| GISdevelopment.net ---> Application ---> Natural Resource Management--> Overview |
Towards a framework for quantifying water resources in india
D. Shankar
e-mail: shankar@darya.nio.org
Vidya Kotamraju
e-mail:vidyakr@darya.nio.org
S. R. Shetye
Physical Oceanography Division, National Institute of Oceanography
Dona Paula, Goa 403004, India.
Abstract
While issues related to water attract considerable attention in India, very little quantitative information is available on her water budget. There are two reasons for this lacuna: the dearth of information on hydrological variables, and the absence of an easily accessible quantitative framework to put these variables in perspective. In this paper, we assemble a framework to address both issues. At its core is a hydrological routing model; the basic data needed for implementing the framework are a digital elevation model and data on precipitation and evapotranspiration. We demonstrate the viability of the framework by applying it to the hydrology of the Mandovi river in Goa. The model output mimics the observed discharge well.
Introduction
India, as one of the countries affected by both population explosion and the Green Revolution, hardly sees a day without some problem related to water figuring in the news. The problems are twofold (Agarwal and Narain, 1997, 1999; Agarwal and Chak, 1991; Agarwal et al., 1999; Rao, 1975). First, there is the problem of scarcity, resulting in drought or famine; second, there is the problem of surplus (mostly during the summer monsoon, when most of India receives almost 80% of the annual precipitation), resulting in floods. It is also evident that certain regions of the country, especially in the south and northwest, are more prone to drought, while other regions, especially in the east and northeast, are more prone to floods. With each disaster comes the clamour for some long-term solution. One solution that has been proposed is the transfer of water, through the medium of a network of canals, from river basins with a surplus (mostly the rivers in the north and northeast) to river basins with a deficiency (Radhakrishna, 2003; Rao, 1975) (mostly in the south and northwest). This idea received a major boost last year following one of the worst droughts of the century (Gadgil et al., 2002).
The financial implications of networking rivers over an area the size of India are enormous: a tentative estimate for this project is Rs. 560000 crores (Radhakrishna, 2003). Commissioning a project of such magnitude and then executing it successfully will require a vast database on hydrology and a framework that can use the data to quantify the water budget for each river basin. This apart, given the pressure that exists on the resources of fresh water, there is an urgent need to evolve better, sustainable strategies for managing the country's water resources. This, in turn, requires hard data on the basis of which decisions can be made: we need quantitative estimates of water budgets for the river basins at a resolution small enough for evolving strategies for, say, an average Indian district, yet large enough to make possible an estimate of water resources on the scale of the Indian subcontinent.
The estimates of water resources available in India today are often on too small a scale, like a small watershed, or are on too large a scale, like gross statistics based on the average precipitation over the entire country and the measurements made at a few stream-flow gauges on major rivers and in the neighbourhood of some dams (Nag and Kathpalia, 1975; Agarwal and Narain, 1999; Anon, 1988; Rao and Ramaseshan, 1985; Chaturvedi, 1985; Rao et al., 1971; Rao, 1975). For individual river basins, the tendency has been to adopt an empirical systems approach (Chaturvedi and Srivastava, 1981; Rao, 1975). What is lacking is an overview that provides a reliable quantitative estimate of the water resources of the country. There are two reasons for the absence of such estimates. First, there is a dearth of information on the basic hydrological variables of interest -- groundwater recharge rates, stream-flow, evapotranspiration, etc. Only precipitation data have been collected systematically over a long period; evapotranspiration data are scanty (Rao, 2001); stream-flow data are confined to a few rivers, and are often not easily accessible (The Ad Hoc Group on Global Water Data Sets, International Association of Hydrological Sciences, 2001; Radhakrishna, 2003). Second, a quantitative framework that can put these variables in perspective is often missing. Though several hydrological software packages are available, Indian academic institutions are reluctant to invest in them because these packages are usually expensive and are geared to handle situations in developed countries where the hydrological databases are much more advanced. They also rely too often on commercial GIS (Geographical Information Systems) packages, which too are expensive for Indian academic institutions. Hence, in view of the present state of continental-scale hydrology of the subcontinent, it seems that what is needed today is a quantitative framework that satisfies the following conditions.
The framework is described in detail in Coe (2000) and Shankar et al. (2003). Here, we demonstrate its viability by applying it to a river basin. Our choice for testing the framework is the Mandovi river (also called Mahadayi and Mhadei over some stretches), which has its source in the Sahyadri range and flows into the eastern Arabian Sea (Figure 1). Its catchment area is mostly in Goa, but also extends into the neighbouring states of Karnataka and Maharashtra. This choice is dictated by the following considerations. First, being one of the two rivers that flow past Panaji, the capital of Goa and the location of the National Institute of Oceanography, it is one of the best studied estuarine stretches in India, and is typical of the rivers along the west coast of India. Second, it is a small river, with a channel width less than 1 km over much of its 105 km length, making it a rigorous test of the DEM's accuracy: a DEM that works in this region should also work elsewhere in India. Third, discharge data are available for the Mandovi at one point, about 50 km from the mouth, making it possible to validate the framework. We conclude the paper by discussing the potential of this quantitative framework for quantifying the water resources of India.
Figure 1: The topography of the region, as seen in GLOBE data (after editing the DEM). The superimposed rivers were digitised from the maps of the Survey of India. The domain of the simulations described in this paper lies mostly in Goa. The Mandovi and the Zuari are the two main rivers of Goa. The Mandovi flows into the Aguada Bay near Panaji, the capital of Goa and the location of its main meteorological observatory. A stream-flow gauge is located on the Mandovi at Ganjem; Valpoi has the easternmost rain gauge in Goa, but data for 1959 from Kulem, farther east, are available for 1959 (Anon, 1964).
Application to the mandovi river in Goa
The Mandovi is one of the two major rivers located in Goa, the other being the Zuari (Figure 1). The two rivers are connected by a narrow channel, the Kumbarjua Canal (Shetye et al., 1995); many tributaries also feed into the two rivers. The Mandovi, the Zuari, the Kumbarjua Canal, and the tributaries form a network. The cross-sectional area of both rivers decreases rapidly upstream (Shetye et al., 1995). Tidal influence is felt in the Mandovi just a little downstream of Ganjem (Figure 1), approximately 50 km inland, during the dry season (Shetye et al., 1995). The estuarine stretches (where at least some salinity intrusion occurs during the dry season) receive freshwater primarily from rivers that originate in the Sahyadris (or Western Ghats), but there is additional inflow at several points along the main channels. A stream-flow gauge is located on the Mandovi at Ganjem, just upstream of the limit of tidal and salt incursion.
