Effects of land-use data resolution on hydrologic modelling, a case study in the upper reach of the Heihe River, Northwest China

Abstract

Land-use/land-cover (LULC) data represent important inputs in hydrologic models. The LULC data can affect modelled watershed hydrologic processes by altering the rates of interception, infiltration, evapotranspiration and groundwater recharge that govern the timing and volume of surface and river runoff. A thorough understanding of the impacts of LULC resolution on hydrologic modelling is thus essential and important. Generally, differences in the resolution of LULC data in hydrologic modelling can result in different interpretation accuracies, LULC classifications and spatial resolutions for one-year LULC datasets; typically, such models only use one-year LULC datasets in the simulation period to simply the calculation process. But how would the model performance change if multiple years of LULC data were input into hydrologic models? To better understand the impacts of LULC resolution on hydrologic modelling, this study input four LULC datasets into the SWAT (Soil and Water Assessment Tool) model and compared the impacts of the LULC datasets on the hydrologic modelling results for a high-elevation, cold and mountainous watershed in Northwest China. The first three datasets describe the LULC in the year 2000 and are based on different interpretation accuracies, LULC classifications and spatial resolutions. The last dataset is the yearly LULC data over a 20-year period. To incorporate the yearly LULC dataset, we modified the HRU (hydrologic response unit) division method and SWAT computational structure. The main findings were as follows. (1) A yearly LULC dataset may result in a higher simulation complexity in the SWAT model because this involves the most LULC patches. (2) If multiple years of LULC data exist, a LULC dataset with smaller time intervals, e.g. yearly is necessary in streamflow modelling in SWAT to get better model performance, the model performance may improve by 2.2%–13.9% compared to one-year LULC datasets. (3) If only one-year LULC datasets exist, we suggest inputting a dataset with a high resolution and remote sensing interpretation accuracy and reduced numbers of LULC classification types into SWAT to improve the hydrologic model performance; with such an approach, the model performance may improve by 1.1%–6.9% compared to other one-year LULC datasets.