Satellite-based precipitation estimates using a dense rain gauge network over the Southwestern Brazilian Amazon: Implication for identifying trends in dry season rainfall

Accurate long-term estimates of rainfall at fine spatial and temporal resolution are vital for hydrometeorology and climatology studies, but such data are often unavailable in remote regions. We assessed the accuracy of three satellite-based precipitation products that have data from 1981 to 2019 over the state of Rondo<SIC>nia in the Brazilian Amazon: (a) satellite-only, using the Climate Hazards Group Infrared Precipitation (CHIRP) product, (b) CHIRP with sparse gauge data (CHIRPS), and (c) CHIRPS calibrated with data from a dense rain gauge network (N = 73) (dnCHIRPS). We evaluated the rainfall products using additional validation gauges (N = 55) at the monthly and seasonal time scales and compared their drought events and temporal trends. Both CHIRP (10.0 mm/month mean error (ME), 23.6% percent bias (PB)) and CHIRPS (-0.08ME, 7.4% PB) underestimate high monthly rainfall in the wet season and overestimate low monthly rainfall during the dry season. dnCHIRPS had a lower error in monthly rainfall (-0.01ME, 1.1%PB) compared with CHIRP and CHIRPS, with the largest percentage difference between dnCHIRPS and the other two datasets in the dry season. dnCHIRPS captured decreasing trends in dry season rainfall over agricultural parts of the state, trends that were missed by the other two products. We conclude that a high density of rain gauges is essential for documenting the spatial pattern and trends in rainfall during the dry season and droughts in this important agricultural region of the Amazon basin.