Abstract. Satellite rainfall products are considered important options for acquiring rainfall estimates in the absence of ground measurements. However, estimates from these products need to be validated as their accuracy can be affected by geographical position, topography, and climate, as well as by the algorithms used to derive rainfall from satellite measurements. Eight satellite-based rainfall products (TRMM, CHIRPS, RFEv2, ARC2, PERSIANN, GPCP, CMAP and CMORPH) were evaluated against ground observations over the complex topography of the upper Tekeze-Atbara basin in Ethiopia. The performance was evaluated at various temporal (daily, monthly, seasonal) and spatial (point, sub-basin, basin) scales over the period 2002–2015.

Results show that CHIRPS, TRMM, and RFEv2 performed well and were able to capture the rainfall measured by rain gauges. The BIAS and correlation of these products were within ±25 % and > 0.5 over different time steps. The remaining products poorly performed at daily time step with higher BIAS (up to ±200 %) and lower correlation (< 0.5). CMORPH, PERSIANN, and ARCv2 were relatively better while CMAP and GPCP performed poorly (r < 0.4) in all conditions. The overall performance of all products was lower in the mountainous areas of the basin with station elevation > 2500 m a.s.l. Compared to the lowlands, the BIAS at highlands increased by 35 % whilst the correlation dropped by 28 %.

Underestimation and overestimation of rainfall dominated in the mountainous and lowland areas, respectively. CMORPH and TRMM overestimated while the remaining products underestimated the rainfall at all spatiotemporal scales. CMAP, ARC2, and GPCP estimates were the most affected by large underestimation. Unlike in temporal scale, the performance of the products did not show a uniform pattern with respect to spatial scale. Their performance improved from point to aerial comparisons in the lowlands whereas it slightly reduced at highland areas. Poor performance in the highlands contributed to a slightly lower performance at basin scale compared to the pixel-to-point comparison.

Our results show that rainfall estimates from CHIRPS and TRMM have a consistently good agreement with ground rainfall at different spatiotemporal scales in the upper Tekeze-Atbara basin. Interpolating the sparse and unevenly distributed rain gauges over the complex terrains however introduces unknown uncertainties with respect to the actual rainfall.