Multi-source Precipitation Observations And Fusion For Hydrological Applications In The Yangtze River Basin

The Yangtze River basin is known to be extremely susceptible to frequent floods in China. The East Asia Monsoon and the effects of complex regional terrains have shaped a complicated rainfall regime in the basin during the flood season, which is characterized by great spatiotemporal variability. Therefore, as a primary way to improve the accuracy of hydrological prediction, adequate monitoring of precipitation has great implications for flood control, disaster mitigation and water resources exploitation and utilization in the Yangtze River basin. To address this essential issue, this dissertation conducts a research on synthesizing applications of multi-source precipitation observations, including ground gauge network, weather radar and space-borne sensors, as well as on data fusion method for multi-source precipitation observations, over the Yangtze River basin.First, over the mountainous Three Gorges Region(TGR), the spatial variation and diurnal cycle of hourly precipitation are analyzed with records from a regional dense network of gauges, which was installed after the Three Gorges Reservoir started to store water. The results show that there are obvious orographic effects on the pattern of space-time variations for hourly precipitation in the TGR. Further, by applying an empirical spatial function, the correlation scale is identified as only 30 km for hourly precipitation in the warm season. The short correlation scale thus indicates that the current gauge network is perhaps insufficient to capture localized rainstorms during flood season. To make up the deficiency by gauge measurement, as an alternative data, China CINRAD weather radar observation is employed to develop the radar quantitative precipitation estimation(QPE) algorithms for the TGR. The radar QPE algorithms contain the procedures to deal with beam blockage and ground clutter, which both are typical errors for radar observation in the mountainous area, while rain type identification as well as multiple Z-R relations are also taken as part of the algorithms. According to the algorithms, a set of hourly radar QPE product at a resolution of 1 km is generated for further evaluation and applications during the 2010 flood season. The evaluation results suggest that the radar QPE product derived total precipitation amount is well correlated with the gauge data during typical rainstorm events, while it tends to slightly underestimate hourly precipitation in general. At the same time, the evaluation work indicates that, beyond the detection range of 150 km, the accuracy of radar QPE will decrease remarkably. For hydrological applications of radar QPE product, a rain-cell identification algorithm is proposed to characterize the geometrical features of summer rainstorm, and the results find a highly localized pattern for summer TGR rainstorms, with typical scales of 13~16 km and 5~17 km in the major and minor radius directions, respectively. Further, by coupling radar QPE data with the TGR distributed hydrological modeling system, the benefits from high-resolution radar QPE for TGR flood simulation are demonstrated.Then the investigation is extended into the whole Yangtze River basin, and for this application purpose, several sets of satellite-based precipitation products are evaluated and intercompared at annual, seasonal and daily scales, respectively. The results suggest that, the gauge-adjusted 3B42 V7 has the best performance in the Yangtze River, while near-real-time products always show error characteristics that depend on terrain, season and land cover. By applying a distributed hydrological model to whole Yangtze River basin, the modeling-based evaluation results indicate that 3B42 RT and CMORPH can be used as the forcing data for hydrological modeling in the midstream and downstream of and in the upstream of the Yangtze River, respectively. Based on the evaluation and error analysis work, the additive model and multiplicative model are further adopted to quantify the errors with satellite estimated monthly precipitation and daily precipitation, respectively. As the prior information of precipitation ground truth can be obtained from the gauge data by Kriging interpolation method, then the Bayesian data fusion approach is developed by combing precipitation observation information from both satellites and gauges. The validation results find that this fusion method can effectively remove those errors of satellite-based precipitation products, while it is able to provide an uncertainty interval for merged precipitation estimates at the same time.