IMPACT OF LAND USE LAND COVER CHANGE ON RUN OFF GENERATION IN TUNGABHADRA RIVER BASIN

Streamflow can be affected by a number of aspects related to land use and can vary promptly as those factors change. Urbanization, deforestation, mining, agricultural practices and economic growth are some of the factors related to these land use changes which alter the stream flow. In the present study, the impact of land use land cover change (LULC) on stream flow is studied by using SWAT model for Tungabhadra river basin, located in the state of Karnataka, India. Tungabhadra river originates in the Western Ghats of Karnataka and flows towards north-east and joins the river Krishna. The land use maps of 1993, 2003 and 2018 are used for assessing the stream flow changes with respect to LULC. Calibration and validation of the model for streamflow was carried out using the SUFI-2 algorithm in SWAT-CUP for the years 1983-1993 and 1994-2000 respectively. Statistical parameters namely Coefficient of Determination (R) & Nash–Sutcliffe (N-S) were used to assess the efficiency and performance of the SWAT model. It was found that the observed and simulated streamflow values are closely matching, which in turn projects that the model results are acceptable. The calibrated model was used for simulation of future dynamic land use scenario to assess the impact on streamflow. The results can be used for conservation of water and soil management.


INTRODUCTION
Land Use Land Cover change is a crucial environmental change which has several impacts on human livelihoods.Management of earth's natural resources remains a critical environmental challenge that society must address because misuse of available resources may lead to severe threat causing scarcity of water resources.Natural life is mainly supported by major resources i.e. water and soil, which play crucial roles in the natural ecosystems.Freshwater which moves from upstream to downstream is mainly supplied by the watersheds.The water quality reaching the downstream is being degraded due to the changes that are occurring in land use and land cover.Changes in land use and land cover mainly drive the changes in watershed hydrology.Deforestation, conversion of vegetation lands to agriculture may increase the economic development but it also affects the environmental status of the society.* Stream flows are sensitive to land use change i.e. minor change in land use causes major changes to stream flows.Numerous studies have been conducted to investigate the impact of LULC change on stream flows ranging from small watersheds to large river basins which ended up exhibiting the causes for stream flow changes is due to conversion of forest land to agricultural lands.Increase in settlements, deforestation, expansion of agricultural area and intensive grazing yields high runoff and sediment yield.These changes enlarge the quantity, velocity and intensity of runoff.Considering this, Loi (2010) used two land use scenarios for assessing the factors that contribute to the change in runoff * Corresponding author -venkateshkolluru95@gmail.com for Dong Nai watershed, Vietnam and Shrestha et al. (2015) used monthly stream flows and sediment yield data for assessing runoff and sediment yield from Da river basin in Northwest of Vietnam.Both of them applied SWAT model for simulating daily, monthly runoff and sediment yield and concluded that there is an increase in runoff and sediment yield when the land had been converted from forest to agriculture.The specific objective of the present study is to analyse the impact of LULC on stream flows from the past three decades which is important to understand the economic and environmental changes in the study area.

STUDY AREA
Tungabhadra River is a major tributary of river Krishna which originates from the confluence of two rivers Tunga and Bhadra which were started at Gangamoola of Western Ghats region of Karnataka at an altitude of 1198 m above MSL flowing towards eastern side and meeting at Holehonnur at an altitude of 610 m in Shimoga.The Tungabhadra river basin has a total catchment area of about 69552 km 2 which includes both upper and lower Tungabhadra river basins but the current study area lies between longitudes 74°00′00″-76°30′00″E and latitudes 13°00′00″-15°30′00″N, with a catchment area of 15393.039km 2 up to the Haralahalli gauge station, which is at the outlet of the catchment as shown in Figure 1.The average annual temperature of the region is around 26˚ C with mean maximum monthly temperature varying from 26.3˚C to 35.5˚C and mean minimum monthly temperature varying from 13.8˚C to 22.3˚C.The average annual rainfall recorded over the region is about 1200 mm (Lo Porto et al. 2010).