We use the GLOBE (GLOBE Task Team and others, eds., 1999) (Global Land One-km Base Elevation; see the GLOBE web site http://www.ngdc.noaa.gov/seg/topo/globe.shtml for more information) DEM, which has a
resolution of 30 arc seconds (~ 1 km). The topography of the region, as seen in the GLOBE data, is shown in Figure 1; the superimposed rivers were digitised from Survey of India maps and the good match shows that the GLOBE DEM is able to capture the watershed geometry fairly well. The elevation data are used to determine the local flow directions for each grid cell. Once the basin geometry is defined, HYDRA can be forced by prescribing the surface runoff and sub-surface drainage at each grid cell and integrating forward in time; for the GLOBE DEM used here, the model time step is 5 minutes.
We force the model with the precipitation and evapotranspiration data from the NCEP/NCAR Reanalyses (Kalnay et al., 1996) (Case 1; see Table 1 for a brief description of the simulation cases). Since the NCEP/NCAR model resolution is 2.5o, a single NCEP/NCAR grid cell encompasses the entire basin, yielding a spatially uniform forcing field; its temporal variation is shown in Figure 2. The runoff required to force the model is the difference between precipitation and evapotranspiration (runoff is zero when the latter is greater), and the partition between surface runoff and sub-surface drainage is assumed to be the same as in Coe (Coe, 2000): 30% surface runoff and 70% sub-surface drainage. The result is that there is almost no discharge into the sea, most of the local runoff tending to accumulate in lakes 30-50 m deep! Since no such ``potential water areas'' (PWAs: regions where water can pile up, such as lakes or wetlands, and slow the flow through the river) are observed in the Mandovi basin upstream of Ganjem, this is due to the failure of the DEM to channel the runoff into and through the river.
Figure 2: Climatology of observed precipitation and evapotranspiration (mm/day) and discharge (m3s-1) in the Mandovi basin. The climatologies are defined over 1980-1987. The precipitation data are from the NCEP/NCAR Reanalyses (Kalnay et al., 1996) and from the rain gauges at Panaji and Valpoi. The evapotranspiration data are from the NCEP/NCAR Reanalyses. The discharge was measured at Ganjem.
Coe (2000,1998) faced the same problem in certain regions in his global simulation. A DEM is unlikely to resolve all the rivers accurately, and it may be necessary to edit it. Coe (1998) used two objective procedures to eliminate spurious PWAs. The spurious PWAs that were left after these objective analyses were those due to the inability of the DEM to resolve narrow river valleys; these were edited subjectively by visually identifying them in a HYDRA simulation and correcting either or both the DEM elevation and the local flow direction
Since our problem is due to the inability of the DEM to resolve the narrow river valleys accurately, we had to edit the DEM visually. We developed a set of tools based on the GRASS GIS to edit the GLOBE DEM. Once the DEM was edited by changing either or both elevation and local flow directions, the spurious PWAs in the Mandovi basin disappeared and the local runoff was routed by HYDRA to the sea. A comparison between the predicted and observed discharge at Ganjem (Figure 3) shows that the NCEP/NCAR runoff are able to capture the seasonal cycle, but considerably underestimate the magnitude. The difference between the simulated and observed discharges is much greater than the estimated error of 10-15% in the measurements (Dickinson, 1967; Coe, 2000; Cogley, 1989). The evapotranspiration during the summer monsoon in the Mandovi basin is negligible in comparison to precipitation (Figure 2) and the NCEP/NCAR evapotranspiration estimates compare favourably with those estimated by the India Meteorological Department (Rao et al., 1971); hence, the large difference between the simulated and observed discharge must be because of errors in the NCEP/NCAR precipitation.
That the NCEP/NCAR Reanalyses considerably underestimate the precipitation over Goa is confirmed by a comparison with the monthly climatology of precipitation at Panaji (Vose et al., 1992) (Figure 2). Forcing the model with the Panaji precipitation and NCEP/NCAR evapotranspiration (Case 2) (note that the forcing is still uniform in space, i.e., the Panaji precipitation is used to compute the runoffs for all grid cells) gives much better results (Figure 3). The simulated discharge now matches that observed during June; it is, however, still much lower than that observed during the peak of the monsoon during July-August and in the months following it, leading to a much lower estimate of annual discharge (Table 1).
A least-squares fit to station gauge data shows that precipitation increases inland over Goa during the peak of the summer monsoon. The climatology of precipitation at Valpoi, about 32 km inland and located within the catchment of the Mandovi at Ganjem (Figure 1), was available to us (Sulochana Gadgil, personal communication, 2002) for the four months that comprise the summer monsoon -- June-September. The precipitation at Valpoi is comparable to that at Panaji during June and September, but is much greater during July and August (Figure 2). Hence, we construct another climatology of precipitation using the data for Valpoi during June-September and the data for Panaji during the rest of the year. Forcing the model with this climatology (Case 3; note that the forcing is spatially uniform) improves the results considerably, especially in the months following the summer monsoon (Figure 3); the discharge during July and August, and therefore the annual discharge (Table 1), however, are still much less than observed .
The reason for this lies again in an underestimate of precipitation in the catchment of the Mandovi at Ganjem. Much of this catchment area lies at altitudes much greater than at Valpoi, which is just 30 m above mean sea level (Figure 1), and the heavy precipitation over the Indian west coast is known to be due primarily to the influence of orography (Sarkar, 1966). There are no rain gauges available today at stations to the east of Valpoi, but data for 1959 (Anon, 1964) show that the trend of increasing precipitation away from the sea holds even farther eastward than Valpoi. Kulem, the easternmost station for which precipitation information (only for 1959) is available, lies just 100 m above mean sea level; there are no stations on the slopes of the Sahyadris, which rise steeply just a few kilometres to the east of Valpoi and Kulem (Figure 1). Hence, since the precipitation at Valpoi differs from that at Panaji mostly during July and August, we assume that this is true farther east also: therefore, the forcing for Case 4 is as in Case 3, but with the Valpoi precipitation for July and August increased by 30%. (Algorithms that extrapolate precipitation data from lowlands into mountain settings are available, but we have not used them in this paper.) The temporal variation of the simulated discharge is shown in Figure 3 and the spatial variation in Figure 4; the annual discharge now matches the observations (Table 1).
