Hydrologic Modeling of a Sub-continental Scale Mountainous River Basin - Case Study of the Head Area of Yellow River

The mountainous head area of Yellow River contribute about 1/3 of total water yield in Yelow River Basin, while the area of this region account only for 1/8 approximately. Water yield in the head region decreased from 695 m3 (1960-1989) to 526 m3 (1990s). The water yield decrease and water use increase lead to no-water events happened every year in the downstream main channel of Yellow River in the 1990s. Deeper and better understanding of water cycle and exploring the relationship between climate change, LULC change and water cycle are necessary. Area of the head area (the upstream above Tangnaihai hydrologic station) is about 114,345 km2. Snowfall-snowmelt process have great impact on hydrologic simulation. The objective of this study is to construct distributed hydrological model that can accurately simulate hydrologic process in such a Sub-continental Scale Mountainous River Basin. Three methods were used to describe the snowfall-snowmelt component in the framework of SWAT. The first and second methods incorporate the change of precipitation, temperature and relative humidity with elevation into the degree-day snowfall-snowmelt algorithm already in SWAT. The difference between the first and second method is: the first method use the same elevation bands, which is extracted from the whole basin's elevation distribution, for the all subbasins; while the second method extract elevation band for each subbasin according to each subbasin's elevation distribution. The first and second methods operate on HRU (Hydrological Response Unit) scale. The third method is SNOW17 snowmelt algorithm, which an energy budget approach is accounting for the energy balance at the ground surface. In order to implement SNOW17, the basin needs to be discretized at grid scale and extract each grid's elevation, aspect and slope information. The SNOW17 method is more complex and physically-based than the first and second methods. Below, we use SWAT-ElevBand denote first method, SWAT-SubbasinElevBand denote second method and SWAT- SNOW17 denote third method. The three methods were evaluated with monthly runoff at Tangnaihai hydrologic station during 1975 and 1990, SCE-UA optimization method was used to calibrate relevant parameters. The evaluation parameter RE show that SWAT- ElevBand, which underestimate about 30% runoff with default parameters and 18% runoff even with calibrated parameters, is not good enough to accurately predict runoff in the study area. SWAT- SNOW17 is the best model to predict runoff with default parameters, whose relative error is about -11%. While SWAT-SubElevBand is better than SWAT- SNOW17 with the calibrated parameters, whose relative error is -3.02% for 1975 to 1985 and only -0.31% for 1986 to 1990. Coefficient of Determination and Nash-Sutcliffe coefficient for SWAT- SNOW17 and SWAT-SubElevBand are both equal to or above 0.79, which reveal that these two model can capture the hydrograph shape well. Generally, SWAT- ElevBand model can't simulate runoff accurately even after being calibrated. SWAT- SNOW17 can predict runoff more accurate than SWAT-SubElevBand with default parameters, while SWAT-SubElevBand can give similar or better results than SWAT- SNOW17 if there are enough data for parameters calibration.