Correcting for model changes in statistical postprocessing – an approach based on response theory

Abstract

Abstract. For most statistical postprocessing schemes used to correct weather forecasts, changes to the forecast model induce a considerable reforecasting effort. We present a new approach based on response theory to cope with slight model changes. In this framework, the model change is seen as a perturbation of the original forecast model. The response theory allows us then to evaluate the variation induced on the parameters involved in the statistical postprocessing, provided that the magnitude of this perturbation is not too large. This approach is studied in the context of a simple Ornstein–Uhlenbeck model and then on a more realistic, yet simple, quasi-geostrophic model. The analytical results for the former case help to pose the problem, while the application to the latter provides a proof of concept and assesses the potential performance of response theory in a chaotic system. In both cases, the parameters of the statistical postprocessing used – the Error-in-Variables Model Output Statistics (EVMOS) method – are appropriately corrected when facing a model change. The potential application in an operational environment is also discussed.

Verification: Assessment of Calibration and Accuracy

“…Lang et al ( 2020 ) have recently shown that sliding training period windows that incorporate at least some historical data have an advantage over the conventional approach. In an interesting proof-of-concept, Demaeyer and Vannitsem ( 2020 ) investigate a technique based on response theory to incorporate small model changes into the estimation of post-processing coefficients.…”

Calibrated ensemble forecasts of the height of new snow using quantile regression forests and ensemble model output statistics

“…The forecasts are obtained by a chain of ensemble numerical simulations. The 10-member reforecasts of the PEARP ensemble NWP (Descamps et al, 2015;Boisserie et al, 2016) are downscaled by the SAFRAN system (Durand et al, 1999) to obtain a meteorological forcing adjusted in elevation. The Crocus multilayer snowpack model, part of the S2M modelling chain (Vernay et al, 2019), is forced by these forecasts to provide ensemble simulations of HN, accounting for all the main physical processes explaining the variability in HN for a given precipitation amount, namely the dependence of falling snow density on meteorological conditions, the mechanical compaction over time depending on snow weight, the microstructure and wetness of the snow, a possible surface melting, and so on.…”

“…Time-adaptive training based on operational systems is an alternative to favour the homogeneity of the data set. Although new theories are emerging to face the challenge of model evolutions (Demaeyer and Vannitsem, 2020), several consistent recent studies show that the length of the calibration period is more critical than the strict homogeneity of data sets to forecast rare events (Lang et al, 2020;Hess, 2020). In the case of HN forecasts from EMOS (Nousu et al, 2019), even a 4-year calibration period was detrimental for the reliability of severe snowfall events compared to a longer heterogeneous reforecast.…”

Calibrated ensemble forecasts of the height of new snow using quantile regression forests and Ensemble Model Output Statistics

“…Time adaptative training based on operational systems is an alternative to favour the homogeneity of the dataset. Although new theories are emerging to face the challenge of model evolutions (Demaeyer and Vannitsem, 2020), several consistent recent studies show that the length of the calibration period is more critical than the strict homogeneity of datasets to forecast rare events (Lang et al, 2020;Hess, 2020). In the case of HN forecasts from EMOS (Nousu et al, 2019), even a 4-year calibration period was detrimental for the reliability of severe snowfall events compared to a longer heterogeneous reforecast.…”