Getting the seasonal cycle of forcing right, however, is not sufficient for simulating correctly the peak discharge in July and August or the discharge during the `lean' season that precedes the monsoon (Figure 3); it is important to simulate correctly both discharges because the former is a measure of the flood potential and the latter is a measure of the potential yield of wells that tap groundwater. These discharge errors are due to the use of constant residence times for the surface runoff (2 hours) and sub-surface drainage (15 days). The large
Figure 3: Comparison between simulated and observed discharge (m3s-1) at Ganjem for the four cases listed in Table 1. The simulation forced by NCEP/NCAR precipitation (Case 1) considerably underestimates the discharge. The discharge estimates improve on forcing the model with precipitation data from the rain gauge at Panaji (Case 2) and on combining the precipitation data for Valpoi with that for Panaji (Case 3); Increasing the Valpoi precipitation for July and August by 30% (Case 4) to account for the possible increase in precipitation on the hill slopes, which form a large fraction of the catchment area of the Mandovi at Ganjem, shows the best match with the observations. With this forcing, the annual discharge (Table 1) matches the observations. The vertical lines indicate the range of discharge at Ganjem during 1980-1987.
amplitude of the seasonal cycle of precipitation (Figure 2) implies a large seasonal cycle of soil moisture. At the time of monsoon onset in June, the soil in the region is dry; hence, much more of the rain water penetrates into the soil column than later during the summer monsoon. This would imply a larger fraction of runoff going into the sub-surface drainage reservoir and also an increase in its residence time, allowing the local runoff during June to move more slowly toward the river. This would not only decrease the discharge at Ganjem in May and June, but also increase it in July, by when the soil is almost saturated. Similarly, the precipitation in September must be retained in the basin much longer than permitted by the residence time of 15 days. It is this precipitation toward the end of the monsoon that is released as sub-surface flows or groundwater in the months preceding the next summer monsoon, when evapotranspiration exceeds precipitation and all the discharge is due to the sub-surface drainage still available in the basin.
Figure 4: The runoff (m3s-1) simulated for July in Case 4, in which HYDRA is forced with the precipitation at Valpoi (June-September) and Panaji (other months), but with the Valpoi precipitation for July and August increased by 30%. The evapotranspiration data are from the NCEP/NCAR Reanalyses, and the forcing is uniform in space. The DEM used for this simulation was edited -- either or both elevation and local flow direction estimated by HYDRA were modified at 25% of the grid cells encompassing the Mandovi basin, whose catchment area is in colour (compare with Figure 1).
It is in this period preceding the summer monsoon that the scarcity of fresh water is felt most in India. At this time, the availability of fresh water is dependent on two sources: storage of surface water and ground water. Storage of surface water occurs in natural and man-make lakes. In HYDRA, this is strongly dependent on the DEM; hence, care must be taken to account for these water bodies when editing it. In HYDRA, groundwater is represented by the sub-surface drainage or baseflow. Therefore, a reasonable estimation of the available fresh water during the months preceding the rains requires better parameterisation of the residence times and of the separation between surface runoff and sub-surface drainage, and therefore, of soil moisture.
The sort of behaviour seen above with respect to soil moisture should also hold for the rest of India. Given appropriate data on local soil conditions, vegetation cover, and surface meteorological observations, this can be attempted with a model of land surface processes. One such model is IBIS (Integrated Biosphere Simulator (Foley et al., 1996; Kucharik et al., 2000)), which has also been developed at SAGE, where HYDRA was developed. The data required for this also exist today: the meteorological observations are made at the several observatories of the India Meteorological Department, and data on soil type and vegetation cover are available from land surveys and satellite remote sensing. The challenge is to use the models and available data to simulate both the peak discharge in July, which is critical for estimating the potential of floods, and the two-orders-of-magnitude smaller discharge before the monsoon, which is critical for estimating the potential yield of fresh water from wells.
Discussion
We have demonstrated that the framework consisting of a freely available hydrological routing algorithm like HYDRA, the GLOBE DEM, the GRASS GIS, and the precipitation and evapotranspiration data available today can successfully simulate the observed monthly-mean discharge in the Mandovi river in Goa. A corollary is that it is possible to use the framework to estimate the discharge at any point along the river. This river being typical of those along the Indian west coast, the framework is equally applicable to the rest of the west coast, permitting an estimate of the water budget of the region. The framework is also applicable to the other rivers of India; the global simulations of Coe (Coe, 2000) on a 5' grid showed a reasonable match between the discharge simulated for the Ganga and the Brahmaputra and the discharge measured at the stream-flow gauges on the rivers.
The GLOBE DEM may also require less editing in the broader river valleys in the rest of India; some of the problems encountered with the DEM, like the presence of pits, can also be handled with available algorithms. The relative sparsity of rain gauges can be compensated for by the precipitation data from AGCMs and satellites; the NCEP/NCAR data, which underestimate the precipitation over the west coast owing to the inability of the model grid to resolve the steep Sahyadris, compare well with the all-India precipitation (Parthasarathy et al., 1995) estimated for the non-hilly subdivisions of the India Meteorological Department based on rain-gauge observations (Shankar et al., 2003). Along the west coast, where the AGCM is unable to account for the orographic effect because of its coarse resolution, statistical downscaling of AGCM precipitation (Wilby et al., 1998) will have to be used in basins without rain gauges. A similar approach will be needed in the Himalayas, where several of the rivers of northern India originate.
The importance of this can hardly be overestimated. There is a dearth of stream-flow gauges in the country (Rao, 1975), and, like gauges elsewhere in the world, they are declining in number (The Ad Hoc Group on Global Water Data Sets, International Association of Hydrological Sciences, 2001; Stokstad, 1999). The framework described in this paper, once validated using existing stream-flow data in the country, can be applied to all basins. It will also help fill the many gaps that exist in available stream-flow data, resulting in a more continuous data set; this will permit an analysis of interannual variability of river discharge over longer times than is possible with the observations alone. Since HYDRA also allows explicit inclusion of dams and irrigation outflows (Coe, 2000), the framework makes possible an estimate of a nationwide water budget. That HYDRA scales well is seen from the range of resolutions at which it has been used; hence, this approach can be used to obtain a budget estimate for an Indian district, the average size of a district being of the same order as that of Goa.
