Deep-learning-based post-processing for probabilistic precipitation forecasting

Abstract

Ensemble prediction systems (EPSs) serve as a popular technique to provide probabilistic precipitation prediction in short- and medium-range forecasting. However, numerical models still suffer from imperfect configurations associated with data assimilation and physical parameterization, which can lead to systemic bias. Even state-of-the-art models often fail to provide high-quality precipitation forecasting, especially for extreme events. In this study, two deep-learning-based models—a shallow neural network (NN) and a deep NN with convolutional layers (CNN)—were used as alternative post-processing approaches to further improve the probabilistic forecasting of precipitation over China with 1–7 lead days. A popular conventional method—the censored and shifted gamma distribution-based ensemble model output statistics (CSG EMOS)—was used as the baseline. Re-forecasts run using a frozen EPS—Global Ensemble Forecast System version 12—were collected as the raw ensembles spanning from 2000 to 2019. The re-forecast data were generated once per day and consisted of one control run and four perturbed members. We used the calendar year 2018 as the validation period and 2019 as the testing period, and the remaining 18 years of data were used for training. According to the results, in terms of the continuous ranked probability score (CRPS) and the Brier score, the CNN model significantly outperforms the shallow NN model, as well as the CSG EMOS approach and the raw ensemble, especially for heavy or extreme precipitation events (those exceeding 50 mm/day). A remarkable degradation was seen when reducing the size of training samples from 18 years of data to two years. The spatial distribution of the CRPS shows that the stations in central China were better calibrated than those in other regions. With a lead time of 1 day, the CNN model was found to be superior to the other models (in terms of the CRPS) at 74.5% of the study stations. These results indicate that deep NNs can serve as a promising approach to the statistical post-processing of probabilistic precipitation forecasting.

Sub-Seasonal Precipitation Bias-Correction in Thailand Using Attention U-Net With Seasonal and Meteorological Effects

“…Moreover, Attention U-net architect [8] has been used on precipitation bias-correction tasks and has shown promising results. Ji Y. et al [9] implemented two CNN networks as a post processing method for probabilistic forecasting.…”

“…Due to their simplicity and explainability, statistical approaches such as statistical downscaling [4] and quantile mapping [5] have also been among the most often used methods for precipitation bias-correction for decades. Recently, however, deep learning has significantly advanced and excelled in many meteorological applications, especially convolutional neural networks (CNN) [6] [7][8] [9]. Further studies have also been carried out concerning the impact of various meteorological variables [10] [11] on S2S precipitation.…”

Deep Learning for Daily 2‐m Temperature Downscaling

“…It shows promise in terms of both accuracy and efficiency, surpassing previous methods (Höhlein et al., 2020). However, the use of deep learning methods in the field of meteorological downscaling is still in its early stages and faces challenges such as inadequate description of complex features and poor performance in extreme events (Baño‐Medina et al., 2020; Y. Ji, Gong, et al., 2022; Y. Ji, Zhi, et al., 2022; Y. Ji, Zhi, et al., 2023; Vandal et al., 2019). Therefore, further practical exploration and research are needed to address these issues.…”

“…He et al., 2016; Wilby et al., 1998). With the advent of the big data era, deep learning has the potential to discover features in high‐dimensional data and capture the underlying nonlinear relationships between various meteorological variables (Y. Ji, Gong, et al., 2022; Y. Ji, Zhi, et al., 2022; Yuan et al., 2020; Zhi & Wang, 2023; Zhi et al., 2022). It shows promise in terms of both accuracy and efficiency, surpassing previous methods (Höhlein et al., 2020).…”

Adapting a deep convolutional RNN model with imbalanced regression loss for improved spatio-temporal forecasting of extreme wind speed events in the short to medium range

“…Other authors have explored the potential of machine learning in this context. Ji et al (2022), for example, investigate two deep-learning-based post-processing approaches for ensemble precipitation forecasts and compare these against the censored and shifted gamma-distribution-based ensemble model output statistics (CSG EMOS) method. The authors report significant improvements of the DL-based approaches over the CSG EMOS and the raw ensemble, particularly for extreme precipitation events.…”