Seamless Hydrological prediction of east Indian summer monsoon and Variance Analysis of its meteorological and hydrological uncertainty (SHIVA)

点击翻译按钮获取中文摘要

Conducting predictions seamlessly, in a streamlined and uniform way for lead times ranging from one day to one season, is a relatively novel method in meteorological research; it provides the mutual benefit that is often found when two formerly separated research branches are merged. Probabilistic predictions by means of ensemble simulations, on the other hand, is the de-facto standard in operational meteorological forecasting, and it is becoming a standard in hydrological forecasting as well. The underlying physics and statistics of meteorological and hydrological ensembles are very different, however, especially along the full range of lead times. The main constraint for meteorological predictability is the uncertain initial state: once initialized, each ensemble member evolves according to deterministic dynamics, which largely characterizes meteorological predictability as a function of lead time. Naturally, that is one major constraint for hydrological predictability as well, but other factors, collectively termed hydrological uncertainty, are at least as important: an imperfect modeling structure, uncertain parameters, and (like in meteorology) uncertain initial states. Structural and parameter uncertainty are so important because a river basin is a far more heterogeneous object of study than it is the global atmosphere for meteorology. Simultaneous treatment of meteorological and hydrological uncertainties in the framework of seamless prediction poses a major challenge, accordingly, and coping with this challenge represents two out of three tasks in our proposed project SHIVA.The third task deals with a given seamless probabilistic streamflow forecast. We analyze the relative contributions of meteorological and hydrological uncertainty to the overall prediction. They can be quantified using a two-way analysis of variance (ANOVA) of the predicted streamflow for each lead time. Three aspects deserve attention: 1) we use dedicated predictands for the short, medium, and long range; 2) for any single lead time the meteorological and hydrological uncertainties reveal important sources of errors and, perhaps, possible remedies; 3) while the uncertainty estimates for the single lead times are likely affected by data insufficiencies, it should be possible to define a smooth function that maps lead times to relative uncertainties, revealing the principal limitations of seamless probabilistic streamflow prediction. The Mahanadi Catchment (A_C = 141,500km²) of East India constitutes an ideal site for the planned research. First, due to the strong atmosphere-ocean interaction in the tropical Monsoon belt, the potential for making skillful long-term predictions is high compared to, for example, the mid latitudes. Second, streamflow forecasts covering lead times from days to months are of utmost importance for effective water resources management in the Mahanadi Basin, including flood warning and control, reservoir operation, and irrigation.