Successful application of the framework will demand the collective efforts of several hydrologists. For this to materialise, the framework has to satisfy the four conditions listed in the beginning of the paper. All the ingredients of the framework -- HYDRA and its ancillary software, the GLOBE DEM, and the GRASS GIS -- meet these requirements. These tools, however, only provide a means to sythesise available data into a coherent and quantitative picture. As the simulations described in the preceding section show, the quality of the synthesis is dependent on the quality of data it brings together. In the case of the Mandovi river, the unsatisfactory aspect is the absence of reliable data on precipitation on the slopes of the Sahyadris. As the assembly of tools described here is enlarged (for example, by incorporating IBIS), it is likely that we will encounter the unsatisfactoriness of the other hydrology data (for example, soil condition or vegetation cover) available in the country. We hope that application of the framework to tackling one of the major problems in the country, that of fresh water resources, will motivate collection of hydrology data, and more important, their open distribution. The data also have to be integrated into a common database to ensure uniformity (condition 4). Building hydrologic databases is a major task and is being done by several groups worldwide (Ramankutty and Foley, 1998; Vörösmarty et al., 1996; Graham et al., 1999). For this also, the data that exist have to be available openly to facilitate interaction among various groups.
The magnitude of the task can be gauged by comparing the area of the Mandovi basin with that of the Indian subcontinent. This is but a modest beginning in simulating hydrology on the scale of the Indian subcontinent. Such an endeavour, beginning appropriately in the International Year of Freshwater, would be a fitting tribute to those who have, over the years, tried to understand and find viable solutions to India's water crises.
Acknowledgements
We thank Michael Coe for making available HYDRA and its ancillary codes, and for answering our questions about the model. Sulochana Gadgil and S. R. Purohit made available the climatology of Valpoi precipitation and Ganjem discharge, respectively. G. S. Michael digitised the rivers from the maps of the Survey of India and helped with the figures, which were made using GMT and Ferret. All the work reported in this paper was done using free software and we express our gratitude to those who have developed these programs, those who maintain them, and those who offer support on the mailing lists; in particular, the support available on the GRASS Developers' mailing list is gratefully acknowledged. Michael Coe, Ravishankar Najundiah, A. Ghosh Bobba, and V. K. Ghanekar reviewed the paper and their comments helped improve the manuscript. This work was supported by grants from the Department of Ocean Development under their INDOMOD programme and from the Department of Science and Technology under their ARMEX programme.
References
D. Shankar
e-mail: shankar@darya.nio.org
Vidya Kotamraju
e-mail:vidyakr@darya.nio.org
S. R. Shetye
Physical Oceanography Division, National Institute of Oceanography
Dona Paula, Goa 403004, India.
Abstract
While issues related to water attract considerable attention in India, very little quantitative information is available on her water budget. There are two reasons for this lacuna: the dearth of information on hydrological variables, and the absence of an easily accessible quantitative framework to put these variables in perspective. In this paper, we assemble a framework to address both issues. At its core is a hydrological routing model; the basic data needed for implementing the framework are a digital elevation model and data on precipitation and evapotranspiration. We demonstrate the viability of the framework by applying it to the hydrology of the Mandovi river in Goa. The model output mimics the observed discharge well.
Introduction
India, as one of the countries affected by both population explosion and the Green Revolution, hardly sees a day without some problem related to water figuring in the news. The problems are twofold (Agarwal and Narain, 1997, 1999; Agarwal and Chak, 1991; Agarwal et al., 1999; Rao, 1975). First, there is the problem of scarcity, resulting in drought or famine; second, there is the problem of surplus (mostly during the summer monsoon, when most of India receives almost 80% of the annual precipitation), resulting in floods. It is also evident that certain regions of the country, especially in the south and northwest, are more prone to drought, while other regions, especially in the east and northeast, are more prone to floods. With each disaster comes the clamour for some long-term solution. One solution that has been proposed is the transfer of water, through the medium of a network of canals, from river basins with a surplus (mostly the rivers in the north and northeast) to river basins with a deficiency (Radhakrishna, 2003; Rao, 1975) (mostly in the south and northwest). This idea received a major boost last year following one of the worst droughts of the century (Gadgil et al., 2002).
The financial implications of networking rivers over an area the size of India are enormous: a tentative estimate for this project is Rs. 560000 crores (Radhakrishna, 2003). Commissioning a project of such magnitude and then executing it successfully will require a vast database on hydrology and a framework that can use the data to quantify the water budget for each river basin. This apart, given the pressure that exists on the resources of fresh water, there is an urgent need to evolve better, sustainable strategies for managing the country's water resources. This, in turn, requires hard data on the basis of which decisions can be made: we need quantitative estimates of water budgets for the river basins at a resolution small enough for evolving strategies for, say, an average Indian district, yet large enough to make possible an estimate of water resources on the scale of the Indian subcontinent.
The estimates of water resources available in India today are often on too small a scale, like a small watershed, or are on too large a scale, like gross statistics based on the average precipitation over the entire country and the measurements made at a few stream-flow gauges on major rivers and in the neighbourhood of some dams (Nag and Kathpalia, 1975; Agarwal and Narain, 1999; Anon, 1988; Rao and Ramaseshan, 1985; Chaturvedi, 1985; Rao et al., 1971; Rao, 1975). For individual river basins, the tendency has been to adopt an empirical systems approach (Chaturvedi and Srivastava, 1981; Rao, 1975). What is lacking is an overview that provides a reliable quantitative estimate of the water resources of the country. There are two reasons for the absence of such estimates. First, there is a dearth of information on the basic hydrological variables of interest -- groundwater recharge rates, stream-flow, evapotranspiration, etc. Only precipitation data have been collected systematically over a long period; evapotranspiration data are scanty (Rao, 2001); stream-flow data are confined to a few rivers, and are often not easily accessible (The Ad Hoc Group on Global Water Data Sets, International Association of Hydrological Sciences, 2001; Radhakrishna, 2003). Second, a quantitative framework that can put these variables in perspective is often missing. Though several hydrological software packages are available, Indian academic institutions are reluctant to invest in them because these packages are usually expensive and are geared to handle situations in developed countries where the hydrological databases are much more advanced. They also rely too often on commercial GIS (Geographical Information Systems) packages, which too are expensive for Indian academic institutions. Hence, in view of the present state of continental-scale hydrology of the subcontinent, it seems that what is needed today is a quantitative framework that satisfies the following conditions.
- The framework should include a simple hydrological model that can provide a reliable water balance of a river system;
- demands on the database required by the model should be consistent with the realities in the country;
- the packages that incorporate the model should be able to handle a range of spatial scales, from small rivers to continental scales, to enable many groups working independently on different river basins to dovetail their analyses into a coherent picture on the larger scale; and
- the models and their ancillary software should be freely accessible.
