Abstract ID: 218
On the next generation (NextGen) seasonal precipitation forecast in Chile
Lead Author: Diego A. Campos Díaz
Direccion Meteorologica de Chile, Chile
Keywords: nextgen seasonal forecast, precipitation, calibrated multi-model ensemble, climate services
Abstract: Since 2019, the Chilean Meteorological Directorate (DMC) began a process of updating its seasonal precipitation forecast, going from a forecast based exclusively on sea surface temperature in the El Niño 3.4 region to a multi-model assembled forecast. using the NextGen methodology designed by the International Research Institute for Climate and Society (Columbia University) and promoted by the World Meteorological Organization.
The Chilean NextGen is based on a calibrated multi-model ensemble (CMME) approach that uses, on the one hand, state-of-the-art general circulation models (GCM) from the North American Multi-Model Ensemble Project (CCSM4, CFSv2, FLOR-A06, and CanCM3). A canonical correlation analysis-based regression (CCA) is used to calibrate the predictions from the GCMs against observations and bring the forecasts from grid resolution to ground station resolution. On the other hand, a set of statistical models are built using CCA based on the historical relationship between sea surface temperature in the Niño 3.4 (ENSO model) region and the South-Western Pacific (SWP model) and the monthly rainfall in Chile.
The CMME is created using an equal-weight average of the calibrated GCMs and statistical models that present sufficient historical ability (using the 2AFC > 40% as threshold) for each ground station after a cross-validation process. The CMME shows the probabilistic tercile forecast and is disseminated to the public in a monthly presentation and bulletin.
Multiple verification metrics were used to evaluate these probabilistic seasonal forecasts between 2019 and 2021. The results show that, among GCMs, CFSv2 performs better in terms of resolution (based on the Heide Skill Score), discrimination (using the Critical Success Index), and reliability (according to Brier Score), especially during the winter. The best-evaluated statistical model was the SWP model.
The CMME forecast exhibits better performance than individual models and an experimental forecast based on the simple average of all models. A notable advantage of CMME is the consistency in performance throughout the year; the individual models present a significant seasonal variability, and CMME is less variable, adding confidence to the forecast.
Finally, the latitudinal extension of Chile adds difficulty to the realization of seasonal forecasts based on spatial patterns (CCA). The results show that the central zone concentrates the best results, while a significant decline in verification metrics is observed towards the extreme south and north of the country.