South America has long been underrepresented in global numerical weather prediction verification, leaving a gap in understanding forecast performance over the continent.
To address this, a pilot project was launched in early 2024 by the Working Group on Numerical Experimentation (WGNE) and the Joint Working Group on Forecast Verification Research (JWGFVR), in response to the UN Early Warnings for All initiative. Led by Argentina's National Meteorological Service (SMN) and ECMWF, the pilot evaluates global and regional forecast performance over South America. This article summarises the main results, including statistical verification of ECMWF's Integrated Forecasting System (IFS), comparisons with other global and limited-area models, and case studies of three extreme weather events in Argentina: a destructive convective storm, strong Zonda winds and a prolonged heat wave.
General verification results
Over South America, statistical verification showed positive IFS performance. Upper-air results demonstrated skill comparable to Europe, an unexpected outcome given the Southern Hemisphere's unfavourable land-ocean distribution and sparse upper-air observations.
Surface verification indicated generally low bias across most of Argentina. The model performed particularly well in the Pampas and along the Atlantic coast. More challenging conditions were found over the Amazon, where complex land-atmosphere interactions and dense vegetation make modelling difficult, and in the Andes, where steep topography is difficult to represent. These features are illustrated, for example, in the 2-metre dew point temperature station verification maps.
Inter-model comparisons showed the IFS performing competitively with the GFS, ICON, CMC and UKMO models, and achieving skill comparable to Argentina's operational WRF system, which is GFS-coupled, for most meteorological variables.
High-impact weather case studies
Three contrasting extreme weather cases illustrate IFS performance across different physical mechanisms and predictability challenges.
On 21 July 2023, a Zonda wind event tested the IFS against one of Argentina's most challenging topographic phenomena. The model reproduced downslope winds with high skill, matching or exceeding the WRF. Enhanced IFS experiments with finer resolution and an improved convection scheme showed little benefit over the ENS control, except for some improvement in 10-metre wind gusts. All models, however, struggled to capture the extremely low surface humidity, with 2-metre dew points reaching –40 °C.
From 16 to 17 December 2023 in northeastern Argentina, a convective windstorm highlighted the dual challenge of predicting synoptic evolution and extreme surface winds. CAPE-shear Extreme Forecast Index (EFI) values signalled severe potential up to six days ahead, but the IFS was unable to resolve damaging gusts, a limitation of scale and resolution that was overcome by SMN's higher-resolution model.
In January 2024, a prolonged heatwave in central Argentina assessed predictability of performance from medium-range to sub-seasonal scales. IFS ensemble forecasts captured its temporal and spatial evolution with good skill.
Future collaboration
This pilot project is the first documented effort to assess IFS operational performance over South America and improve severe weather prediction across forecasting systems. A key outcome is a comprehensive Technical Memorandum (https://www.ecmwf.int/en/elibrary/81664-verification-global-and-regional-nwp-models-over-south-america) detailing methodology, systematic verification and high-impact case studies. Future work will explore ensemble and machine-learning-based forecasts, such as those from the Artificial Intelligence Forecasting System (AIFS) and provide more detailed verification of extreme events. A journal article on Zonda predictability using different models and products is also in preparation. From 2026, South America will join the WMO WIPPS standardised verification score exchange (Manual No. 485, https://library.wmo.int/records/item/35703-manual-on-the-global-data-processing-and-forecasting-system), strengthening collaboration with global numerical weather prediction centres and supporting the Early Warnings for All initiative. This pilot may also serve as a blueprint for similar efforts across other regions of the world.