The framework is described in detail in Coe (2000) and Shankar et al. (2003). Here, we demonstrate its viability by applying it to a river basin. Our choice for testing the framework is the Mandovi river (also called Mahadayi and Mhadei over some stretches), which has its source in the Sahyadri range and flows into the eastern Arabian Sea (Figure 1). Its catchment area is mostly in Goa, but also extends into the neighbouring states of Karnataka and Maharashtra. This choice is dictated by the following considerations. First, being one of the two rivers that flow past Panaji, the capital of Goa and the location of the National Institute of Oceanography, it is one of the best studied estuarine stretches in India, and is typical of the rivers along the west coast of India. Second, it is a small river, with a channel width less than 1 km over much of its 105 km length, making it a rigorous test of the DEM's accuracy: a DEM that works in this region should also work elsewhere in India. Third, discharge data are available for the Mandovi at one point, about 50 km from the mouth, making it possible to validate the framework. We conclude the paper by discussing the potential of this quantitative framework for quantifying the water resources of India.
Figure 1: The topography of the region, as seen in GLOBE data (after editing the DEM). The superimposed rivers were digitised from the maps of the Survey of India. The domain of the simulations described in this paper lies mostly in Goa. The Mandovi and the Zuari are the two main rivers of Goa. The Mandovi flows into the Aguada Bay near Panaji, the capital of Goa and the location of its main meteorological observatory. A stream-flow gauge is located on the Mandovi at Ganjem; Valpoi has the easternmost rain gauge in Goa, but data for 1959 from Kulem, farther east, are available for 1959 (Anon, 1964).
Application to the mandovi river in Goa
The Mandovi is one of the two major rivers located in Goa, the other being the Zuari (Figure 1). The two rivers are connected by a narrow channel, the Kumbarjua Canal (Shetye et al., 1995); many tributaries also feed into the two rivers. The Mandovi, the Zuari, the Kumbarjua Canal, and the tributaries form a network. The cross-sectional area of both rivers decreases rapidly upstream (Shetye et al., 1995). Tidal influence is felt in the Mandovi just a little downstream of Ganjem (Figure 1), approximately 50 km inland, during the dry season (Shetye et al., 1995). The estuarine stretches (where at least some salinity intrusion occurs during the dry season) receive freshwater primarily from rivers that originate in the Sahyadris (or Western Ghats), but there is additional inflow at several points along the main channels. A stream-flow gauge is located on the Mandovi at Ganjem, just upstream of the limit of tidal and salt incursion.
We use the GLOBE (GLOBE Task Team and others, eds., 1999) (Global Land One-km Base Elevation; see the GLOBE web site http://www.ngdc.noaa.gov/seg/topo/globe.shtml for more information) DEM, which has a
resolution of 30 arc seconds (~ 1 km). The topography of the region, as seen in the GLOBE data, is shown in Figure 1; the superimposed rivers were digitised from Survey of India maps and the good match shows that the GLOBE DEM is able to capture the watershed geometry fairly well. The elevation data are used to determine the local flow directions for each grid cell. Once the basin geometry is defined, HYDRA can be forced by prescribing the surface runoff and sub-surface drainage at each grid cell and integrating forward in time; for the GLOBE DEM used here, the model time step is 5 minutes.
We force the model with the precipitation and evapotranspiration data from the NCEP/NCAR Reanalyses (Kalnay et al., 1996) (Case 1; see Table 1 for a brief description of the simulation cases). Since the NCEP/NCAR model resolution is 2.5o, a single NCEP/NCAR grid cell encompasses the entire basin, yielding a spatially uniform forcing field; its temporal variation is shown in Figure 2. The runoff required to force the model is the difference between precipitation and evapotranspiration (runoff is zero when the latter is greater), and the partition between surface runoff and sub-surface drainage is assumed to be the same as in Coe (Coe, 2000): 30% surface runoff and 70% sub-surface drainage. The result is that there is almost no discharge into the sea, most of the local runoff tending to accumulate in lakes 30-50 m deep! Since no such ``potential water areas'' (PWAs: regions where water can pile up, such as lakes or wetlands, and slow the flow through the river) are observed in the Mandovi basin upstream of Ganjem, this is due to the failure of the DEM to channel the runoff into and through the river.
| Table 1: Observed and simulated annual discharge (106m3) at Ganjem. In Case 1, the precipitation forcing is based on the NCEP/NCAR Reanalyses (Kalnay et al., 1996); in case 2, the precipitation forcing is based on rain gauge data for Panaji (Vose et al., 1992); in case 3, the forcing is a combination of rain gauge data for Valpoi (June-September) and Panaji (other months); and in case 4, the forcing is as in case 3, except that the Valpoi precipitation for July and August is increased by 30%. The precipitation and evapotranspiration forcing in all four cases is uniform in space. The annual discharge simulated by case 4 compares well with that observed at Ganjem, providing some justification for the assumption that precipitation on the foothills of the Sahyadris, which form a large fraction of the catchment area of the Mandovi at Ganjem, exceeds that at Valpoi during July-August. |
Discharge 3425 842 1861 2866 3476 |
Figure 2: Climatology of observed precipitation and evapotranspiration (mm/day) and discharge (m3s-1) in the Mandovi basin. The climatologies are defined over 1980-1987. The precipitation data are from the NCEP/NCAR Reanalyses (Kalnay et al., 1996) and from the rain gauges at Panaji and Valpoi. The evapotranspiration data are from the NCEP/NCAR Reanalyses. The discharge was measured at Ganjem.
Coe (2000,1998) faced the same problem in certain regions in his global simulation. A DEM is unlikely to resolve all the rivers accurately, and it may be necessary to edit it. Coe (1998) used two objective procedures to eliminate spurious PWAs. The spurious PWAs that were left after these objective analyses were those due to the inability of the DEM to resolve narrow river valleys; these were edited subjectively by visually identifying them in a HYDRA simulation and correcting either or both the DEM elevation and the local flow direction
Since our problem is due to the inability of the DEM to resolve the narrow river valleys accurately, we had to edit the DEM visually. We developed a set of tools based on the GRASS GIS to edit the GLOBE DEM. Once the DEM was edited by changing either or both elevation and local flow directions, the spurious PWAs in the Mandovi basin disappeared and the local runoff was routed by HYDRA to the sea. A comparison between the predicted and observed discharge at Ganjem (Figure 3) shows that the NCEP/NCAR runoff are able to capture the seasonal cycle, but considerably underestimate the magnitude. The difference between the simulated and observed discharges is much greater than the estimated error of 10-15% in the measurements (Dickinson, 1967; Coe, 2000; Cogley, 1989). The evapotranspiration during the summer monsoon in the Mandovi basin is negligible in comparison to precipitation (Figure 2) and the NCEP/NCAR evapotranspiration estimates compare favourably with those estimated by the India Meteorological Department (Rao et al., 1971); hence, the large difference between the simulated and observed discharge must be because of errors in the NCEP/NCAR precipitation.
