In: Wagner W„ Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
contain fluxes of surface runoff, évapotranspiration, baseflow, 
soil moisture etc. produced at that location. In order to simulate 
streamflow at an outlet, routing of runoff component was done 
using a routing model developed by Lohmann et. al (1998). 
Routing was done for 6 sub-basins namely Mundali (main 
outlet), Kantamal, Andhiyarkore, Simga, Baminidihi and 
Sundergarh. Daily and monthly streamflows in Cusec and mm 
for each outlet location was obtained as the output. 
Calibration of a hydrological model is an iterative process 
which involves changing the values of sensitive model 
parameters to obtain best possible match between the observed 
and simulated values. In general, before conducting numerical 
simulations, six model parameters of the VIC-2L model need to 
be calibrated because they cannot be determined well based on 
the available soil information (Yuan, 2004). These six model 
parameters are the depths of the upper and lower soil layers (di, 
i = 2, 3); the exponent (B) of the VIC-2L soil moisture capacity 
curve, which describes the spatial variability of the soil moisture 
capacity; and the three subsurface flow parameters (i.e., Dm, 
Ds, and JFs, where Dm is the maximum velocity of base flow, 
Ds is the fraction of Dm, and Ws is the fraction of maximum 
soil moisture). Three criteria were selected for model 
calibration: (i) Relative error (Er in percent), (ii) The Nash- 
Sutcliffe coefficient (Ce) (Nash and Sutcliffe, 1970), and (iii) 
Coefficient of Determination, (R 2 ). 
4. RESULTS AND DISCUSSION 
This study attempts to model the hydrology of Mahanadi river 
basin and assess landcover change impacts on streamflows at 
various locations along the river in the basin. For this purpose, 
mapping of landuse/ landcover was carried out in detail to 
represent the present and historical landcover conditions and 
changes that have taken place over whole of the basin in a span 
of three decades. Analysing landuse changes from 1972 to 
2003, it may be concluded that the total forest cover has 
declined by 5.71% of the total area of the basin. A reduction in 
barren land (0.64%) is followed by increase in areas of surface 
water bodies (0.47%), built up land (0.22%), river bed (0.11%) 
and most prominently agriculture (5.55%). This implies that the 
total forest cover and barren land has declined at the expense of 
increase in water body, river bed, agriculture and built up land 
in a span of 30 years.The simulation results obtained while 
calibrating and validating the VIC land surface hydrologic 
model were analysed and simulated streamflows were compared 
with the observed discharge at outlet station to look for the 
model efficiency in representing hydrological conditions 
accurately. 
4.1 Hydrological Modelling using VIC land surface model 
4.1.1 Pre-calibration simulation:The vegetation, soil, and 
forcing (meteorological) data as described were applied to the 
VIC-2L model to simulate évapotranspiration, runoff, and soil 
moisture at each grid over the Mahanadi River basin for year 
2003. To compare the VIC-2L model simulated runoff with the 
observed streamflow, the simulated runoff is routed through the 
river network using a simple routing model at the outlet 
Mundali as suggested by Lohmann et al, (1998). The routed 
daily and monthly runoff at these stations was compared with 
the daily and monthly observed streamflows, respectively as 
shown in Fig. 3. The R 2 showing agreement between the trends 
of simulated and observed streamflow records were found to be 
as 0.747, prior to calibration. 
Fig 3. Pre-calibration Comparison b/w Observed & Simulated 
daily discharges 
4.1.2 Model Calibration and validation: 
Since streamflow can be measured with relatively high accuracy 
compared with other water fluxes in the watershed it is mostly 
used to calibrate model parameters. In general, before 
conducting numerical simulations, the six model parameters of 
the VIC-2L model were calibrated and assigned values as: B = 
2.0, Dm = 15.0, Ds = 0.02, fVs = 0.8 and di = 0.5 and 2.0 m for i 
= 2 and 3. The velocity parameter was also adjusted (increased 
to 2.3 m/s) since the simulated runoff was coming delayed. The 
stream discharge at Mundali outlet for a period of 6 months was 
considered as the reference for calibration. 
Post calibration comparison of observed and simulated 
hydrograph at Mundali is shown in Fig 4. A good agreement 
between the observed and simulated values was found with an 
R 2 value of 0.836, Ce of 0.821 and Er of -8. 49 %. It can be seen 
that low flow simulations were overestimated and an 
underestimation was found during high flows. 
VIC is a model primarily designed to assess and evaluate long 
term climate and landcover changes on basin hydrology. It 
therefore essentially ignores the effect due to human induced 
activities. VIC simulates naturalized flows without considering 
any effect of reservoirs, dams or any other structural 
intervention. The Mahanadi basin contains several storage 
reservoirs and diversion structures and the observed 
streamflows are thus bias and are not really appropriate for the 
purpose of calibration. This may be a reason of disagreement 
between observed and simulated discharge. During low flows, 
reservoirs come into play and store most of the river waters 
whereas during high flows a reservoir has to throw out all 
waters coming into it once filled. This may be the possible 
reason of overestimation during low flows and underestimation 
during high flows. 
Better simulation results were obtained for monthly time-step 
when compared with daily and good agreement at Mundali was 
found. Comparisons of observed versus simulated hydrographs 
during model validation (daily) are shown in Fig. 4. Monthly 
comparisons were found good (Fig. 5) with an R 2 value of 
0.920, Ce of 0.890 and Er of -8.70%. The VIC model simulated 
runoff compares well with the daily observed streamflow in 
general, but significant overestimations of the streamflows are 
evident. This may be because of erratic spatial distribution of 
precipitation. Streamflows are most sensitive to vegetation and 
forcing input, thus near perfect simulations require accurate 
estimation of these parameters. In the present simulation, 
precipitation information has spatial resolution of 1 degree 
which is coarser, high resolution is therefore expected to 
improve simulation.