Water Resources

Stream Slopes

Climate, Hydrology, Flood Warning, Hydropower, and Water Quality

The water sector - the distribution of water across the landscape, for multiple purposes - is an integral part of the resource base of a country. Progressive land use and climate changes raise important question, such as -

  • What would be the impacts of changes in agriculture (including irrigation) and forestry practices on local and regional water balances?
  • How would changes in land use practices affect water supply and water quality?
  • Can floods or droughts be predicated, or at least anticipated, one or two months into the future, as an early-warning system?
  • What might effects might changing climate have on water supply and hydropower?

Resolving these questions requires merging information from multiple sources, and lies at the heart of DrukDIF. The approach taken here is to build from ground and satellite observations, within the structure of a computer model of how all the different elements produce the balances of water and energy across the landscape.

Two scales of models are developed - the VIC model for the country as a whole, and the DHSVM model for the Wangchhu.

Direct Measurements of Hydrology and Meterology

The core information for the evaluation of the hydrologic cycle is “reality” - direct discharge (flow measurements) at specific gauges sites on the main rivers. An important point to remember with models is that if there is satisfactory performance at gauged sites, then at least reasonable confidence can be expected for rivers or sites where there are few or no records. The Hydromet Services Division under the Department of Energy, Ministry of Economic Affairs (MoEA) is the government agency responsible for the hydro-meteorological data collection in Bhutan. Surface water data are collected from the hydrological monitoring network established from all the major river basins across the Country, starting in 1991. These critical data are organized by the major river basins, and made available through publishing a yearbook, the Surface Hydrologic Data of Bhutan Yearbook 2008.


A Dynamic Model of Climate, the Landscape, and Water

The relation between climate, landscape structure, and water distribution (hydrology) across Bhutan is being simulated by the geospatially-explicit/process based simulation model, VIC (Variable Infiltration Capacity). Very briefly, VIC is a semi-distributed grid-based land surface hydrologic model which parameterizes the dominant hydrometeorological processes taking place at the land surface-atmosphere interface. The model consists of two major components, vertical and horizontal. The vertical component calculates the water and energy balance components for each individual grid cell. The horizontal component is a convolution integral, which routes the runoff generated at each grid cell to basin outlet (tributary or main stem) channels. A mosaic representation of land cover, and sub-grid parameterizations for infiltration and the spatial variability of precipitation and temperature, account for sub-grid scale heterogeneities in key hydrological processes. The model uses three soil layers and one vegetation layer with energy and moisture fluxes exchanged between the layers. The application of VIC requires the development of a set of input data files, including meteorological forcing (land surface climatology of daily precipitation, minimum and maximum temperature, and winds), vegetation attributes by vegetation class, a river network derived from a digital elevation model, and river discharge history at select stations.

Incorporation of Topography, Soils, and Vegetation Data

The data sets described under Topography and Landcover were set up and incorporated into the VIC model structure, along with climate. The specific development of these datsets for the modeling is summarized below. The model was set up at 1/24-degree (about 4-km)  resolution.


With the physical properties of the basin established, the next critical information layer is surface climate; specifically daily precipitation, minimum and maximum temperature, solar radiation, and winds. Obviously acquisition of such data presents a serious challenge. The Hydromet Service Division maintains a network of ~85 met stations, but coverage of the entire country is extremely difficult by such stations alone. Relative to the needs of this project, we assembled a climate data set based on a combination of satellites and so-called global re-analysis products.

For precipitation, two sources of data were used. The TRMM (Tropical Rainfall Measuring Mission) satellite is a "flying rain bucket." Using a combination of instruments, it produces a precipitation product every 3 hours, at 25 km resolution, beginning in 1998. While accuracy at any one point may be questionable, the product is the best solution over the country. Data for Bhutan were extracted from the global TRMM rainfall data, using the 3B42 V.6 version. Su et al (2008) showed that this version has the potential to be valuable in streamflow prediction in data sparse regions. The rainfall data was accumulated to daily and extracted with a spatial resolution of 0.25 x 0.25° and 0.5 x 0.5°.

