VIC Setup & Cal-Val

VICtop

VIC Set up for Espirito Santo & Rio Doce. Model calibration-validation relies on historical data, as part of the process of developing confidence

    VIC requires information on topography, river networks, soil properties, and vegetation properties. It then requires meteorological forcings (inputs of daily precipitation, minimum and maximum temperature, and winds). The dataframe to implement these requirements utilized the information provided in the section on River Basins.

     Briefly, the 90-m SRTM (Shuttle Radar Topography Mission) conditioned by HydroSheds was acquired and masked to Espírito Santo. Since the local Brazilian topographic information was available only for Espirito Santo and did not include the upper Rio Doce basin, the SRTM data helped fill in the missing data.  The VIC domain, as derived from the topographic data, was set at a basin scale of 1/16th degrees (~ 6km) to capture the finer nuances in land cover and topography. Soil texture files were obtained from the Brazilian Corporation of Agricultural Research (EMBRAPA) product of soil maps (EMBRAPA, 2011). Soil percentage and profile was obtained from the RADAM-BRASIL Project (Cooper, 2005). Where soil data were unavailable for the parameterization of VIC, the corresponding parameters for each soil type was obtain from the United States Department of Agriculture (USDA) soil database. Landcover for the entire simulated region was obtain from the 2010 MODIS product.

     For the meteorological forcings,  a global half-degree gridded meteorological forcing dataset as described in Adam et al. (2006) was used to generate daily historical gridded forcings of temperature minima and maxima, precipitation, and wind speed at one-sixteenth degree latitude-longitude resolution from observed station data using methods described in Maurer et al. (2002). This dataset is from hereon referred to as the UW dataset. Daily values of precipitation and temperature were then disaggregated into 3-hourly time steps according to methods outlined in Nijssen et al. (2001) and Wang et al. (2009). Similar to Maurer (2002), other meteorological and radiation variables are calculated from established relationships, for example downward solar and longwave radiation and dew point were derived from the daily temperature and temperature range using methods described in Nijssen et al. (2001). The UW dataset is available for 1948 – 2008, however the period  1950 to 2006 was selected as the period of analysis with the first twenty years used to spin-up the model and specify the initial conditions.

     The model was spun up for 20 years (1950 – 1969) to achieve stable moisture conditions, and subsequently calibrated from 1970 – 1979 and validated from 1980 – 2006, which is the extent overlap between the available dataset and observed data. Model performance was evaluated using both the Nash-Sutcliffe Efficiency (NSe), and linear correlation coefficient, R. The NSe gives a view of how well the model is able to predict the flows while the correlation coefficient quantifies how well the model matches the observed data as defined by:

 

 

In both cases of statistical analyses measures, a value of ‘1’ indicates a good match between simulated and observed data.

     As expected, the largest basins produce the highest correlation coefficients due to heterogeneity (figure, below). In general, VIC captures streamflow seasonality very well (R is between 0.7 to 0.95). The gauge with the lowest NSe was Leopoldina, which is located at the mouth of the Rio Doce and is preceded by numerous small reservoirs and dams (Section on Dams). The low NSe and R can therefore be explained by the presence of these dams and the difficulty in modeling un-naturalized streamflow. The station with the second lowest NSe and R was Itabapoana, located at the foothills of a steep terrain towards the southern edge of the Espirito Santo. Streamflow from mountainous regions is particularly difficult to model since the steep topography makes the basin highly sensitive to distribution of precipitation and complexity in runoff events (Rahman et al., 2012). While the UW dataset used in this study accounts for orographic effects and is biased corrected against gauge data, many variables such as accuracy of streamflow records, temperature lapse rate and anthropogenic influence will affect the modeling capabilities.

VIC simulated (red) vs observed (black) flows, with Nash-Sutcliffe index of simulation efficiency (NSe) and correlation coefficient R