The LAI calculated by the original SWAT was primarily affected by land use/cover types, water stress, temperature stress, nitrogen stress, and phosphorus stress. Only one meteorological station exist in the Bayin River Basin, and all of the vegetation is natural and without fertilisation. Therefore, in the original SWAT model, the spatial heterogeneity of LAI was determined by the land use/cover types; that is, the simulated LAI values were the same for each land cover type (Figure 7). This does not conform to the spatial heterogeneity of vegetation coverage within land cover types. However, the modified SWAT incorporated the remotely sensed LAI (that is, the vegetation coverage data from the Bayin River Basin) and thus overcame the limitation of the original SWAT.
Most remotely sensed LAI products have gaps and missing values, and either underestimate or overestimate the LAI in many areas (Fensholt et al., 2004; Sun et al., 2014; Li et al., 2018). The GLASS-based LAI used in this study was spatially and temporally continuous with no gaps and missing values and had high quality and accuracy in China (Zhao et al., 2013; Li et al., 2018; Xie et al., 2019). Furthermore, the GLASS LAI was downscaled, and higher resolution LAI data would reduce the mismatch boundary of HRUs in SWAT models (Ma et al., 2019).
Coupling the SWAT model with the MODIS LAI product has been used for enhanced modelling of green vegetation dynamics (in tropical or subtropical areas) and crop patterns (in semi-arid areas), and may improve the applicability of SWAT in corresponding areas (Ma et al., 2019; Pual et al., 2020). The improvement of the SWAT in these studies was catalogued by the remotely sensed LAI, which could properly capture a more realistic plant phenology and patterns at a higher resolution. The remotely sensed LAI could capture the vegetation coverage of the grassland and barren land, which are the main land cover types for the whole study area. However, considerable areas of barren land have been converted to grassland under the impact of climate change and artificial vegetation restoration, and the vegetation coverage of grassland has increased in the Bayin River Basin (Jin et al., 2019). These changes can be elucidated in the LAI with high spatial and temporal resolution derived in this study.
The improved LAI can impact the simulation of canopy interception loss, soil water content, water budget in SWAT model (Jaromir et al., 2010; Zheng et al., 2018; Ma et al., 2019). Further, the erosion process would be impacted because the SWAT model compute erosion caused by rainfall and runoff with the Modified Universal Soil Loss Equation (MULSE). In MUSLE, the average annual gross erosion is predicted as a function of runoff factor. The spatial and temporal accuracy of the LAI was confirmed as being of crucial importance for SWAT predictoins. Indeed, streamflow, sediment yield and ET which are the vegetation coverage-affected processes were more accuracy when using the remotely sensed high temporal and spatial resolution LAI.
SWAT models are commonly calibrated and validated with streamflow and sediment yield data at the outlet of a watershed, which improves the reliability of the model simulations (Gassman et al., 2007; Jin et al., 2015; Yang et al., 2020). However, model performance should be analysed at a more detailed scale as well, especially when testing high-resolution input data (Alemayehu et al., 2017; Pual et al., 2020). This study used high-resolution (1 km) remotely sensed ET data (SSEBop) to validate the original and modified SWAT models at the subbasin and HRU levels. SSEBop is also known to have a higher quality than several other datasets in grass covered surface (Herman et al., 2018; Dembélé et al., 2020).
The major limitations of this study are as follows: (i) There is only one meteorological station in the study area, which presented the only set of meteorological data. Although, two paraneters (PLAPS, precipiattion lapse rate; TLAPS, temperature lapse rate) were used to modify the precipitaion and temperature of the study area. This would still impose uncertainties in hydrological process modelling, especially in mountainous areas. (ii) Piecewise linear interpolation was used to interpolate the 8 d and 30 m LAI data to the daily time interval. However, the LAI may not vary linearly, and this may cause some uncertainties.