That the NCEP/NCAR Reanalyses considerably underestimate the precipitation over Goa is confirmed by a comparison with the monthly climatology of precipitation at Panaji (Vose et al., 1992) (Figure 2). Forcing the model with the Panaji precipitation and NCEP/NCAR evapotranspiration (Case 2) (note that the forcing is still uniform in space, i.e., the Panaji precipitation is used to compute the runoffs for all grid cells) gives much better results (Figure 3). The simulated discharge now matches that observed during June; it is, however, still much lower than that observed during the peak of the monsoon during July-August and in the months following it, leading to a much lower estimate of annual discharge (Table 1).
A least-squares fit to station gauge data shows that precipitation increases inland over Goa during the peak of the summer monsoon. The climatology of precipitation at Valpoi, about 32 km inland and located within the catchment of the Mandovi at Ganjem (Figure 1), was available to us (Sulochana Gadgil, personal communication, 2002) for the four months that comprise the summer monsoon -- June-September. The precipitation at Valpoi is comparable to that at Panaji during June and September, but is much greater during July and August (Figure 2). Hence, we construct another climatology of precipitation using the data for Valpoi during June-September and the data for Panaji during the rest of the year. Forcing the model with this climatology (Case 3; note that the forcing is spatially uniform) improves the results considerably, especially in the months following the summer monsoon (Figure 3); the discharge during July and August, and therefore the annual discharge (Table 1), however, are still much less than observed .
The reason for this lies again in an underestimate of precipitation in the catchment of the Mandovi at Ganjem. Much of this catchment area lies at altitudes much greater than at Valpoi, which is just 30 m above mean sea level (Figure 1), and the heavy precipitation over the Indian west coast is known to be due primarily to the influence of orography (Sarkar, 1966). There are no rain gauges available today at stations to the east of Valpoi, but data for 1959 (Anon, 1964) show that the trend of increasing precipitation away from the sea holds even farther eastward than Valpoi. Kulem, the easternmost station for which precipitation information (only for 1959) is available, lies just 100 m above mean sea level; there are no stations on the slopes of the Sahyadris, which rise steeply just a few kilometres to the east of Valpoi and Kulem (Figure 1). Hence, since the precipitation at Valpoi differs from that at Panaji mostly during July and August, we assume that this is true farther east also: therefore, the forcing for Case 4 is as in Case 3, but with the Valpoi precipitation for July and August increased by 30%. (Algorithms that extrapolate precipitation data from lowlands into mountain settings are available, but we have not used them in this paper.) The temporal variation of the simulated discharge is shown in Figure 3 and the spatial variation in Figure 4; the annual discharge now matches the observations (Table 1).
Getting the seasonal cycle of forcing right, however, is not sufficient for simulating correctly the peak discharge in July and August or the discharge during the `lean' season that precedes the monsoon (Figure 3); it is important to simulate correctly both discharges because the former is a measure of the flood potential and the latter is a measure of the potential yield of wells that tap groundwater. These discharge errors are due to the use of constant residence times for the surface runoff (2 hours) and sub-surface drainage (15 days). The large
Figure 3: Comparison between simulated and observed discharge (m3s-1) at Ganjem for the four cases listed in Table 1. The simulation forced by NCEP/NCAR precipitation (Case 1) considerably underestimates the discharge. The discharge estimates improve on forcing the model with precipitation data from the rain gauge at Panaji (Case 2) and on combining the precipitation data for Valpoi with that for Panaji (Case 3); Increasing the Valpoi precipitation for July and August by 30% (Case 4) to account for the possible increase in precipitation on the hill slopes, which form a large fraction of the catchment area of the Mandovi at Ganjem, shows the best match with the observations. With this forcing, the annual discharge (Table 1) matches the observations. The vertical lines indicate the range of discharge at Ganjem during 1980-1987.
amplitude of the seasonal cycle of precipitation (Figure 2) implies a large seasonal cycle of soil moisture. At the time of monsoon onset in June, the soil in the region is dry; hence, much more of the rain water penetrates into the soil column than later during the summer monsoon. This would imply a larger fraction of runoff going into the sub-surface drainage reservoir and also an increase in its residence time, allowing the local runoff during June to move more slowly toward the river. This would not only decrease the discharge at Ganjem in May and June, but also increase it in July, by when the soil is almost saturated. Similarly, the precipitation in September must be retained in the basin much longer than permitted by the residence time of 15 days. It is this precipitation toward the end of the monsoon that is released as sub-surface flows or groundwater in the months preceding the next summer monsoon, when evapotranspiration exceeds precipitation and all the discharge is due to the sub-surface drainage still available in the basin.
Figure 4: The runoff (m3s-1) simulated for July in Case 4, in which HYDRA is forced with the precipitation at Valpoi (June-September) and Panaji (other months), but with the Valpoi precipitation for July and August increased by 30%. The evapotranspiration data are from the NCEP/NCAR Reanalyses, and the forcing is uniform in space. The DEM used for this simulation was edited -- either or both elevation and local flow direction estimated by HYDRA were modified at 25% of the grid cells encompassing the Mandovi basin, whose catchment area is in colour (compare with Figure 1).
It is in this period preceding the summer monsoon that the scarcity of fresh water is felt most in India. At this time, the availability of fresh water is dependent on two sources: storage of surface water and ground water. Storage of surface water occurs in natural and man-make lakes. In HYDRA, this is strongly dependent on the DEM; hence, care must be taken to account for these water bodies when editing it. In HYDRA, groundwater is represented by the sub-surface drainage or baseflow. Therefore, a reasonable estimation of the available fresh water during the months preceding the rains requires better parameterisation of the residence times and of the separation between surface runoff and sub-surface drainage, and therefore, of soil moisture.
The sort of behaviour seen above with respect to soil moisture should also hold for the rest of India. Given appropriate data on local soil conditions, vegetation cover, and surface meteorological observations, this can be attempted with a model of land surface processes. One such model is IBIS (Integrated Biosphere Simulator (Foley et al., 1996; Kucharik et al., 2000)), which has also been developed at SAGE, where HYDRA was developed. The data required for this also exist today: the meteorological observations are made at the several observatories of the India Meteorological Department, and data on soil type and vegetation cover are available from land surveys and satellite remote sensing. The challenge is to use the models and available data to simulate both the peak discharge in July, which is critical for estimating the potential of floods, and the two-orders-of-magnitude smaller discharge before the monsoon, which is critical for estimating the potential yield of fresh water from wells.