Topography:
Topography is mainly represented in the form of Digital Elevation Model (DEM) as shown in fig 2. Shuttle Radar Topography Mission (SRTM) DEM which represents the topography of the study area with a spatial resolution of 30m is downloaded from USGS Earth Explorer.DEM gives elevation values for each pixel and it is used for delineating the watershed in SWAT model.SRTM DEM obtained from USGS Earth Explorer has some voids which should be filled for processing into SWAT.In order to fill these voids ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) DEM was used which has the same spatial resolution of 30m.Raster calculator in ArcGIS is used for filling these voids by overlaying ASTER DEM and SRTM DEM.Slope map was generated from DEM depending upon the steepness of the surface.The study area is divided into 5 slope classes as shown in figure 3 , viz. 0-10, 10-20, 20-30, 30-40 and >40.

Land Use Land Cover Map:
Three land use land cover datasets are created for the years 1993, 2003 and 2018 by downloading, layer stacking and mosaicking Landsat 5, 7 and 8 satellite images from USGS Earth Explorer which are free from cloud cover (Details are shown in Table 9).The mosaicked images are further processed for land use land cover classification using ERDAS Imagine software using Maximum Likelihood algorithm.The land use land cover datasets are divided into 7 classes namely agriculture (AGRL-All varieties of crops and plantations are considered as agriculture), barren (BARR-Rocks, Hills, Wastelands), built up (URBN), cultivated land (RNGE-Agricultural land which was left unseeded for some years), forest (FRST), mining (SWRN) and water body (WATR).The agricultural land will intercept at least some part of the rain whereas cultivated lands were vacant which contributes more runoff as no interception occurs when compared to agricultural lands.10 where barren land was decreased in 2018 when compared to 1993 and some part of the barren land was converted to mining area and some percentage to urban and agricultural lands.Three LULC datasets of different years are used in this study to identify whether the changes in land use affect the quantity of stream flow.The average annual runoff values during the calibration period were 363.44, 361.39, 350.89 mm and 435.76, 433.95, 424.46 during the validation period for the years 1993, 2003 and 2018.From these results it is observed that even though there are larger changes in percentage occupancies of land use, there is less influence on runoff.
Figure 10 Percentage changes of LULC from 1993 to 2018 The water area was decreased from 29634 ha to 14814 ha from 1993 to 2003 and was increased to 22259 ha in 2018.
The overall accuracy was found to be 84.53%,86.33% and 85.94% for the years 1993, 2003 and 2018.Kappa statistics was determined which was found satisfactory with a result of 0.774, 0.744 and 0.789 for the 3 years respectively.

Calibration and Validation:
Sensitivity analysis was performed prior to the calibration of the model.Out of 18 parameters obtained from the previous literature, 10 parameters were found to be sensitive.20, it is evident that the peak values during the Observed period were much larger than the Simulation period for both daily and monthly phases and can also be observed in figures 18 and 22.The SWAT model was unable to match the peaks since there is a larger deviation between the observed and simulated values in the calibration phase.Table 21 exhibits that the peaks are closely matching and the deviation between the observed and simulated values are also less.The observed values have more standard deviation than the simulated values in the validation phase, due to which the N-S values during the validation phase in all the 3 years was less when compared to the calibration phase.
The overall results exhibited good performance in simulating runoff using SWAT for Tungabhadra river basin during the 3 time periods.It is observed that, the change in LULC in 3 time periods did not show much difference between the simulated streamflow values.The accuracy can further be improved by implementing a soil map with better classification and high-resolution LULC maps.

CONCLUSIONS
The following conclusions are drawn from this study based on the SWAT model.
Based on LULC classification, the predominant classes are barren and cultivated land.Both the classes were decreased in 2018 when compared to 1993 which was accompanied by the increase in agriculture and urban area.
So many studies (Loi et al. 2010, Ngo et al. 2015) concluded that the conversion of forest to agricultural land increases the runoff.In the present study, even though there are significant changes in the LULC for the 3 decades, especially the decrease of forest and increase of agricultural land during the years 2003 and 2018, there was no significant change in the average annual runoff during the calibration and validation phases for the years Based on sensitivity analysis CH_N2, GW_DELAY, GWQMN, ALPHA_BF, CH_K2, SOL_AWC and ALPHA_BNK, ESCO, CN2, SOL_K were found to be sensitive for SWAT model employed in Tungabhadra river basin.
For daily simulations the results are good (R 2 = 0.727, 0.729, 0.73 during calibration phase and R 2 = 0.753, 0.754, 0.75 during validation phase) for the years 1993, 2003 and 2018 At monthly time step the results are further improved for runoff (R 2 = 0.8, 0.804, 0.8 during calibration phase and R 2 = 0.852, 0.854, 0.85 during validation phase) for the 3 years respectively.
The statistical coefficients (R 2 and N.S) were proved effective which exhibits that the SWAT model is capable of simulating runoff in the study area accurately.

