Estimation of Hydraulic Modeling Parameters of Poorly Gauged Basins Using Remote Sensing and GIS

H. GONCA COSKUN Department of Geomatic Engineering Istanbul Technical University 34469 Maslak, Istanbul TURKEY gonca@itu.edu.tr SEZEL KARAYUSUFOGLU Department of Satellite Communication and Remote Sensing Istanbul Technical University 34469 Maslak, Istanbul TURKEY sezel.karayusufoglu@gmail.com EBRU ERIS Department of Civil Engineering Istanbul Technical University 34469 Maslak, Istanbul TURKEY
eriseb@itu.edu.tr

UGUR ALGANCI Center for Sattelite Communications and Remote Sensing Istanbul Technical University 34469 Maslak, Istanbul TURKEY ugur@cscrs.itu.edu.tr
Abstract: - According to the increasing population, pollution of the natural reserves, global climate changing effects, distractions due to disasters make it indispensable to explore and research the Earth faster and cheaper. In comparison to conventional methods, integration of Remote Sensing (RS) and Geographical Information Systems (GIS) offer faster and economical solutions. Recently, for the hydrology and water resource studies and researches, RS and GIS are being used. Eastern Black Sea region where the study area is located in is a mountainous area. This type of places is hard or impossible to reach to acquire data by terrestrial measurement. However, by using integration of RS and GIS even those kinds of basins can be modeled. In small and poorly gauged basins by using different kinds of geostatistical methods, isohyets can be generated. By Digital Elevation Model (DEM), topographic features of the area can be obtained. Besides, by applying image classification to multispectral satellite data, current land use can be determined. Integrating all these data, basin area can be modeled assisted by GIS. In this study, for the classification of the study area IRS P6 multispectral satellite data with 5.8 m spatial resolution are used and to derive the Digital Elevation Model IRS P5 stereo satellite data with 2.5 m spatial resolution is used. The basin characteristics are mathematically modeled. Using DEM, which is generated from satellite data, isohyets map which is created by using one of the geostatistical methods Kriging, and basin characteristics, mean annual flows are estimated by the help of multiple regression analysis. Key-Words: - Hydrologic Modeling, Kriging, Remote Sensing, GIS, Classification, Image Processing.

1 Introduction
The more the technology improves the more the energy requirement for the daily life needs grow. For the developed countries the energy requirement reaches to the top level meanwhile the emerging countries increases day by day. It is important for one of the emerging countries, Turkey, to produce its energy provided by national natural resources, independent of other countries, and assure it sustainable. Therefore, in the sense of producing energy without the need for raw materials, providing an efficient use of renewable sources such as water, wind and sun takes importance. Turkeys rough geographical characteristics such as being generally mountainous and the considerable drop heights of the rivers make it substantial to determine the hydro-electric potential. It is getting indispensable to explore and research the Earth in the fastest and the cheapest way due to the growing population, violated natural reserves, effects of global climate changing, disasters causing distractions. Admitted, efficient and useful way to detect, manage and analyse earth surface is the integration of Remote Sensing (RS) and Geographic Information System (GIS). In comparison to the conventional methods integrated work of Remote Sensing (RS) and Geographical Information Systems (GIS) offer faster and economical solutions. Recently, for the hydrology and water resource studies and researches, RS and GIS are being used. Furthermore with the power of the recent advances in computer science and technology scientists, engineers and managers employ the two effective tools, the RS and the GIS, to obtain the needed parameters for the prediction of the stream flow in modelling. The stream flow estimation plays significant role for the water resources projects such as dam and hydroelectric power plant (HPP) constructions. Stream flow values take an important place at both determination of the hydroelectric potential and also the dam and HPP dimensions. For poorly gauged basins stream flow prediction is a difficult and time consuming process. Conventionally hydrological and meteorological data in addition to the physical basin parameters are necessary to predict stream flow. Briefly, the flow is the basic and the most considerable factor in the cost determination of such kind of expensive structures. In combination with appropriate rainfall-flow models, the RS and GIS provide ideal tools for the estimation of direct runoff volume, peak flows and hydrographs [1; 2; 3]. Generally, RS data provide a source of input parameters for the models in stream flow estimation. Thematic information on land use, soil, vegetation, drainage, etc., and topographic parameters (area, elevation and slope) which combined with conventionally measured climatic parameters (rainfall, temperature, etc.), constitute the necessary input data for