Discussion
We have demonstrated that the framework consisting of a freely available hydrological routing algorithm like HYDRA, the GLOBE DEM, the GRASS GIS, and the precipitation and evapotranspiration data available today can successfully simulate the observed monthly-mean discharge in the Mandovi river in Goa. A corollary is that it is possible to use the framework to estimate the discharge at any point along the river. This river being typical of those along the Indian west coast, the framework is equally applicable to the rest of the west coast, permitting an estimate of the water budget of the region. The framework is also applicable to the other rivers of India; the global simulations of Coe (Coe, 2000) on a 5' grid showed a reasonable match between the discharge simulated for the Ganga and the Brahmaputra and the discharge measured at the stream-flow gauges on the rivers.
The GLOBE DEM may also require less editing in the broader river valleys in the rest of India; some of the problems encountered with the DEM, like the presence of pits, can also be handled with available algorithms. The relative sparsity of rain gauges can be compensated for by the precipitation data from AGCMs and satellites; the NCEP/NCAR data, which underestimate the precipitation over the west coast owing to the inability of the model grid to resolve the steep Sahyadris, compare well with the all-India precipitation (Parthasarathy et al., 1995) estimated for the non-hilly subdivisions of the India Meteorological Department based on rain-gauge observations (Shankar et al., 2003). Along the west coast, where the AGCM is unable to account for the orographic effect because of its coarse resolution, statistical downscaling of AGCM precipitation (Wilby et al., 1998) will have to be used in basins without rain gauges. A similar approach will be needed in the Himalayas, where several of the rivers of northern India originate.
The importance of this can hardly be overestimated. There is a dearth of stream-flow gauges in the country (Rao, 1975), and, like gauges elsewhere in the world, they are declining in number (The Ad Hoc Group on Global Water Data Sets, International Association of Hydrological Sciences, 2001; Stokstad, 1999). The framework described in this paper, once validated using existing stream-flow data in the country, can be applied to all basins. It will also help fill the many gaps that exist in available stream-flow data, resulting in a more continuous data set; this will permit an analysis of interannual variability of river discharge over longer times than is possible with the observations alone. Since HYDRA also allows explicit inclusion of dams and irrigation outflows (Coe, 2000), the framework makes possible an estimate of a nationwide water budget. That HYDRA scales well is seen from the range of resolutions at which it has been used; hence, this approach can be used to obtain a budget estimate for an Indian district, the average size of a district being of the same order as that of Goa.
Successful application of the framework will demand the collective efforts of several hydrologists. For this to materialise, the framework has to satisfy the four conditions listed in the beginning of the paper. All the ingredients of the framework -- HYDRA and its ancillary software, the GLOBE DEM, and the GRASS GIS -- meet these requirements. These tools, however, only provide a means to sythesise available data into a coherent and quantitative picture. As the simulations described in the preceding section show, the quality of the synthesis is dependent on the quality of data it brings together. In the case of the Mandovi river, the unsatisfactory aspect is the absence of reliable data on precipitation on the slopes of the Sahyadris. As the assembly of tools described here is enlarged (for example, by incorporating IBIS), it is likely that we will encounter the unsatisfactoriness of the other hydrology data (for example, soil condition or vegetation cover) available in the country. We hope that application of the framework to tackling one of the major problems in the country, that of fresh water resources, will motivate collection of hydrology data, and more important, their open distribution. The data also have to be integrated into a common database to ensure uniformity (condition 4). Building hydrologic databases is a major task and is being done by several groups worldwide (Ramankutty and Foley, 1998; Vörösmarty et al., 1996; Graham et al., 1999). For this also, the data that exist have to be available openly to facilitate interaction among various groups.
The magnitude of the task can be gauged by comparing the area of the Mandovi basin with that of the Indian subcontinent. This is but a modest beginning in simulating hydrology on the scale of the Indian subcontinent. Such an endeavour, beginning appropriately in the International Year of Freshwater, would be a fitting tribute to those who have, over the years, tried to understand and find viable solutions to India's water crises.
Acknowledgements
We thank Michael Coe for making available HYDRA and its ancillary codes, and for answering our questions about the model. Sulochana Gadgil and S. R. Purohit made available the climatology of Valpoi precipitation and Ganjem discharge, respectively. G. S. Michael digitised the rivers from the maps of the Survey of India and helped with the figures, which were made using GMT and Ferret. All the work reported in this paper was done using free software and we express our gratitude to those who have developed these programs, those who maintain them, and those who offer support on the mailing lists; in particular, the support available on the GRASS Developers' mailing list is gratefully acknowledged. Michael Coe, Ravishankar Najundiah, A. Ghosh Bobba, and V. K. Ghanekar reviewed the paper and their comments helped improve the manuscript. This work was supported by grants from the Department of Ocean Development under their INDOMOD programme and from the Department of Science and Technology under their ARMEX programme.
References
- Agarwal, A. and Chak, A., editors. (1991). Floods, flood plains, and environmental myths. State of India's environment (3). Centre for Science and Environment, New Delhi, India.
- Agarwal, A. and Narain, S., editors. (1997). Dying wisdom: Rise, fall, and potential of India's traditional water harvesting systems. State of India's environment (4). Centre for Science and Environment, New Delhi, India.
-
Agarwal, A. and Narain, S., editors. (1999). The citizens' fifth report -- Part II: Statistical database.
State of India's environment (5). Centre for Science and Environment, New Delhi, India. -
Agarwal, A., Narain, S. and Sen, S., editors. (1999). The citizens' fifth report -- Part I: National overview.
State of India's environment (5). Centre for Science and Environment, New Delhi, India. - Anon. (1964). "Techno-economic survey of Goa, Daman, and Diu". Technical report, National Council of Applied Economic Research, New Delhi.
- Anon. (1988). "Water resources of India". CWC Publication No 30/88, Central Water Commission, New Delhi, India.
- Chaturvedi, M. C. (1985). "Water resources of India -- an overview". Sadhana, 8, 38.
- Chaturvedi, M. C. and Srivastava, D. K. (1981). "Study of complex river systems with screening and simulation models". Water Resour. Res., 17, 796.
