Subseasonal Forecasts over the Caribbean, Central America, and Mexico and Verifications

Over the past several years, there has been a growing demand for operational forecasts on the sub-seasonal time scale from week 1 to week 3-4 with lead time ranging from one day to two weeks. This is particular due to the importance of this time scale in socio-economic sectors such as agriculture, water resource management, health, energy and transportation. In response to these escalating needs and requirements, NOAA’s Climate Prediction Center within the framework for the WMO Regional Climate Centers (RCCs) is developing a set of forecasting tools to address the gap in sub-seasonal forecasting. This paper focuses on the southern part of the WMO Regional Association (RA) IV, encompassing the Caribbean, Central America, and Mexico, hereafter referred to as S-RAIV. The NCEP Global Ensemble Forecast System (GEFS) and the Climate Forecast System version 2 (CFSv2) are used in the prediction of precipitation (P) and two meter air temperature (2mT) at week 1, week 2, and week 3-4. Both calibrated and non-calibrated forecasts are presented. The CPC Unified Precipitation data and ensemble regression are used to calibrate the forecasts. Forecasts are presented in two categorical probabilities of above and below average P and 2mT. We further use standard verification metrics including Root Mean Square Error (RMSE), Brier Score (BS), the Heidke Skill Score (HSS) and the Receiver Operating Characteristic (ROC), to provide an objective evaluation of the forecast quality. Both verifications through the hindcast period of 1999-2016 and for forecasts generated in 2017 and 2018 are presented. Results are quite consistent with findings in the literature and reveal skillful forecasts at lead times of week 1 and week 2. It is also noted that calibration helps improve the week 3-4 forecast skills. Finally tailored forecasts to heat waves in the S-RAIV and their verifications are presented. Heat waves are defined in two ways: (1) three consecutive days in a week for the minimum T or the maximum T to exceed the 90^th percentile in the 30-year climatological record from 1981 to 2010; or (2) the heat index, combination of 2mT and relative humidity to exceed a certain threshold, 38⁰C, for example, for three consecutive days . Results suggest that the NCEP models perform reasonably well in depicting heat wave events in the S-RAIV. We show that this information when made available in real time can help mitigate the impact of heat on human health in vulnerable populations.