the rainfall-flow models are provided by satellite data. Using high resolution satellite imagery, Digital Elevation Model (DEM) and land cover maps can be obtained. Information can be provided in a quick, accurate and reliable way compared to classical terrestrial survey methods. In GIS media, RS data and other information, can be stored as geographically referenced [4; 5]. DEM is produced using stereo satellite images and digital topographic maps and used for determination of the elevation, slope, and aspect information of the basin. Additionally, processing the DEM data in a specific software module, area of each subbasins and synthetic drainage network data can be derived. In GIS environment the whole information can be combined with hydro-meteorological data in order to estimate the flow [6; 7; 8]. Integrated RS and GIS studies have been used in water resources assessment in the literature and these studies provide more practical and less time consuming results [9; 10; 11]. In this study, DEM of a poorly gauged stream basin is generated and basin characteristics are obtained by using RS and GIS. By means of these data, a multiple linear regression model is performed to estimate flow values In this manner, a multi-disciplinary study consisting RS, GIS methods and hydraulic academicians performances has been done.

2 Problem Formulation
The study area, Solakli Basin is placed in the city of Trabzon at the Eastern Black Sea Region and located at 41o 00' North latitude and 39o 43' East longitude (Fig. 1). It covers an area of 767.25 km2 . From the RS analysis, it is found that the basin area consists of 22.6% coniferous forest area, 20.4% broadleaf forest area, 1.2% water covered surface, 1.4% settlement area, 12.2% rocky area, 15.8% plain area, 0.5% roads, 13.9% grassland, 4.6% tea plant covered area and hazel-nut area covers 7.4% of the total basin area. Haldizen and Soganli Mountains rise in the south part of the basin where the highest point Demirkapi Hill reaches to 3376 m elevation. Solakli stream is the main stream of the basin. Especially during the spring, both snow covering starts to melt and the severe convectional activities in the atmosphere causes heavy rainfall lead floods and landslides. Dense forest cover also a considerable factor results in difficulties for the conventional terrestrial measurements.

rainfall values using one of the mostly used geostatistical methods, Kriging. Thus the mean areal rainfall values are obtained in the form of isohyteal map. Consequently, area, mean basin elevation values are calculated for the sub basins of Solakli Basin in GIS media. All these information is then used to derive a regression equation to estimate stream flow for the entire Solakli Basin.

Fig.1 Study area Meteorological and hyrological measurements are limited with three rainfall-stream flow station pairs. The data are obtained from these gauging station pairs: Uzungol-Haldizen, Koknar-Ogene and CaykaraUlucami. Table 1 shows the properties of the rainfall and stream flow gauging stations. Mean annual rainfall and stream flow data is used through the regression analyses. Table 1 Rainfall and stream flow stations used in this study.
Station No 1962 22-016 1801 22-07 22-57 22-52 Station Name Uzungol Koknar Caykara Haldizen Ogene Ulucami Station Type Rainfall Rainfall Rainfall Streamflow Streamflow Streamflow Coordinates X (m) 609071 602839 604880 609060 602599 606083 Y (m) 4497483 4502007 4515760 4497459 4502760 4513714 Elevation (m) 1110 1218 264 1116 675 260

Fig.2 Rainfall gauging stations used to generate isohyetal map and mean areal rainfall.

