SEAS5 skilfully predicts late wet‐season precipitation in Central American Dry Corridor excelling in Costa Rica and Nicaragua

Abstract

Better drought preparedness is critically needed in the Central American Dry Corridor (CADC). Seasonal forecasts can be used to build this preparedness but need localized evaluations to ensure they are relevant and useful. This study provides a CADC-focused assessment of the SEAS5 seasonal forecasting system produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). We evaluate SEAS5 predictions of the mean, variability, and extremes of precipitation across the CADC at 1-7-month lead times. We assess differences in regional forecast quality across seasons and lead times by evaluating spatial and temporal associations with El Niño-Southern Oscillation (ENSO) phase, topography, and continentality. Results show that SEAS5 precipitation forecasts often have better skill primarily during the mid to late wet season (July-October). In these months, low/normal precipitation forecasts outperform the climatological mean up to 5-or 6-month lead times in some subregions. Forecast skill is often worse, however, for predicting precipitation during the early wet season, primarily in June. Forecast skill varies spatially across the region, with higher skill concentrated in the southeast (Costa Rica and Nicaragua). Forecast skill is significantly related to continentality and topography, and together these factors account for at least a quarter of the spatial variance in annual skill at all lead times. Forecast accuracy varies depending on ENSO phase: predictions are often worse in El Niño (warm ENSO) periods during the early wet season when ENSO also has a weaker association with cumulative precipitation relative to the later wet season.SEAS5 could be a particularly useful tool during the second half of the wet season in the southeast CADC, skilfully alerting of upcoming precipitation variability with over 3-month lead times.

A comparison of seasonal rainfall forecasts over Central America using dynamic and hybrid approaches from Copernicus Climate Change Service seasonal forecasting system and the North American Multimodel Ensemble

“…We generate indirect and statistical forecasts using two known teleconnection regions—Niño3.4 (170°–120°W, 5°N–5°S) for the ENSO teleconnection (Trenberth, 1997; Trenberth and Stepaniak, 2001) and Tropical North Atlantic (TNA; 55°–15°W, 5°–25°N) for the TNA teleconnection (Enfield et al ., 1999). TNA is selected partly because of its relatively strong association with Central American rainfall in the early wet season (Spence et al ., 2004; Alfaro, 2007; Alfaro et al ., 2016a; Maldonado et al ., 2017), a time of year when some GCMs have had relatively lower skill (e.g., Kowal et al ., 2021) to see how this SST zone could potentially enhance their skill. The indirect forecasts are constructed in pyCPT like the direct rainfall forecasts, but SST is selected as the predictor for either the Niño3.4 or the TNA region and then the same CCA method is used to transform the raw 1° GCM forecasts to 0.25° and statistically relate the SST predictors from a selected zone (e.g., Niño3.4) in a given period (e.g., MJJ) to rainfall over Central America in that period (e.g., MJJ).…”

“…Evaluation of seasonal rainfall predictions over Central America is needed because GCMs that perform well globally may not necessarily forecast rainfall well over a given region (Hagedorn et al ., 2005; Amador and Alfaro, 2009; Hidalgo and Alfaro, 2012; 2015; Doblas‐Reyes et al ., 2013; Kharin et al ., 2017; Almazroui et al ., 2021). Some evaluations of GCMs have showcased their potential over Central America and in nearby regions (e.g., Kirtman et al ., 2014; Weisheimer and Palmer, 2014; Carrão et al ., 2018; Khajehei et al ., 2018; Slater et al ., 2019; Becker et al ., 2020; Gubler et al ., 2020; Kowal et al ., 2021). Very few of these studies target Central America specifically and they typically compare one model, ensemble, or forecasting method, using different forecast verification metrics and time‐periods of evaluation.…”

“…Although ENSO is a well‐known indicator of rainfall over Central America, and its phases are publicized by meteorological bulletins to inform stakeholders (e.g., IMN, 2019), this phenomenon is unlikely to provide the full rainfall picture. El Niño phases of ENSO only explain some of the drought events in the CADC, for instance, and drought events differ in intensity between similar El Niño phases (Muñoz‐Jiménez et al ., 2019; Kowal et al ., 2021). The isthmus is affected by a complex interaction of weather patterns that arise from processes in both the Pacific and Atlantic oceans (Durán‐Quesada et al ., 2020), including seasonal migration of the inter‐tropical convergence zone (ITCZ), tropical cyclones, and movement of the Atlantic warm pool (Enfield and Alfaro, 1999; Poveda and Mesa, 1999; Wang and Enfield, 2001; Amador et al ., 2006; Wang, 2007; Amador, 2008; Hidalgo et al ., 2015; Sori et al ., 2015; Durán‐Quesada et al ., 2017; 2020).…”

Process‐Informed Subsampling Improves Subseasonal Rainfall Forecasts in Central America

“…We test our process‐informed subsampling method in Central America, a region in need of further MME optimization where AOGCM skill varies across locations, times of year, and lead times (e.g., Carrão et al., 2018 ; Hidalgo & Alfaro, 2012 , 2015 ; Kowal et al., 2021 , 2023 ; Maldonado, Alfaro, Amador, & Rutgersson, 2018 ; Maldonado, Alfaro, & Hidalgo, 2018 ). Central America presents prediction challenges in part due to the complex interaction of weather patterns originating from both the Pacific and Atlantic oceans and marked topography that moderates moisture transport over the region (Durán‐Quesada et al., 2017 , 2020 ).…”