The use of geostationary satellite based rainfall estimation and rainfall-runoff modelling for regional flash flood assessment

The availability of rainfall triggered hazard information such as flash flood is crucial in the flood disaster management and mitigation. However, providing that information is mainly hampered by the shortage of data because of the sparse, uneven or absence the hydrological or meteorological observation. Remote sensing techniques that make frequent observations with continuous spatial coverage provide useful information for detecting the hydrometeorological phenomena such as rainfall and floods. This study aims to develop and evaluate geostationary satellite based rainfall estimation by considering cloud types and atmospheric environmental conditions. Furthermore, the satellite rainfall estimation is coupled with rainfall-runoff model for regional flash flood assessment. First, a simple rainfall estimation method using geostationary satellite i.e. Multifunctional Transport Satellite (MTSAT) blended with Tropical Rainfall Measuring Mission (TRMM) 2A12 is performed for Java Island, Indonesia and its surrounding area. The blending process is conducted by developing statistical relationship between cloud top temperature from MTSAT 10.8μm channel (TIR1) which is collocated with rainfall rate (RR) acquired by TRMM 2A12. Inter comparison with TRMM Multi Precipitation Analysis (TMPA) data product is conducted. Temporal validation result shows that TMPA demonstrated better statistical performance than TIR1 and RR statistical relationship model. However for the spatial correlation, TIR1 and RR statistical relationship model shows reasonably better performance than TMPA. Second, the rainfall estimation method basically uses an assumption the lower cloud top temperature is associated with heavier rainfall, particularly for convective cloud type. To fulfill such assumption, the statistical relationship is developed mainly for cumulonimbus (Cb) cloud type. A new two-dimensional threshold diagram (2D-THR) has been developed based on maximum likelihood cloud classification results, which can readily be applied for MTSAT split window datasets. The study area is Japan and its surrounding area. By integrating the cloud type classification especially by separating Cb cloud type from other cloud types can improve the TIR1 and RR statistical relationship, which is indicated by increasing correlation coefficient and the gradient of regression line. Therefore, underestimating rainfall intensity can be avoided by applying rainfall rate and cloud top temperature relationship that uses Cb cloud type only rather than using all cloud types. A good agreement between estimated and measured storm rainfall also has been demonstrated when use this approach. The geostationary satellite based rainfall estimation then applied for characterizing the storm severity. The Hosking-Wallis Regional Frequency Analysis (HW-RFA) method is used to define the frequency distribution of long-term hourly maximum rainfall over Hokkaido Island. HW-RFA indicates that Generalized Normal/Log Normal three parameters (GNO/LN3) is suitable to describe the frequency distribution of long-term hourly maximum rainfall over Hokkaido Island. A return period map during heavy rainfall event is generated by using MTSAT based rainfall estimation, based on the GNO/LN3 distribution. A comparison with AMeDAS return period of the same rainfall even demonstrates that that the return period information shown by MTSAT rainfall is comparable with AMeDAS rainfall return period. For assessing the return period of an extreme event in the area that observed rainfall is lacking, the use of geostationary satellite based is proved useful to overcome such problem. Third, total Precipitable Water Vapor (PWV) as a product of Global Positioning System observation and atmospheric vertical instability were considered to represent the atmospheric environmental conditions during the development of TIR1 and RR statistical models. The results demonstrated that the models that considered the combination of total PWV and atmospheric vertical instability were relatively more sensitive to heavy rainfall than were the models that considered no atmospheric environmental conditions. Intercomparison results with the TRMM 3B42 rainfall estimation product confirmed that MTSAT-based rainfall estimations made by considering atmospheric environmental conditions were more accurate for estimating rainfall generated by Cb cloud. Lastly, a regional flash flood assessment is conducted based on two rainfall-runoff models: (i) empirical regression model approach and (ii) physical based approach using land surface model. The empirical model uses the multiple regression approach to draw a relationship between the flash flood severity and hydrological, morphometrical and meteorological conditions. Particularly for flash flood severity related to hydrological condition the statistical relationship is strongly determined by initial soil moisture condition. The resulted empirical models shows that flash flood severity as the function of morphometrical factors can provide flash flood potential information. Moreover, flash flood severity as the function of hydrological and meteorological factors demonstrate more dynamic pattern since they are related to rainfall intensity distribution. The physical based approach for flash flood assessment had been conducted by implementing river flow simulation the minimal advance treatments of surface interaction and runoff (MATSIRO). The result indicates that the river flow simulated by MTSAT downscaled with relatively sparse rainfall observation is comparable with the river flow simulation using more dense rain observation network.