- Coe, M. T. (1998). "A linked global model of terrestrial hydrologic processes: Simulation of modern rivers, lakes and wetlands". J. Geophys. Res., 103, 8899.
- Coe, M.T. (2000). "Modeling terrestrial hydrological systems at the continental scale: Testing the accuracy of an atmospheric GCM". J. Clim., 13, 704.
-
Cogley, J. G. (1989). "Runoff from the world's landmasses: Amounts and uncertainties at 2-degree resolution".
Trent Climate Note 89-3, 25 pp. Available from Department of Geography, Trent University, Peterborough, ON K9J 7B8, Canada. -
Dickinson, W. T. (1967). "Accuracy of discharge determinations". Hydrology Paper 20, 54 p,
Available from Colorado State University, Fort Collins, CO 80523, USA. -
Foley , J. A., Prentice, I. C., Ramankutty, N. , Lewis, S. , Pollard, D. , Sitch, S. and Haxeltine, A. (1996).
"An integrated biosphere model of land surface processes, terrestrial carbon balance, and vegetation dynamics".
Global Biogeochem. Cycles, 10, 628. -
Gadgil, S., Srinivasan, J., Nanjundiah, R. S., Krishnakumar, K., Munot, A. A. and Rupakumar, K.(2002).
"On forecasting the Indian summer monsoon: The intriguing summer of 2002". Curr. Sci., 83, 403. - GLOBE Task Team and others, eds., The Global Land One-kilometer Base Elevation (GLOBE) Digital Elevation Model, Version 1.0, 1999. National Oceanic and Atmospheric Administration, National Geophysical Data Center, 325 Broadway, Boulder, Colorado 80303, U.S.A. [Digital data base on the World Wide Web (URL: http://www.ngdc.noaa.gov/seg/topo/globe.shtml) and CD-ROMs].
- Graham, S. T., Famiglietti, J. S. and Maidment, D. R. (1999). "5-minute, 0.5-degree, and 1-degree data sets of continental watersheds and river networks for use in regional and global hydrologic and climate system modeling studies". Water Resour. Res., 35, 587.
- Kalnay, E. , Kanamitsu, M. , Kistler, R. , Collins , W. , Deaven, D., Gandin, L. , Iredell, M. , Saha, S. , White , G. , Woolen , J. , Zhu, Y. , Chelliah, M. , Ebisuzaki, W. , Higgins, W. , Janowiak, J. , Mo, K. C., Ropelewski, C. , Wang, J. , Leetmaa, A. , Reynolds, R. , Jenne, R. , and Joseph, D. (1996). "The NCEP/NCAR 40-year reanalysis project". Bull. Amer. Meteorol. Soc., 77, 471.
- Kucharik, C. J., Foley, J. A., Delire, C. , Fisher, V. A., Coe, M. T., Lenters, J. D., Young-Molling, C. , Ramankutty, N. , Norman, J. M., and Gower, T. S. (2000). "Testing the performance of a dynamic global ecosystem model: Water balance, carbon balance, and vegetation structure". Global Biogeochem. Cycles, 14, 825.
- Nag, B. S. and Kathpalia, G. N. (1975). "Water resources of India". In Proceedings of the Second World Congress on Water Resources, volume 2. Central Board of Irrigation and Power, New Delhi, India.
- Neteler, M. and Mitasova, H. .(2002). Open Source GIS: A GRASS GIS approach. Kluwer Academic Publishers, Dordrecht, The Netherlands.
- Parthasarathy, B. , Munot, A. A., and Kothawale, D. R.. (1995). "Monthly and seasonal rainfall series for all-India homogeneous regions and meteorological subdivisions: 1871-1994". IITM Research Report No. RR-065, ISSN 0252-1075, Indian Institute of Tropical Meteorology, Pune.
- Radhakrishna, B. P.(2003). "Linking of major rivers of India -- bane or boon?". Curr. Sci., 84, 1394.
- Ramankutty, N. and Foley, J. A.. (1998). "Characterizing patterns of global land use: An analysis of global croplands data". Global Biogeochem. Cycles, 12, 685.
- Rao, K. L.(1975). "India's water wealth". Orient Longman.
- Rao, K. N., Raman, C. R. V., Daniel, C. E. J. and Venkataraman S. (1971)."Evaporation over India". Technical Report PPSR No. 146, India Meteorological Department, Meteorological Office, Pune.
- Rao, P. S. (2001)."Large-scale field experiments and modelling land surface processes: A review". J. Agromet., 3, 21.
- Rao, P. S. and Ramaseshan, S. (1985). "Study of Bhakra reservoir operation". Sadhana, 8, 206.
- Sarkar, R. P..(1966). "A dynamical model of orographic rainfall". Mon. Wea. Rev., 94, 572.
- Shankar, D., Kotamraju, V. and Shetye, S.R. (2003) "A quantitative framework for estimating water resources in India" Current Science (communicated).
- Shetye, S. R., Gouveia, A. D., Singbal, S. Y., Naik, C. G., Sundar, D., Michael, G. S., and Nampoothiri, G. .(1995). "Propagation of tides in the Mandovi-Zuari estuarine network". Proc. Ind. Acad. Sci. (Earth Planet. Sci.), 104, 682.
- Stokstad, E. .(1999). "Scarcity of rain, stream gages threatens forecasts". Science, 285, 1200.
-
The Ad Hoc Group on Global Water Data Sets, International Association of Hydrological Sciences.( 2001).
"Global water data: An endangered species". EOS, Transactions, American Geophysical Union, 82, 58. - Vörösmarty, C. J., Fekete, B. M., and Tucker, B. A. (1996). Global river discharge database (RivDis Version 1.0), Vol. 0: Introduction, overview, and technical notes. International Hydrological Program Tech. Document in Hydrology SC. 96/WS/26. Technical report, UNESCO, Division of Water Sciences, 1 rue Miollis, 75732 Paris, France, 1996.
-
Vose, R. S., Schmoyer, R. L., Steurer, P. M., Peterson, T. C., R. Heim, T. R. Karl, and Eischeid, J. . (1992).
"The Global Historical Climatology Network: Long-term monthly temperature, precipitation, sea level pressure, and station pressure data". Technical Report ORNL/CDIAC-53, NDP-041, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA. -
Wilby, R. L., T. Wigley, M. L., Conway, D. , Jones, P. D., Hewitson, B. C., Main, J. , and Wilks, D. S.. (1998).
"Statistical downscaling of general circulation moldel output: A comparison of results.". Water Resour. Res., 34, 3008.