Validation of satellite- and gauge-based gridded rainfall products over Ghana (West Africa)

Africa is known for its notoriously poor coverage of available surface meteorological observations. This pertains also to rainfall, one of the most important meteorological parameter on a continent, on which many farmers rely on rain-fed agriculture and which is known to experience large interannual to decadal rainfall fluctuations. This is also true for the country of Ghana, our study region, situated at the Guinea Coast and affected by the seasonal cycle of the West African monsoon. Due to the paucity of surface observations, many agricultural, hydrological and climate change studies and applications use satellite derived rainfall estimations. These rely on the use of satellite radiances at different wavelength, are therefore indirect measurements and are known show improvements if calibrated by surface gauge observations. However, in the case of Ghana, the availability of observations from the 22 synoptic stations on the WMO Global Telecommunication System has been quite poor in recent decades, thus limiting the possibility of ground calibration used in several satellite products. Obviously, the paucity of surface observations may also impact on widely used gauge-only products.

In the present study, we make use of a recently created, 0.25°x0.25° latitude-longitude gauge-only product that is based on a dense network of 113 stations over Ghana for the 15-year period 1998-2012 (Aryee et al. 2017, Internat. Journal of Climatology). Monthly rainfall totals from this product are compared to six satellite rainfall estimates (CMAP, GPCP, ARC2, TAMSAT V3, TRMM 3b43 and CHIRPS V2) and two gauge-only products (GPCC V5 and CRU). Comparisons are made for annual totals and the seasonal cycle for the entire territory of Ghana as well as for four ecological zones in Ghana, viz. the coastal, forest, transition, and Savannah zones. Linear Pearson correlations between monthly anomalies of the Ghana gauge and the eight tested products range between 0.52 for TAMSAT to 0.77 for GPCC at the coast. CHIRPS and TAMSAT show lower correlation at the coast with somewhat higher values inland. Correlations rise to about 0.9 for all products if the seasonal cycle is not removed. ARC2 and CMAP substantially underestimate the standard deviations, with TAMSAT and CRU being very close to observations. Regarding the unsystematic bias, CMAP, ARC2 and CHIRPS perform poorest for Ghana and the former two have the largest (negative) bias. If split up into the four ecological zones, the tested products perform poorest for the forest and coastal zones where rainfall is known to be related to warmer cloud tops and land-sea breeze convection. In these most densely populated regions of Ghana, the standard deviations are also underestimated, the biases are mostly negative and the centered root mean square errors are the largest. Overall, the ARC2, CMAP and GPCP products show the poorest performance with CHIRPS showing mixed results with a poorer performance near the coast. TAMSAT and TRMM show best performance with the gauge-only CRU product being comparable to the latter two. Our results show deficiencies of all products in the wet coastal zone and corroborate the need to include more surface rain gauge information into the product in zones with more copious rainfall from warmer cloud tops.