Assessment of land cover resolution impact on flood modeling uncertainty

The main objective of this paper is to evaluate the impact of land cover resolution, in comparison with the digital elevation model (DEM) resolution, on hydrological modeling outputs. Three different basins in the various resolutions of DEM (12.5, 25, 50, 100, 500 and 1,000 m) and land-use maps (250, 1,000 and 2,500 m) were collected in this study, and the hydrological modeling process was performed using the Soil and Water Assessment Tool (SWAT) model. The soil type resolution was 1,000 m for all basins, and the runoff modeling was done based on the Soil Conservation Service Curve Number (SCS-CN) method. The final model outputs showed that the DEM cell size variations affect significantly the topographical characteristics of a catchment such as area, mean slope, river network and time to concentration which alter the flood modeling outputs especially in hilly watersheds (mean slope more than 15%) up to 15% for a DEM cell size of 1,000 m in comparison to 12.5 m. Also, the resolution and spatial distribution of land cover maps which directly specify SCS-CN values, can change the output simulated runoff results up to 49% for a land cover cell size of 2,500 m in comparison to 250 m. These results indicated that the quality of the land cover map is more important than the quality of DEM in hydrological modeling. Also, the results showed that for an identical land-use cell size, the differences between model outputs using DEM cell sizes less than 100 m were not very significant. Furthermore, in all models by increasing the DEM cell size, the simulated runoff depth was decreased.


INTRODUCTION
Hydrological models are useful to identify present and future water resources, and they help researchers and water managers to estimate the spatial variability in resource over watersheds. Simulation models can be classified into two different types; physical-based models and empirical models. There are various physical-based models designed to simulate hydrological processes, such as Areal Nonpoint A decade ago, the lack of DEM availability was a big problem; however, it has been solved recently by using global remotely sensing DEM products (Lin et al. ). In some studies, the impact of DEM resolution on other aspects of hydrological modeling was investigated. Chaplot () obtained better results in runoff and sediment modeling by using DEM with 50 m resolution in comparison to 500 m, and finally, DEM with (100-300 m) resolution was suggested in large watersheds. Similar results were also achieved by Reddy & Reddy (). Yang et al. () found that higher DEM resolution could provide a more accurate representation of topographic features. Yang & Chu (a, b) showed that the simulation results were

MATERIALS AND METHODS
The impact of input DEM and LULC map quality on hydrological modeling have been investigated using three schematic watersheds which have similar areas but different properties (mean slope, topography and LULC) in Also, three land-use maps with different grid cell sizes as 250, 1,000 and 2,500 m have been used for CN estimation in each watershed (Figures 1-3). These maps were downloaded from various digital map websites (www.FAO.org, www. Globallandcover.com and www.Mapcruzin.com).
The climate data (precipitation, relative humidity, solar radiation, temperature and wind speed) which are needed for every SWAT model, were collected as daily data for all 22 stations in the vicinity of the watersheds for more than 30 years . The climate data as raster files were calculated using 22 stations and based on the inverse distance weight technique in ArcGIS. The mean annual rainfall maps (30 years) are represented in Figure 4 for each watershed.

Study area
The study areas are located in a semi-arid region in northeastern China (Figures 1-3). Table 1 summarizes the topographical properties of the three selected study areas.
In addition to the slope, the main difference between  Figure 4 shows the variation of land cover components in various grid resolutions in the study areas.

Hydrological modeling
The hydrological modeling of these watersheds has been implemented using the SWAT model. SWAT is a conceptual, fully distributed, physically based, continuous model which evaluates all components of the water cycle (runoff, evapotranspiration and percolation) within a watershed.
The capability of the SWAT model in analyzing various detailed input data is an advantage in hydrological modeling.
Using input maps (DEM, LULC and soil type), SWAT dissects the watershed into multiple HRUs so that each contains almost uniform land cover and soil type; then, assigns a mean CN to HRU based on input LULC and soil type maps. In fact, a CN grid was made first in which, each pixel value was calculated based on corresponding The SCS-CN method is expressed as follows (Mishra & Singh ): where Q is the discharge depth (mm), P is the daily rainfall depth (mm), CN is the curve number (dimensionless), S is water balance equation can model the water cycle across a watershed.
For each watershed, a complete SWAT model was implemented and calibrated using SWAT-CUP software (total 18 models for each watershed).

RESULTS AND DISCUSSION
Determining hydrological responses of a watershed is a very Area and slope   Also, by increasing the DEM cell size, the distribution of slope in a watershed was changed. This distribution in hilly regions is very important because this factor directly affects the amount and shape of the runoff hydrograph. For comparison, Figure 6 shows the slope distribution within each watershed for two DEM cell sizes of 12.5 and 1,000 m. As can be seen, almost all areas with slopes of more than 30% in the high-resolution DEMs (12.5 m), are shifted to the slopes of less than 30% in the low-resolution DEMs (1,000 m).

River network and sub-basins
In addition to the area and slope of the watersheds, changes in DEM cell sizes also affect the river network and consequently, delineation of sub-basins. Because the slope and pixel elevation were changed when the cell size of DEM was changed, the formation and even direction of reaches were completely deformed. This deformation might change river network properties such as the length of the longest flow path, which affects the time of concentration and model output significantly, especially in large watersheds.
For comparison, Figure 7 shows this deformation in the derived river network and also sub-basins within the watershed WSA in different cell sizes (SWAT outputs). This result   However, the ratio numbers which are indicated in Figure 8 may be altered in large watersheds. This result is in agreement with other studies (Munoth & Goyal ).

Runoff modeling
The SWAT outputs showed that the resolution of the input maps can intensively affect the amount of basin runoff. Although hydrological modeling was implemented for 30 years, the results showed that maximum variation has occurred when the maximum rainfall has occurred which was in agreement with other studies (Mishra & Singh ; Galoie et al. ). Due to this, the maximum variation is shown in Figure 9.
As can be seen in Figure 9, the following results can be obtained: 1. In all model outputs, when the DEM cell size increased, the runoff depth was decreased which is in agreement with Nazari-Sharabian et al. (), Al-Khafaji & Saeed () and Sharma & Tiwari (). This is because when DEM cell size is increased, the mean slope of the area is decreased as shown in Figure 5. Therefore, by decreasing the mean slope, the runoff is decreased.
Since the maximum mean slope reduction was occurred in strong slope watershed (WSA), the maximum reduction of runoff depth has also occurred in this watershed ( Figure 9, top, left).
2. Regardless of LULC resolution, the amount of runoff reduction for DEM cell sizes smaller than 100 m was less than 10%. This shows that for hydrological modeling and deriving physical characteristics of DEM using SWAT, pixel size less than 100 m is sufficient.  is increased; therefore, the runoff is also increased. However, when the LULC cell size for gentle slope DEM (WSC) is increased, the dominant LULC is changed so that the mean CN value is decreased, and therefore, the runoff is decreased.
According to the above results, it can be said that in comparison to DEM cell size changes, runoff values are more sensitive to the land-use resolution. This is because the most sensitive parameter in the SCS-CN relationship is CN which is directly evaluated based on land use, soil types and mean slope within a basin. Therefore, for successive hydrological modeling, although a finer DEM cell size is better, the resolution of land use is more important. As it can be seen in Figure 9, for a land-use cell size of 2,500 m, the runoff change rate for WSA was between À30 and À50%.
Also, it seems that for any hydrological modeling, a DEM with a cell size of up to 100 m is enough for most cases. As can be seen in Figure 9, for a given land-use map, the differences between percentage changes of runoff using DEM cell sizes as 12.5, 25, 50 and 100 are not significant.

CONCLUSIONS
In this paper, the impact of the quality of input DEM and LULC maps on hydrological modeling output using SWAT is investigated. To do this, DEM maps in three kinds of strong, moderate and gentle slope DEM were collected as 12.5, 25, 50, 100, 500 and 1,000 m and land-use maps as 250, 1,000 and 2,500 m. The output results showed that runoff values were changed when DEM cell size or landuse resolution was changed, but the sensitivity of runoff to land-use resolution was more than DEM cell size. In addition to the LULC resolution, the analysis showed that the distribution pattern of the land use across a watershed can change the amount of modeled runoff. Also, in all models, when the DEM cell size was increased, the runoff depth was decreased, but this reduction for the strong slope watershed was more than moderate and gentle slope watersheds. Also, the results showed that, for a given landuse map in a watershed, the differences between runoff values for various DEM cell sizes less than 100 m are not significant, so for hydrological modeling, a DEM with cell size up to 100 m seems to be sufficient.
For future works, it is suggested that the impact of DEM and LULC resolution on groundwater level and evapotranspiration are evaluated. Also, the effect of soil type map resolution on hydrological modeling is investigated.