Post-processing multi-ensemble temperature and precipitation forecasts through an Exchangeable Gamma Normal model and its Tobit extension

TL;DR

This paper introduces a novel exchangeable gamma normal model with a Tobit extension for post-processing multi-ensemble temperature and precipitation forecasts, improving calibration and sharpness of probabilistic weather predictions.

Contribution

It proposes a new collaborative mixed effect model based on exchangeability and invariance, utilizing multi-ensemble data for enhanced probabilistic weather forecasting.

Findings

Post-processed ensembles are better calibrated.

Forecasts are sharper after post-processing.

Model performs well on real meteorological data.

Abstract

Meteorological ensembles are a collection of scenarios for future weather delivered by a meteorological center. Such ensembles form the main source of valuable information for probabilistic forecasting which aims at producing a predictive probability distribution of the quantity of interest instead of a single best guess estimate. Unfortunately, ensembles cannot generally be considered as a sample from such a predictive probability distribution without a preliminary post-processing treatment to calibrate the ensemble. Two main families of post-processing methods, either competing such as BMA or collaborative such as EMOS, can be found in the literature. This paper proposes a mixed effect model belonging to the collaborative family. The structure of the model is based on the hypothesis of invariance under the relabelling of the ensemble members. Its interesting specificities are as follows: 1) exchangeability, which contributes to parsimony, with a latent pivot variable synthesizing the essential meteorological features of the ensembles, 2) a multi-ensemble implementation, allowing to take advantage of various information so as to increase the sharpness of the forecasting procedure. Focus is cast onto Normal statistical structures, first with a direct application for temperatures, then with its Tobit extension for precipitation. Inference is performed by EM algorithms with recourse made to stochastic conditional simulations in the precipitation case. After checking its good behavior on artificial data, the proposed post-processing technique is applied to temperature and precipitation ensemble forecasts produced over five river basins managed by Hydro-Qu$\'e$bec. These ensemble forecasts were extracted from the THORPEX Interactive Grand Global Ensemble (TIGGE) database. The results indicate that post-processed ensemble are calibrated and generally sharper than the raw ensembles.