Site adaptation of CAMS without a local station in West Africa: transfer cost and bias structure

Ground measurements of solar irradiance are scarce in West Africa, and most sites have no station with which to adapt satellite estimates. A CAMS site adaptation learned at existing stations can instead be transferred to such a site. The cost of this transfer is evaluated at 31 West African stations, with corrections learned on one year of data and tested on the following year. For global horizontal irradiance (GHI), having no local measurement adds about one percentage point to the mean absolute error of the annual total. The north–south trend in the bias transfers, but each station retains an offset of its own. Part of the observed excess error comes from the sensors. With independent calibration errors of 0.95 %, the excess of the trend corrections is no longer detectable with this network. Assuming a centred error, their one-sided 90 % bound on the annual error is 2.6–2.7 %, compared with 2.2 % for one year on site. Measured against the true irradiance, only the two transfers based on the nearest stations remain worse than on-site adaptation across the whole network (permutation test with Holm correction). On the Tier 1 stations alone, no transfer is worse. A bias–latitude line with prediction intervals is provided for sites within the study domain. For direct normal irradiance (DNI), no transfer cost is established after correction for multiple comparisons. One year of measurement is not enough to pin down its bias.