5:15 PM - 7:15 PM
[AAS01-P10] The predictability of global river discharge forecast at sub-seasonal to seasonal (S2S) timescale
Keywords:river discharge, sub-seasonal to seasonal, predictability
Accurate sub-seasonal to seasonal (S2S) hydrological quantities predictions greatly benefit for variety of applications, such as disaster preparedness, water resources management, and agriculture. However, S2S timescale is considered as new frontier for predictability research and is previously limited-explored forecasting timescale for flood prediction. In this study, we conduct a comparative analysis of river discharge predictability at the sub-seasonal scale across global river basins. By leveraging extended-range S2S forcings into the Integrated Land Surface (ILS) model, which couples emulator of MATSIRO land surface model and hydrodynamic processes from CaMa-Flood, we generate sub-seasonal river discharge forecasts with lead times of up to 45 days.
This research evaluates S2S river discharge forecasts by comparing simulations driven by S2S-ECMWF forcings and post-processed S2S precipitation data from support vector machine (SVM)-based method against observed river discharge dataset from GRDC, used as a benchmark. Instead of relying solely on ensemble means, we utilize the full set of 51 ensemble members from the S2S-ECMWF forecast then conduct the simulation using ILS. To enhance the forecast's applicability, we also perform a probabilistic flood risk based on return period assessments for further analysis.
Prior to S2S simulation, we performed a historical simulation of daily river discharge and evaluation to establish the initial conditions. The results of the forecasted S2S simulation suggests that applying SVM-based post-processing approach to precipitation is also applicable to enhance river discharge forecasts in some river stations to some extent. Despite it is not a significant improvement, the SVM-forced river discharge forecast demonstrates higher skill than ECMWF-forced forecast in later lead times. According to KGE and R2 globally averaged value, skill enhancement becomes noticeable between 10 and 45 days of lead time. Moreover, as S2S forecast has longer lead time, the skillful predictions are observed up to 20 days, as indicated by nRMSE value less than 1. Additionally, among of 588 river stations analyzed as sample data, there are 205 stations showed improvement in week 4.
Overall, the accuracy of S2S prediction is essential to provide seamless hydrological prediction, further works we will try to enhance the sites and implement the forecasted data for flood risk probability assessment.
This research evaluates S2S river discharge forecasts by comparing simulations driven by S2S-ECMWF forcings and post-processed S2S precipitation data from support vector machine (SVM)-based method against observed river discharge dataset from GRDC, used as a benchmark. Instead of relying solely on ensemble means, we utilize the full set of 51 ensemble members from the S2S-ECMWF forecast then conduct the simulation using ILS. To enhance the forecast's applicability, we also perform a probabilistic flood risk based on return period assessments for further analysis.
Prior to S2S simulation, we performed a historical simulation of daily river discharge and evaluation to establish the initial conditions. The results of the forecasted S2S simulation suggests that applying SVM-based post-processing approach to precipitation is also applicable to enhance river discharge forecasts in some river stations to some extent. Despite it is not a significant improvement, the SVM-forced river discharge forecast demonstrates higher skill than ECMWF-forced forecast in later lead times. According to KGE and R2 globally averaged value, skill enhancement becomes noticeable between 10 and 45 days of lead time. Moreover, as S2S forecast has longer lead time, the skillful predictions are observed up to 20 days, as indicated by nRMSE value less than 1. Additionally, among of 588 river stations analyzed as sample data, there are 205 stations showed improvement in week 4.
Overall, the accuracy of S2S prediction is essential to provide seamless hydrological prediction, further works we will try to enhance the sites and implement the forecasted data for flood risk probability assessment.