Skill of Global Raw and Postprocessed Ensemble Predictions of Rainfall in the Tropics


Precipitation forecasts in the tropics potentially have a high economic value, but the quality of current numerical weather prediction (NWP) ensemble forecasts has not been evaluated in detail. In this study, we investigate the skill of two ensemble prediction systems (EPSs), namely those of the European Centre for Medium-Range Weather Forecasts (ECMWF) and of the Meteorological Service of Canada (MSC), to issue predictions for the amount and occurrence of precipitation as well as for extreme precipitation events. We consider accumulation periods of 1–5 days and focus on the broad tropical belt between latitudes 30°S and 30°N. EPS data for the years 2009–2017 are taken from the TIGGE archive and are evaluated with Tropical Rainfall Measuring Mission (TRMM) satellite rainfall estimates. In order to evaluate the full potential of ensemble forecasts, we apply the state-of-the-art statistical postprocessing technique Ensemble Model Output Statistics. Both raw and postprocessed forecasts are then compared to a static observation-based forecast, the so-called extended probabilistic climatology (EPC). The results are aggregated for different Köppen-Geiger climatic regions in the tropics. Raw ensemble forecasts from both models are uncalibrated and unreliable, with MSC forecasts often being slightly better calibrated and more reliable than their ECMWF counterparts. Raw ensemble forecasts for the amount and occurrence of precipitation as well as extreme precipitation of both models are slightly skillful for most climatic regions within the tropics, but not for northern arid and tropical Africa and regions with complex terrain. After postprocessing, forecasts are calibrated and skill is increased almost everywhere, with ECMWF forecasts being more skillful than MSC forecasts. Still, amount and occurrence of precipitation and extreme precipitation remain unpredictable for northern arid and tropical Africa and in complex terrain, while clear skill can be observed for most other regions. Over the investigation period 2009–2017, the ECMWF raw ensemble forecasts show a period of clear improvement between 2009 to 2011 due to better calibration and some smaller improvements afterwards. The increase in horizontal resolution in 2010 probably contributed to this improvement. Postprocessed forecasts show little change in skill, indicating that the earlier forecasts benefitted more from the statistical corrections. There are, however, some regional differences. It is concerning that for several tropical regions, particularly the drier climates, no improvement in forecast skill, neither for raw nor postprocessed forecasts, can be observed for a period of almost one decade. In particular, as for many atmospheric variables clear improvements can be observed for the same period, this standstill in forecast improvement for precipitation forecasts needs further attention and investigation.

100th American Meteorological Society Annual Meeting