Diverse skill of seasonal dynamical models in forecasting South Asian monsoon precipitation and the influence of ENSO and IOD

Abstract

The South Asia Seasonal Climate Outlook Forum (SASCOF) issues seasonal tercile precipitation forecasts to provide advance warning of anomalously dry or wet monsoon seasons in South Asia. To increase objectivity of the SASCOF seasonal outlook, the World Meteorological Organisation recommends using a multi-model ensemble combining the most skilful dynamical seasonal models for the region. We assess the skill of 12 dynamical models at forecasting seasonal precipitation totals for 1993–2016 for the southwest (June–July–August–September) and northeast (October–November–December) monsoon seasons at regional and national levels for Afghanistan, Bangladesh, Nepal, and Pakistan, using identical forecast periods, hindcast initialisation months and domain used at the SASCOF. All models demonstrate positive skill when regionally-averaged, especially for the southwest monsoon season, noting considerable spatial differences. Models exhibit highest skill where correlation between observed precipitation and El Niño Southern Oscillation (ENSO) is highest, e.g., central/north India and Nepal during the southwest monsoon, and Afghanistan and north Pakistan during the northeast monsoon. Model skill is especially low in northwest India and northeast of South Asia during the southwest monsoon, e.g., Bangladesh (despite high precipitation totals) coinciding with a weak ENSO teleconnection. The Indian Ocean Dipole teleconnection is less pronounced in the southwest monsoon season, whereas the spatial pattern for the northeast monsoon closely resembles that of ENSO. Due to high variability in model skill, we recommend basing the SASCOF forecast on a multi-model ensemble of all models but discounting poorly performing models at the national level.

Performance‐based evaluation of NMME and C3S models in forecasting the June–August Central African rainfall under the influence of the South Atlantic Ocean Dipole

“…In our findings, regions, where models demonstrate higher skill, are generally those with less complex terrain, such as the relatively homogeneous Gabon and Chad regions. Generally, models exhibit poorer skill in areas characterized by a wide range in elevation (as noted in studies by Muthoni et al (2023) and Stacey et al (2023)). To enhance model skill, it is essential to ensure that the high mountains and extensive coastlines of CA are wellrepresented in the models to capture the intricate precipitation effects.…”

“…To identify the extent of the variability in relation to the total (mean) precipitation (SST), the coefficient of variance (CV) for year-to-year rainfall/SST is useful (Stacey et al, 2023;Zhang & Han, 2017). The spatiotemporal variabilities of the reference and the model data were studied using the empirical orthogonal function (EOF) analysis (Wallace & Dickinson, 1972;Wilks, 2019), which yields the dominant modes of spatial SSTAs variabilities along with their temporal evolution patterns, that is, the principal component (PC) time series.…”