Abstract
Skillful prediction of the Mid‐Summer Drought (MSD) is important for various socioeconomic sectors in southern Mexico, Central America and the Caribbean. However, operational forecasting errors of the MSD have rarely been evaluated systematically. In this study, we address this research gap by examining operational forecasts of the MSD derived from the North American Multimodel Ensemble (NMME; 1991–2020). We assess these forecasts before and after applying a Model Output Statistics (MOS) scheme based on Canonical Correlation Analysis (CCA). Before applying MOS, only a couple of forecasts exhibited a bimodal signal between July and September, but none of them reproduced the particular signals of the MSD; they generally suffered from two main errors: excessively weak precipitation during the June–September season (particularly in June) and a weak peak in July instead of a relative minimum. While the root cause of these errors can be associated with warm sea surface temperature (SST) bias in the tropical eastern Pacific and cold SST bias in the Gulf of Mexico, Caribbean Sea and northwestern tropical Atlantic, their immediate cause is an erroneous evolution of the eastern Pacific Intertropical Convergence Zone (ITCZ). During June and September, the eastern Pacific ITCZ remains too far south near the equator, while in July and August it expands and intensifies but stays too close to the coasts of southern Mexico and Central America, failing to migrate westward. After applying MOS, the forecasts showed high skill scores during the onset of the MSD (July and August) but not in the months before and after (June and September). As revealed by the CCA analysis, this improved skill is due to the MOS‐corrected forecasts' improved representation of a key relationship: drier MSD episodes are associated with a stronger westward SST gradient between the eastern tropical Pacific Ocean and the Caribbean Sea. Models that adequately capture this relationship exhibit reduced uncertainty in forecasting MSD intensity. This finding can provide a valuable pathway to mitigate uncertainties in projecting future changes in MSD intensity under different climate change scenarios.