3.1 Determination of topographic parameters

DEM is produced in order to determine the topographic parameters of Solakli Basin. As it is indicated before Stereo Cartosat - 1 (IRS P5) satellite data with 2.5 meter spatial resolution (Table 2) is used for this procedure. Table 2 Characteristics of Stereo Cartosat - 1 (IRS P5) satellite
PAN-Fore Tilt Along Track +26 deg 2.5 m 30 km 10 bit 500-850nm 1945 mm 1 * 12000 PAN-Aft -5 deg 2.5 m 27 km 10 bit 500-850nm 1945 mm 1 * 12000

3 Problem Solution
Basins hydrological modelling is done trough three different stages consist of obtaining topographic parameters such as basin area and the mean basin elevation assisted by GIS, then mean annual areal rainfall values generated again using GIS and lastly mathematical model of the basin generated through the regression equation and so the mean annual flow values are determined. Firstly, Stereoscopic IRS P5 satellite data with 2.5 m spatial resolution is used to generate 1:10 000 scaled Digital Elevation Model (DEM) of Solakli Basin to determine the flow accumulation and direction of stream using both RS techniques and GIS. Secondly, because the rainfall gauges are limited in the basin, additional neighbouring gauges (Fig. 2) of the Solakli basin are used in order to generate mean areal

Spatial Resolution Swath-width Radiometric Resolution, Quantisation Spectral Coverage Focal Length CCD Arrays (no. of arrays * no. of elements) CCD Size Integration Time

7mx7m 0.336 ms

7mx7 m 0.336 ms

Automated stereo correlation has become a standard method for generating DEMs from digital stereo images. Automatic image matching is the basis of the stereo

correlation. Although approaches may vary according to the software employed, the procedures normally include the collection of Ground Control Points (GCPs), determination of parallax values on a per pixel or per DEM post basis using automatic image matching techniques and, finally, post-processing to remove the anomalies from the DEMs [12]. Using dual frequency GPS instruments, 65 GCPs based on 11 points from National Network Stations are collected in order to generate DEM. Accuracy of these points reaches 10 cm (Fig. 3a) after processing the collected data. Using ground control points measured with static GPS equipment and the tie points collected from the stereo pairs an absolute DEM with 5 m cell size is generated (Fig. 3b, c). Both in planimetric position and elevation, the expected accuracy of the DEM is approximated to be 4 m.

(a)

(b)

(c)

Fig.4 (a)Flow direction map; (b) Flow accumulation map; (c) Synthetic drainage network of the basin.

3.2 Determination of mean areal rainfall

In the literature several interpolation techniques for sparse data have been discussed. On the other hand, geostatistical techniques are progressively more preferred because they allow one to capitalize on the spatial structure/correlation between neighbouring observations to predict attribute values and to quantify prediction uncertainty at unsampled locations [14]. In this study, to create the isohyetal map using one of the mostly preferred geostatistical methods, Kriging. Therefore the mean areal rainfall values for the related sub basins are estimated. In brief, Kriging is based on the estimated spatial covariance structure of the observed data and implemented as two- or three-dimensional interpolation technique. There are several types of Kriging such as ordinary, simple, universal etc. According to the earlier studies ordinary Kriging is selected to assemble the isohyetal map and to estimate the mean areal rainfall values [15; 16]. Due to the poor rainfall gauging of the study area, mean areal rainfall is determined using total of 11 point observations of both neighbouring rainfall gauges and the ones located in the basin area (Fig.2). The generated isohyetal map for Solakli Basin is shown in Fig. 5.

(a)

(b)

(c)

Fig.3 (a) Static stations used in GPS measurements; (b); (c) Generated DEM DEMs should be cleared of pits or ponds, where water gathers when drainage network is extracted, before being used in the hydrological modelling [13]. Using an algorithm known as sink filling a sign of errors, pits, can be removed from DEMs. After removing the errors by filling sinks algorithm, by calculating the steepest slope and by encoding into each cell for possible eight flow directions toward the surrounding cells a flow direction map is produced (Fig. 4a) [5]. Then flow direction is used to produce the flow accumulation map by addressing each cell of the DEM, counting how many upstream cells contribute to flow through the given cell (Fig. 4b). After all, both flow accumulation and flow direction maps are used to create synthetic drainage network of the basin (Fig. 4c).

