38 Development of Multimodel Streamflow Forecasts for Various River Basins in Different Hydroclimatic Settings

Monday, 8 January 2018
Exhibit Hall 3 (ACC) (Austin, Texas)
Mohamed Elansary, Texas A&M, Kingsville, TX; and T. Sinha

Handout (2.0 MB)

Seasonal and sub-seasonal streamflow forecasts, developed using climate-based forecasts, can be valuable in planning and management of water resources for meeting municipal and industrial water demands, scheduling agricultural operations, optimizing hydropower generation, and creating emergency plans for drought mitigation. However, the forecasting skill of monthly to seasonal streamflow forecasts varies between different seasons as well as over different geographical locations. Therefore, the objectives of this study are: a) Develop monthly updated retrospective streamflow forecasts for multiple River basins using climate forecasts information and evaluate the predictability skill during different seasons, and b) Reduce uncertainty in developed streamflow forecasts by multimodel combinations.

The Principal Component Regression (PCR) modeling is used to develop monthly updated streamflow forecasts using the International Research Institute of Climate and Society (IRI)'s Climate Predictability Tool (CPT). For each month, different PCR models are developed utilizing precipitation and air temperature forecasts from the ECHAM4.5 General Circulation Model (GCM) and Coupled Forecast System Model Version 2 (CFSv2) as well as previous month’s streamflow as predictors.The model is calibrated using the data from 1958 to 1985 for each month and the retrospective streamflow forecasts are developed over the 1986 to 2015 time period for the lead time of 1 to 7 months. The Spearman rank correlation and the Mean Square Skill Score (MSSS) are used to compare the performance of the developed forecasts with respect to the observed streamflow. The streamflow forecasts using ECHAM4.5 and CFSv2 models will be combined in order to reduce the uncertainty in forecasts. The streamflow forecasts will also be developed using the Variable Infiltration Capacity (VIC) land surface model to quantify the improvements in forecasting skills over statistical based streamflow forecasts.

Supplementary URL: https://docs.google.com/presentation/d/1nexWN33_KMbgQkWUEhdICFxx7NBkN1oQGaFY6pQi-Rk/export/pdf

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner