IRD-LOCEAN c/o SAEES, Center for Water Ressources Research, University of Kwazulu-Natal, Rabie Sanders Building, Box X01, 3209 Scottsville, South Africa
Chaplot, V., IRD-LOCEAN c/o SAEES, Center for Water Ressources Research, University of Kwazulu-Natal, Rabie Sanders Building, Box X01, 3209 Scottsville, South Africa
The need to precisely describe the characteristics of a landscape is well-known in mathematical modeling from different environmental disciplines. Because spatial input data, such as climate, relief and soil maps are costly to obtain, especially when large areas are considered, several research studies have investigated the extent to which the resolution of these can be reduced. Yet, a consensus has not been reached on the question of models' sensitivity to the whole range of spatial input data and for different environmental conditions. This issue was illustrated with the analysis of existing results from 41 watersheds from 30 research studies using the Soil and Water Assessment Tool (SWAT). Because these studies were not consistent in the type of spatial input data considered and the range of resolutions, an application of SWAT was performed in a flat 2612ha flat watershed of central Iowa (USA) where the sensitivity of runoff (R), NO3-N (N) and sediment (SED) yields was tested for changes in the resolution of all the required spatial input data (digital elevation model: DEM: 20-500m; n=12; number of rain gauge: NRAIN from 1 to 13; n=8; soil map: SOIL: 1/25,000-1/500,000; n=3) and in the number of watershed sub-divisions (NSW from 4 to 115; n=4). At the flat watershed, a Canonical Correlation Analysis with 67.4% of data variance explained by the two first variates, revealed that R and SED predictions were affected, mostly by NSW (r=0.95), followed by SOIL (r=0.18). N loads were the most sensitive to RAIN (r=0.76) and DEM (r=0.41), followed by SOIL (r=0.23) and NSW (r=-0.17). The Kolmogorov-Smirnov statistic (KS), that describes the significance of resolution changes for a considered spatial input data, showed that the model's sensitivity was greater for SSW below 261ha, for 30<DEM<100m and across the whole range of NRAIN. Finally, the analysis of watersheds with different sizes and environmental conditions revealed that the minimum spatial input data resolution needed, to achieve accurate modeling results can be predicted from watersheds' terrain declivity and mean annual precipitation. These results are expected to help modelers weight the level of investment to be made in generating spatial input data and in subdividing their watersheds as a function of both watersheds' environmental conditions and desired level of accuracy in the output variables. © 2013 .
Soils; Watersheds; Canonical correlation analysis; Global assessment; Kolmogorov-Smirnov statistics; Mean annual precipitation; Model parameterization; NPS pollution; OAT; Soil and water assessment tool; Input output programs; annual variation; canonical analysis; digital elevation model; environmental assessment; environmental conditions; erosion; hydrological modeling; nonpoint source pollution; resolution; sediment yield; spatial data; watershed