In this study for Solakli Basin, in order to estimate the flow a multi regression analysis is performed. Following the study of Vogel et al. [17], both linear and logarithmic regressions shown in Table 3 are employed considering area, mean basin elevation and point rainfall observations as the independent variables. In Solakli Basin there are only three rainfall-stream flow station pairs to be correlated. The locations of rainfall stations are close to flow stations. Determination coefficient (R2), the mean absolute error (MAE) and the mean relative error (MRE) are the performance evaluation criteria. The determination coefficient is a measure of the goodness of fit provided by the estimated regression equation and represents the proportion of the explained variance. The more closer the R2 values to one the much satisfactory the model performance is. MAE and MRE are calculated as:
MAE = 1 N

xi

Fig.5 Isohyetal map for Solakli Basin. To calculate areal rainfall of sub basins, the area in between the two isohyets is multiplied with the average value of them (Figure 6).

(2)

MRE = 100

y x 1 ix i N i

(3)

where; xi and yi represent observed and estimated values, respectively. Using point rainfall observations and basin characteristics, multiple linear and logarithmic regression equations are shown in Table 3. As seen from the table, R2 value of logarithmic equation is higher, MAE and MRE values are lower than that of linear one. Table 3 Linear and logarithmic regression equations.
Models Linear R2 0.900 MAE (m3/s) 1.23 MRE (%) 16.0 Regression Relations for Flow (Q) Q= -7.078 + (0.0126 * P) + (0.0251 * A) (0.00207 * H) Q= (10-11.459)* (P1.337)* (A1.203)* (H1.629 )

Fig.6 Calculating areal rainfall of sub basins.

Logarithmic 0.937 0.95 11.7

In order to calculate mean areal rainfall the equation below is used in this study:
(P +P) (P+P) (P +P) (P +P) (P +P) 0 1 A+ 1 2 A + 2 3 A + 3 6 A + 6 7 A 1 2 3 4 5 2 2 2 2 2 Pavg = A +A +A +A +A 1 2 3 4 5

(1)

3.3 Regression analysis for stream flow

Relation between mean annual stream flow and the geomorphic characteristics, land use, basin characteristics for region scale can be found simply using GIS [17]. In the literature there are several studies available especially for the poorly gauged basins [17; 18; 19; 20].

To validate the logarithmic regression equation, in addition to basin characteristics, mean areal rainfall values of the sub basins are used. As expected, mean areal rainfall of an ungauged point can be obtained from a isohyetal map, thus, flow value of this point can be estimated from the regression equation. Mean areal rainfall values of sub basins are derived in GIS environment (Fig. 7).

Fig.7 Deriving of mean areal rainfall values in GIS media. Using this time, mean areal rainfall and basin characteristics, the logarithmic equations are validated. The results of validation stage is shown in Table 4. Table 4 Results of regression Stream Flow Station/Basin Haldizen Alcakkoru Ulucami Measured flow (m3/s) 4.21 5.21 15.28 Estimated flow (m3/s) 4.27 5.47 14.07 Relative Error (%) 1.27 4.87 -7.89

4 Conclusion
In this study, Remote Sensing and GIS co work provided a macro-scale hydrology model which is tested and verified by using all three sub basins runoff gauging stations. Relation between flow, rainfall, and topographic parameters obtained to create a multiple linear regression equation. This regression model gives the satisfactory results even mean areal rainfall values are used instead of point rainfall values. It means that, for an ungauged point, which has also no rainfall measurement, mean areal rainfall value can be derived from the isohyteal map and using this data, flow can be estimated. The employment of the GIS and RS techniques for the basin flow determination is a quite useful way especially for poorly observed basins. As a result, for practical purposes particularly, this model can be used to estimate flows for poorly gauged basins.

