Ensemble precipitation forecasts (EPFs) derived from single numerical weather predictions (NWPs) often miss extreme events, and individual hydrological models (HMs) often fail to accurately capture all types of flows, including flood peaks. To address these shortcomings, this study introduced four “EPF + HM” schemes for ensemble flood forecasting (EFF) by combining two EPFs and two HMs. A generator-based post-processing (GPP) method was applied to correct biases and under-dispersion within the raw EPF data. The effectiveness of these schemes in delivering high-quality flood forecasts was assessed using both deterministic and probabilistic metrics. The results indicate that, once post-processed by GPP, all proposed schemes show improvements in both deterministic and probabilistic performances, with skillful flood forecasts for 1–7 lead days. The deterioration in forecast performance with extended lead times is also lessened. Notably, the results indicate that uncertainty within hydrological models has a more pronounced impact on capturing flood peaks than uncertainty in precipitation inputs. This study recommends combining individual EPF with multiple hydrological models for reliable flood forecasting. In conclusion, effective flood forecasting necessitates employing post-processing techniques to correct EPFs and accounting for the uncertainty inherent in hydrological models, rather than relying solely on the uncertainty of the input data.

The number of numerical weather prediction (NWP) models is on the rise, and they are commonly used for ensemble precipitation forecast (EPF) and ensemble streamflow prediction (ESP). This study evaluated the reliabilities of two well-behaved NWP centers in the Observing System Research and Predictability Experiment (THORPEX) Interactive Grand Global Ensemble (TIGGE), the European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP), in EPF and ESP over a mountain river basin in China. This evaluation was carried out based on both deterministic and probabilistic metrics at a daily temporal scale. The effectiveness of two postprocessing methods, the Generator-based Postprocessing (GPP) method, and the Bayesian Model Averaging (BMA) method were also investigated for EPF and ESP. Results showed that: (1) The ECMWF shows better performances than NCEP in both EPF and ESP in terms of evaluation indexes and representation of the hydrograph. (2) The GPP method performs better than BMA in improving both EPF and ESP performances, and the improvements are more significant for the NCEP with worse raw performances. (3) Both ECMWF and NCEP have good potential for both EPF and ESP. By using the GPP method, there are desirable EPF performances for both ECMWF and NCEP at all 7 lead days, as well as highly skillful ECMWF ESP for 1~5 lead days and average moderate skillful NCEP ESP for all 7 lead days. The results of this study can provide a reference for the applications of TIGGE over mountain river basins.

Based on the ensemble precipitation forecast data in the summers of 2014–2018 from the Observing System Research and Predictability Experiment (THORPEX) Interactive Grand Global Ensemble (TIGGE), a comparative study of two multi-model ensemble methods, the Bayesian model average (BMA) and the logistic regression (LR), was conducted. Meanwhile, forecasts of heavy precipitation from the two models over the Wujiang River Basin in China for the summer of 2018 were compared to verify their performances. The training period sensitivity test results show that a training period of 2 years was the best for BMA probability forecast model. Compared with the BMA method, the LR model required more statistical samples and its optimal length of the training period was 5 years. According to the Brier score (BS), for precipitation events exceeding 10 mm with lead times of 1–7 days, the BMA outperformed the LR and the raw ensemble prediction system forecasts (RAW) except for forecasts with a lead time of 1 day. Furthermore, for heavy rainfall events exceeding 25 and 50 mm, the RAW and the BMA performed much the same in terms of prediction. The reliability diagram of the two multi-model ensembles (i.e., BMA and LR) was more reliable than the RAW for heavy and moderate rainfall forecasts, and the BMA model had the best performance. The BMA probabilistic forecast can produce a highly concentrated probability density function (PDF) curve and can also provide deterministic forecasts through analyzing percentile forecast results. With regard to the heavy rainfall forecast in mountainous areas, it is recommended to refer to the forecast with a larger percentile between the 75th and 90th percentiles. Nevertheless, extreme events with low probability forecasts may occur and cannot be ignored.

