arXiv subjects

Frederic Vitart

Publications and source records attributed to Frederic Vitart.

5 recordsLinked to original sources

arXiv2026

AIFS-SUBS: Extending Data-Driven Forecasting to Sub-Seasonal Timescales

Data-driven models now rival numerical weather prediction in the medium range, but extending them to sub-seasonal lead times raises challenges absent at shorter horizons. Errors accumulate over long autoregressive rollouts, systematic biases grow with lead time, and several years of data must be held out for independent verification, even though machine-learning models otherwise benefit from longer training records. To address these challenges, we adapt ECMWF's AIFS-CRPS medium-range model. AIFS-SUBS adopts a 24h autoregressive time step to reduce error accumulation, adds stratospheric levels and top-of-atmosphere thermal radiation as predictors, and reserves 2007--2011 as an independent verification window. We evaluate two config-durations: AIFS-SUBS, fine-tuned on operational analyses, and AIFS-SUBS-ERA5, trained on ERA5 alone. Across weeks 2--6, AIFS-SUBS matches the operational Integrated Forecasting System (IFS) in probabilistic skill while reducing systematic biases. For the convective (OLR) component of the Madden--Julian Oscillation (MJO), AIFS-SUBS extends skilful forecasts (correlation > 0.5) by eight days relative to the IFS, while matching or exceeding the IFS for the full multivariate RMM index. AIFS-SUBS also reproduces the observed MJO modulation of tropical cyclone activity comparably. Stratospheric skill is particularly strong with AIFS-SUBS reproducing sudden stratospheric warming (SSW) frequency and surface impact. In the AI Weather Quest, AIFS-SUBS-ERA5 attains a variable-averaged ranked probability skill score slightly ahead of the IFS at weeks 3 and 4. At inference, AIFS-SUBS uses about 200 times less energy than the IFS, opening the door to much larger real-time ensembles. AIFS-SUBS is ECMWF's first machine-learning model targeted at sub-seasonal time-scales.

physics.ao-ph↗

arXiv2025

Regularization of ML models for Earth systems by using longer model timesteps

Regularization is a technique to improve generalization of machine learning (ML) models. A common form of regularization in the ML literature is to train on data where similar inputs map to different outputs. This improves generalization by preventing ML models from becoming overconfident in their predictions. This paper shows how using longer timesteps when modelling chaotic Earth systems naturally leads to more of this regularization. We show this in two domains. We explain how using longer model timesteps can improve results and demonstrate that increased regularization is one of the causes. We explain why longer model timesteps lead to improved regularization in these systems and present a procedure to pick the model timestep. We also carry out a benchmarking exercise on ORAS5 ocean reanalysis data to show that a longer model timestep (28 days) than is typically used gives realistic simulations. We suggest that there will be many opportunities to use this type of regularization in Earth system problems because the Earth system is chaotic and the regularization is so easy to implement.

nlin.CD↗

arXiv2024

Ensemble reliability and the signal-to-noise paradox in ECMWF subseasonal forecasts

Ensemble forecasts can exhibit counterintuitive statistical properties such that the correlation between ensemble means and observations ($r_{mo}$) exceeds the correlation between ensemble means and individual members ($r_{mm}$). This behaviour has been interpreted as a `signal-to-noise paradox' (SNP), which is commonly diagnosed using the ratio of predictable components ($\textnormal{RPC} = \sqrt { r_{mo}^2 / r_{mm}^2 } $). Here, we emphasise the links between ensemble-size-invariant estimates of RPC and other metrics of ensemble reliability and derive a general closed-form expression for RPC in terms of $r_{mo}$, the spread-error ratio (SER), and total variance ratio (VR). Physical constraints on the admissible solutions provide a mechanism to identify statistically paradoxical sample estimates of RPC, $r_{mo}$, SER, and VR that correspond to combinations that are not possible without sampling uncertainty. We evaluate three atmospheric circulation indices in ECMWF subseasonal reforecasts. Large-ensemble NAO forecasts evaluated over 80 start dates satisfy reliability criteria within our estimated sampling uncertainties but exhibit high RPC values at some lead times. These lead times coincide with paradoxical combinations of correlation and reliability metrics that are impossible in the large-sample limit, indicating an important role for sampling uncertainties. Nevertheless, wintertime NAO indices averaged over days 16-45 exhibit more robust evidence for unreliability characterised by RPC$approx1.5$ suggesting that SNP-like behaviour observed in daily data during the period 2001-2020 is not solely attributable to sampling artefacts. However, these results do not generalise to other configurations of the same IFS model evaluated over 3120 start dates for the period 1959-2023. In these extended reforecasts, daily NAO indices are well-calibrated and RPC$\approx1$ for all lead times.

physics.ao-ph↗

arXiv2024

Improving forecasts of precipitation extremes over Northern and Central Italy using machine learning

The accurate prediction of intense precipitation events is one of the main objectives of operational weather services. This task is even more relevant nowadays, with the rapid progression of global warming which intensifies these events. Numerical weather prediction models have improved continuously over time, providing uncertainty estimation with dynamical ensembles. However, direct precipitation forecasting is still challenging. Greater availability of machine learning tools paves the way to a hybrid forecasting approach, with the optimal combination of physical models, event statistics, and user-oriented post-processing. Here we describe a specific chain, based on a random forest pipeline, specialised in recognizing favourable synoptic conditions leading to precipitation extremes and subsequently classifying extremes into predefined types. The application focuses on Northern and Central Italy, taken as a testbed region, but is seamlessly extensible to other regions and timescales. The system is called MaLCoX (Machine Learning model predicting Conditions for eXtreme precipitation) and is running daily at the Italian regional weather service of ARPAE Emilia-Romagna. MalCoX has been trained with the ARCIS gridded high-resolution precipitation dataset as the target truth, using the last 20 years of the ECMWF re-forecast dataset as input predictors. We show that, with a long enough training period, the optimal blend of larger-scale information with direct model output improves the probabilistic forecast accuracy of extremes in the medium range. In addition, with specific methods, we provide a useful diagnostic to convey to forecasters the underlying physical storyline which makes a meteorological event extreme.

physics.ao-ph↗

arXiv2023

Domino: A new framework for the automated identification of weather event precursors, demonstrated for European extreme rainfall

A number of studies have investigated the large-scale drivers and upstream-precursors of extreme weather events, making it clear that the earliest warning signs of extreme events can be remote in both time and space from the impacted region. Integrating and leveraging our understanding of dynamical precursors provides a new perspective on ensemble forecasting for extreme events, focused on building story-lines of possible event evolution. This then acts as a tool for raising awareness of the conditions conducive to high-impact weather, and providing early warning of their possible development. However, operational applications of this developing knowledge-base is limited so far, perhaps for want of a clear framework for doing so. Here, we present such a framework, supported by open software tools, designed for identifying large-scale precursors of categorical weather events in an automated fashion, and for reducing them to scalar indices suitable for statistical prediction, forecast interpretation, and model validation. We demonstrate this framework by systematically analysing the precursor circulations of daily rainfall extremes across 18 regional- to national-scale European domains. We discuss the precursor rainfall dynamics for three disparate regions, and show our findings are consistent with, and extend, previous work. We provide an estimate of the predictive utility of these precursors across Europe based on logistic regression, and show that large-scale precursors can usefully predict heavy rainfall between two and six days ahead, depending on region and season. We further show how for more continental-scale applications the regionally-specific precursors can be synthesised into a minimal set of indices that drive heavy precipitation. We then provide comments and guidance for generalisation and application of our demonstrated approach to new variables, timescales and regions.

physics.ao-ph↗