An experimental seasonal hydrological forecasting system over the Yellow River basin – Part 2: The added value from climate forecast models

Abstract

Abstract. This is the second paper of a two-part series on introducing an experimental seasonal hydrological forecasting system over the Yellow River basin in northern China. While the natural hydrological predictability in terms of initial hydrological conditions (ICs) is investigated in a companion paper, the added value from eight North American Multimodel Ensemble (NMME) climate forecast models with a grand ensemble of 99 members is assessed in this paper, with an implicit consideration of human-induced uncertainty in the hydrological models through a post-processing procedure. The forecast skill in terms of anomaly correlation (AC) for 2 m air temperature and precipitation does not necessarily decrease over leads but is dependent on the target month due to a strong seasonality for the climate over the Yellow River basin. As there is more diversity in the model performance for the temperature forecasts than the precipitation forecasts, the grand NMME ensemble mean forecast has consistently higher skill than the best single model up to 6 months for the temperature but up to 2 months for the precipitation. The NMME climate predictions are downscaled to drive the variable infiltration capacity (VIC) land surface hydrological model and a global routing model regionalized over the Yellow River basin to produce forecasts of soil moisture, runoff and streamflow. And the NMME/VIC forecasts are compared with the Ensemble Streamflow Prediction method (ESP/VIC) through 6-month hindcast experiments for each calendar month during 1982-2010. As verified by the VIC offline simulations, the NMME/VIC is comparable to the ESP/VIC for the soil moisture forecasts, and the former has higher skill than the latter only for the forecasts at long leads and for those initialized in the rainy season. The forecast skill for runoff is lower for both forecast approaches, but the added value from NMME/VIC is more obvious, with an increase of the average AC by 0.08-0.2. To compare with the observed streamflow, both the hindcasts from NMME/VIC and ESP/VIC are post-processed through a linear regression model fitted by using VIC offline-simulated streamflow. The post-processed NMME/VIC reduces the root mean squared error (RMSE) from the post-processed ESP/VIC by 5-15 %. And the reduction occurs mostly during the transition from wet to dry seasons. With the consideration of the uncertainty in the hydrological models, the added value from climate forecast models is decreased especially at short leads, suggesting the necessity of improving the large-scale hydrological models in human-intervened river basins.

An experimental seasonal hydrological forecasting system over the Yellow River basin – Part 1: Understanding the role of initial hydrological conditions

“…While the companion paper will focus on the evaluation of the North American Multimodel Ensemble (NMME)-based seasonal hydrological forecasting (Yuan, 2016), this paper introduces the system and uses it to investigate the role of initial hydrological conditions (ICs) over the Yellow River basin.…”

“…The assessments conditional on the surface and subsurface water state variables, and the dry/wet conditions, are being investigated. Seasonal hydrological forecasting with multiple climate forecast models will be evaluated in a companion paper, by comparison with the ESP-based hydrological forecasting (Yuan, 2016).…”

Potential Threats from Variations of Hydrological Parameters to the Yellow River and Pearl River Basins in China over the Next 30 Years

“…This is because heavy human water consumption has led to the observed streamflow being consistently lower than the naturalized streamflow for the middle and lower reaches of the Yellow River Basin [10,14]. It is difficult to handle the physical processes of irrigation and interbasin water diversion in most hydrological models because of the scarcity of management data and the deficiency regarding the human component in such models [14]. The projection and analysis of natural runoff change are useful for water balance research and water resources management [10].…”

Assessment of an ensemble seasonal streamflow forecasting system for Australia

“…This is a key difference with approaches that post-process streamflow forecasts separately at each lead time (e.g. Yuan, 2016), as it means that each FoGSS time series forecast is a continuous hydrograph that can be summed to produce reliable ensembles of e.g. seasonal inflow totals.…”