Monthly precipitation and temperature time series data were obtained from two gridded observation products. Monthly time series precipitation data for 1900-2006 were obtained from the University of Delware (http://climate.geog.udel.edu/~climate/html_pages/archive.html). The data were adjusted for gauge under catch as described by Adam and Lettenmaier (2003), and for orographic effects as described by Adam et al. (2006). A minimum and maximum monthly temperature data series for 1901-2002 were obtained from the Climate Research Unit (CRU) of the University of East Anglia (http://www.cru.uea.ac.uk/~timm/grid/CRU_TS_2_1.html). Daily time series data were obtained from NCEP/NCAR Reanalysis (National Center for Environmental Prediction /National Center for Atmospheric Research; Kalnay et al. 1996), and re-gridded to 0.5° over global land mask. Quantile mapping between monthly NCAR and observed data is used to fill gaps in the monthly observed data. The NCEP/NCAR data were used to extend the CRU and Delaware data to 2007, by quartile mapping. Then, the daily variability of NCEP/NCAR was used to create daily precipitation and temperature data from the monthly CRU data (for temperature) and U-Delaware (for precipitation), as a control. That is, for a given month the daily precipitation varies as the NCEP/NCAR data, while the amount is controlled by (sums up to) that month's U-Delaware precipitation.

Finally, a script (computer code) was written to take the daily data, to produce gridded forcing data for VIC, and put into the file Bhutan_Forcing.zip (available through DrukDIF). To be consistent with the temperature data, the Delaware precipitation data is used in Bhutan_Forcing.zip. But the TRMM data can be used, as well, as Bhutan_TRMM.zip. The differences between these two products should be evaluated in Phase II; ideally, in comparison with local precipitation data from SEDAT.

Initial VIC Model Setup and Results

For the initial application, VIC was set-up for Basin 16, out of the 17 basins identified from the DEM analysis. This basin essentially cover the Wang Chhu and the Amo Chhu, including those sections of the basins outside of the boundaries of the country.

Initial results, with no calibration, are very promising. For example, simulated results are relatively coherent with observed flow at 3 stations.


 The monthly flow time series for 9 gauging stations were derived, with the R, RMSE, seasonal, and mean 0bserved flow. Since the calibration hasn't been dome yet, only a summary of results to date are warrented.


With the confidence of comparing simulated with observed values at specific gauges, basin wide flows of precipitation, ET, and runoff can be computed, to show regional and seasonal patterns.





In summary, as of October 20, the following can be said about the simulation based on the results available. Even though the calibration of the model has yet to be done and the results for the all stations are not done, the VIC simulation looks very promising with high R values and low RMSE. Thus, the model shows a greater potential for land surface modeling of the kingdom of Bhutan.

For the sub basin considered here, the discharge pattern follows the pattern of precipitation and thus indicating that it could be rainfall dominated than snow dominated.In fact, from the Snow water equivalent data (not shown here), it is only a small fraction of the basin that is covered with snow in part/through out the year.


DHSVM for the Wangchhu

The other model is the basin-scale (or explicit) hydrological model known as Distributed Hydology Soil-Vegetation (DHSVM). In contrast to VIC, DHSVM typically uses a grid resolution of 30 – 150m and is ideally used for basins up to 10000km2. Since DHSVM is a higher resolution saturation excess model, it is more sensitive to changes in vegetation and topography than in VIC.The purpose of this study is to determine the ability of DHSVM to simulate streamflow in the Wangchhu basin and to compare its performance against that of VIC.

The Wangchhu in the Kingdom of Bhutan is a tributary to the Brahmaputra river and its primary economic importance is hydropower. The basin area is roughly 4500 km2. In the Wang Chhu forest covers 43.4% of the total basin area, made up of a variety of fir, coniferous and broadleaf forests (Department of Forests and Park Services, Ministry of Agriculture and Forests Bhutan, 2011).

DHSVM was set up as:

Data sources:
Forcings: NCEP/NCAR reanalysis (forcing data available from 1948 – 2007; see Sonessa et al., 2012)
Soil parameters: FAO Soil Database
Vegetation parameters: Landuse3 Polygon is derived from the 1994 SPOT analysis, via the National Soil Survey Center (NSSC)
Observed streamflow: Hydromet Services Division under the Department of Energy, Ministry of Economic Affairs (MoEA)

Model spinup: 3 times from 1986 – 2006
Model run: 1996 – 2006, 24 hour time step
Results plotted for: 2000 – 2006
Computing time for 10 years: 2.5 hours

As with all hydrological models, model state files must be generated via a spin-up period (initialization period for the system). For DHSVM, the model state files to be generated include a rain interception file and snow, soil and channel state files

The initial results are promising, with DHSVM capturing the seasonality of streamflow very well after minor calibration through varying precipitation lapse rate and channel depth (Figure 2, 3, 4). The peak flows are not captured particularly well, but this may be remedied by more careful or extensive calibration of soil parameters. In comparison with simulations from VIC (Figure 5), the results are comparable. With further calibration, we anticipate that the results would be much improved. Figure 6 shows that inundated areas are captured relatively well in DHSVM i.e. channel flows.