Representation and evaluation of southern Africa's seasonal mean and extreme temperatures in the ERA5-based reanalysis products

Abstract

Over data-sparse regions such as southern Africa, reanalysis products represent valuable proxies for observed weather data. These products are, however, associated with a range of strengths and weaknesses which cannot be overlooked. Hence, for the period 1979–2021, we explored the performance of three ERA5-based reanalysis products (i.e. AgERA5, ERA5 and ERA5-Land) for their spatiotemporal representation of mean summer (November–March) and winter (May–September) temperatures, and coldwave and heatwave characteristics of the seasonal average number of events and days, and magnitude across southern Africa. Compared to the National Oceanic and Atmospheric Administration Climatology Prediction Centre (NOAA CPC) gridded observation-based temperature reference dataset, the reanalysis datasets adequately reproduced spatiotemporal characteristics of mean daily temperatures, with high area average correlations (r > 0.8). Summer (winter) temperatures were, however, typically under-estimated (over-estimated), with area average biases ranging from −0.22 °C (ERA5) to −0.11 °C (AgERA5) for summer and 0.31 °C (ERA5) to 0.51 °C (ERA5-Land) for winter. These biases were reflected in the general tendency for under-estimation (over-estimation) of the average coldwave (heatwave) magnitudes, with biases ranging from an average of 1.04 °C2 (AgERA5) to 1.76 °C2 (ERA5-Land) for coldwave magnitudes and −0.03 °C2 (AgERA5 and ERA5-Land) to −0.09 °C2 (ERA5) for heatwave magnitudes. Of the coldwave and heatwave indices investigated, the ERA5-based coldwave and heatwave seasonally averaged magnitudes were associated with the poorest performance, with the weakest correlations and most statistically significant biases, especially for regions outside South Africa which have substantially fewer observed temperature records. While the reanalysis products were spatiotemporally consistent in their performance for the seasonally averaged number of coldwave and heatwave days and events, they typically over-estimated the number of days, by almost two days for both indices, contributing to these events, while they generally under-estimated (over-estimated) the number of coldwave (heatwave) events by less than one event for both indices. On average, seasonally averaged heatwave characteristics were somewhat better represented than that for coldwaves. Despite this, the ERA5-based reanalysis products performed relatively well overall and are valuable to apply in further research considering coldwaves and heatwaves across southern Africa, provided their limitations are acknowledged. Impact-based studies exploring heatwave and coldwave influences on human health, crop yields, livestock thermal comfort, and water availability, for instance, will, however, likely require higher grid resolutions offered by the AgERA5 and ERA5-Land products.