References: [1]Bellal, M., Sillen, X., Zeck, Y. (1996) Coupling GIS with a distributed hydrological model for studying the effect of various urban planning options on rainfallrunoff relationship in urbanised basins. In: K. Kovar, and

H. P. Nachtnebel, (ed) Application of Geographic Information Systems in hydrology and water resources management: International Association of Hydrological Sciences, Series of Proceedings and Reports. 235: 99106. [2]Demayo, A. and Steel, A. (1996) Data handling and presentation. In: D. Chapman, (ed). Water quality assessments, a guide to the use of biota, sediments and water in environmental monitoring: London, United Nations Educational, Scientific and Cultural Organization, World Health Organization, United Nations Environment Programme, 2nd edition. 10: 511612. [3]Miloradov, M., Marjanovic, P. (1991) Geographic Information System in environmentally sound river basin development. 3rd Rhine-Danube Workshop. 7-8 October Delft the Netherlands Technische Universiteit Delft. [4]Dudhani, S., Sinha, A. K., Inamdar, S. S. (2006) Assessment of small hydropower potential using Remote Sensing data for sustainable development in India. Energy Policy, 34: 3195-3205. [5]Hammouri, N., El-Naqa, A. (2007) Hydrological modeling of ungauged wadis in arid environments using GIS: A case study of Wadi Madoneh in Jordan. Revista, Mexicana de Ciencias Geolgicas 24(2): 185-193. [6]Cheng, Q., Ko, C., Yuan, Y., Ge, Y., Zhang, S. (2006) GIS modelling for predicting river runoff volume in ungauged drainages in Greater Toronto Area, Canada. Computers and Geosciences, 32: 1108-1119. [7]Kumar, A., Singhal, M. K., (1999) Hydropower assessment for small ungauged catchments in Himalayan Region using GIS techniques. Workshop on GIS Applications in Power Sector Map India. [8]Saraf, A. K., Kumar, A., (2006) Spatial technologies in Himalayan small hydropower development. Himalayan Small Hydropower Summit. October 12-13. Dehradun. [9]Coskun, H.G., Alganci, U., Usta, G., (2008) Analysis of Land Use Change and Urbanization in the Kucukcekmece Water Basin (Istanbul, Turkey) with Temporal Satellite Data using Remote Sensing and GIS, Sensors, 8 (11): 7213-7223. [10]Coskun, H. G., (2009) Remote Sensing and GIS Techniques for Temporal Evaluation of Environmental Impacts on Major Drinking Water Dam and Basin of Metropolis Istanbul, Fresenius Environmental Bulletin, 18(3): 261 - 269, Parlar Scientific Publications [11] Coskun, H. G., Alparslan, E., (2009) Environmental Modelling of Omerli Catchment Area in Istanbul, Turkey Using Remote Sensing and GIS Techniques, Environmental Monitoring and Assessment, 153(14):323-332. [12]Kok, A. L., Blais, J. A. R., Rangayyan, R. M. (1987) Filtering of digitally correlated gestalt elevation data.

Photogrammetric Engineering and Remote Sensing 53 (5): 535-538. [13]Ashe, R., (2003) Investigating runoff source area for an irrigation drain system using arc hydro tools, in Proceedings of the Twenty-Third Annual ESRI User Conference. San Diego. [14]Goovaerts, P. (2000). Geostatistical approaches for incorporating elevation into the spatial interpolation of rainfall. Journal of Hydrology 228(1-2): 113-129. [15]Atkinson, P. M. and Lloyd, C. D. (1998) Mapping precipitation in Switzerland with ordinary and indicator Kriging. Journal of Geographic Information and Decision Analysis, 2(2): 65-76. [16]Nour, M. H., Smit D. W., Gamal El-Din M. (2006) Geostatistical mapping of precipitation: implications for rain gauge network design. Water Science & Technology, 53 (10): 101-110. [17]Vogel, R. M., Wilson, I., Daly, C. (1999) Regional regression models of annual stream flow for the United States. ASCE J. Irrigation and Drainage Engineering, 125(3): 148-157. [18]Chiang, S. M., Tsay, T. K., Nix, S. J. (2002) Hydrologic regionalization of watersheds II: Applications. J. Water Resour. Plan. Manage 128: 12-20. [19]Sanborn S. C., Bledsoe B. P. (2006) Predicting streamflow regime metrics for ungauged streams in Colorado, Washington, and Oregon. Journal of Hydrology 325: 241-261. [20]Thomas, D. M., Benson, M. A. (1970) Generalization of stream flow characteristics from drainage-basin characteristics. USGS Water-Supply Paper, 1975, 55 pp.