Figure
Figure 1.Study Area 3. DATA USED IN THE STUDY Soil and Water Assessment Tool (SWAT) model was deployed in the present study for the simulation of runoff for Tungabhadra river basin.SWAT requires raster files such as DEM, land use and slope maps and vector datasets such as outlet points, rainfall and temperature for the generation of runoff.All the input datasets must be projected to WGS 1984 World Mercator for loading them into SWAT.The input datasets are mainly categorized into 4 categories viz.Topography, Land use, Soil and Hydrometeorological datasets for simulating the stream flow processes.

Figure
Figure 4 Soil map

Figure 5 Figure 6 Figure 7
Figure 5 Land use land cover map of 1993

4. 2
SWAT-CUP:Sequential Uncertainty Fitting (SUFI-2) algorithm within SWAT-CUP (Abbaspour et al.) is used for calibration and validation.The streamflow records are obtained from India-WRIS website for Harlahalli gauging stations over a period of 21 years ranging from 1980 to 2000.The entire duration is divided into 3 years for the warm-up period, 12 years ranging from 1983 to 1994 for calibration and 6 years ranging from 1995 to 2000 for the validation period.Calibration and validation are mainly based on sensitive parameters in SWAT-CUP.Parameters are said to be sensitive if a small change in the parameter ranges causes a large change in the runoff.Sensitivity analysis is carried out to identify the parameters which are sensitive to a particular region.Sensitivity analysis is useful for decreasing the number of sensitive parameters if they found insignificant.The t-stat and p-value in the SUFI-2 algorithm is useful for finding the sensitivity of parameters.The p-value determines the significance of sensitivity based on the value ranging from 0 to 1.The value closer to zero is identified as the most sensitive parameter and vice-versa.Different statistical coefficients like the Coefficient of Determination (R 2 ) and Nash-Sutcliffe (N-S) are used for finding the accuracy of model performance.R 2 gives the correlation between observed and simulated values which ranges from 0 to 1 and N-S shows relative difference between observed and simulated values which ranges from -∞ to 1.

Figure 14
Figure 14 Plot showing simulated vs observed runoff at monthly time step for the year 2018(Calibration)

Figure 18
Figure 18 Line graph showing simulated and observed runoff vs time at daily time step for the year 2003 (calibration)

Table 8
Land use land cover statistics The land use, soil map and slope map are overlaid and threshold values are specified to divide into watershed into multiple sub-basins and HRU's.Once the overlaying was completed, Meteorological parameters were inserted and the setup of SWAT was done.
predicting various parameters related to water, sediment and agricultural chemical yields for all types of watersheds at daily, monthly and annual time steps(Arnold et al. 1995).The entire basin is divided into multiple Hydrological Response Units (HRU's) by SWAT which has unique land, soil and slope characteristics.Actual Evapotranspiration and potential transpiration are calculated based on Penmann-Monteith method.Surface runoff and peak runoff are estimated based on modified Soil Conservation Service (SCS-CN) method and the modified rational method.

Table 13
Statistical coefficient values for monthly runoff Scatter plots are plotted for observed against simulated runoff values which are shown in figures 14, 15, 16, 17.Line graphs are plotted for observed and simulated streamflows against time for the years 2003 and 2018 and are shown in Figs. 18, 19, 22, 23.

Table 20 &
21 gives the peak values and standard deviation values for both observed (Obs) and simulated (Sim) runoffs for the years 1993, 2003 and 2018 during the calibration and validation phases.From Table 1993, 2003 and 2018.