 6 / 12