Prof. Xu Liang
University of Pittsburgh, USA
Room 1118, Building 3, IAP
16:39, May 31, 2019
Reliable real-time hydrological forecasting, such as prediction of floods, is invaluable to the society. However, modern high-resolution distributed models have faced challenges when dealing with uncertainties that are caused by large number of parameters and initial state estimations involved. Therefore, to rely on these high-resolution models for critical real-time forecast applications, considerable improvements on parameter and initial state estimation techniques must be made. In addition, issues related to high dimensionality/complexity associated with these high-resolution models must be addressed.
In this talk, I will first present a unified data assimilation algorithm called Optimized PareTo Inverse Modeling through Inverse STochastic Search (OPTIMISTS) to deal with the challenges of having robust flood forecasting for high-resolution distributed models. I will then introduce a dynamic fuzzy clustering approach to reduce the complexity of the state representations for data assimilation for these high-resolution hydrological models.
OPTIMISTS was tested on a low-resolution distributed land surface model using VIC (Variable Infiltration Capacity) and on a high-resolution distributed model using DHSVM (Distributed Hydrology Soil Vegetation Model). OPTIMISTS was also compared with a traditional particle filter and a variational method. Furthermore, performance of our proposed clustering approach is tested using DHSVM. Results of these tests will be presented and discussed.