Aiming at the rainstorm days (≥50 mm/d) in the middle and lower reaches of the Yangtze River during 2010–2020, the obliquely rotated principal component in T-mode (PCT) method is used to classify the daily mean 850 hPa geopotential height, including Type 1 (vortex/shear line), Type 2 (frontal surface), Type 3 (warm shear line), Type 4 (warm inverse trough line), Type 5 (typhoon-westerly trough), and Type 6 (easterly wave). We studied the weather system configurations of different synoptic circulation patterns, their long-term trends, and their impacts on diurnal variations of heavy precipitation and drew the following conclusions: Type 1, Type 2, or Type 3 shows balanced double-peak frequencies of the start time of heavy precipitation during 06:00–08:00 BT and around 17:00 BT, respectively. For Type 1, dynamical lifting and thermal lifting play balanced roles, while for Type 2 and Type 3, dynamical lifting plays a key role. The number of rainstorm stations for Type 1 shows a slight increasing trend, while that for Type 2 or Type 3 shows a significant increasing trend. Type 4, Type 5, or Type 6 show a significant single peak frequency of the start time during 15:00–16:00. Type 5 and Type 6 are mainly affected by dynamical lifting along with favorable cape values, which can trigger rainstorms. The number of rainstorm stations for Type 4 or Type 6 shows a decreasing trend (that for Type 4 is more significant), while that for Type 5 shows a slightly increasing trend.

Peng, T.; Qi, H.X., and Wang, J.C., 2020. Case study on extreme flood forecasting based on ensemble precipitation forecast in Qingjiang Basin of the Yangtze River. In: Guido Aldana, P.A. and Kantamaneni, K. (eds.), Advances in Water Resources, Coastal Management, and Marine Science Technology. Journal of Coastal Research, Special Issue No. 104, pp. 178–187. Coconut Creek (Florida), ISSN 0749-0208.The ensemble precipitation forecasting provides a new idea for flood forecasting. The ensemble forecasting takes into account the uncertainty of initial field and model. A set of forecasting results can be obtained by inputting the ensemble precipitation forecasting products into hydrological model, which can avoid the limitation of single deterministic numerical forecasting results. Taking the typical flood process of Qingjiang Basin in June 2016 as an example, the flood forecast experiment was carried out based on the ensemble precipitation forecast products. Firstly, the flood forecast model was built on the Xin'anjiang hydrological Model, and the ensemble precipitation forecast of the European Center for Medium-Range Weather Forecasting (ECMWF) was estimated and used to drive basin flood forecasting model. The results show that the 51 members of ECMWF ensemble forecast can better capture this rainstorm process, the cumulative precipitation and precipitation process for 72 hours is close to the observation. The precipitation ensemble forecast drive the hydrological model, and can provide more precipitation forecasting information than deterministic forecasting. Moreover, it enriches the input information of hydrological model, to estimate the range of the peak flood discharge and the arrival time of flood peak. Therefore, we can get a probability of flood occurrence in different amounts level through the frequency analysis of flood peak, solves the problem of the accuracy of the single forecast result, and transforms the single deterministic accurate forecast into the probabilistic forecast, which can better meet the demand of risk information for flood control and disaster reduction.

Numerical models often systematically underestimate the intensity of heavy precipitation and struggle to accurately represent the spatial pattern of rainbands over MLYR. This study converts the Spatial Fractional Skill Score (FSS) into a differentiable loss function and the traditional loss function Mean Squared Error (MSE) trained the U-Net deep learning architecture to construct a multi-model ensemble post-processing model that optimizes for spatial pattern and extreme intensity. For heavy precipitation (≥ 25 mm/d), a two-stage data-augmentation fine-tuning strategy is employed to further calibrate the multi-model ensemble outputs. We evaluated the performance of four models, ensemble mean (MEAN), and the multi-model ensemble U-Net(FSS) and U-Net(MSE) for daily precipitation forecasts. The results demonstrate that within 24–240 h lead time, U-Net(FSS) reduces the averaged RMSE by 3–7% compared to the best individual model, with the improvement increasing as the forecast lead time extends. For heavy precipitation(≥25 mm/d), the FSS of U-Net improvesby approximately 10% over the best individual model at lead times of 24–72 h, maintaining an improvement of over 5% through 168 h. For extreme precipitation (95th percentile), the U-Net sustains TS around 0.28 at 168–240h lead times, improving by 10–20% compared to MEAN. A case study of rainstorms in summer 2024 reveals that the U-Net outperforms both individual models and MEAN in capturing the spatial structure of rainbands and the intensity of extreme centers at lead times of 24, 168, and 240 h. The multi-model ensemble U-Net shows significant potential for optimizing the prediction of both the spatial and intensity of heavy precipitation forecasting.The applicability and improvement in heavy precipitation forecasting were discussed to inform future applications of multi-model ensemble techniques.Key words: TIGGE,Multi-Model Ensemble Forecasting,Spatial FSS Loss Function，U-Net Deep Learning Model,Heavy Precipitation Forecasting