87th AMS Annual Meeting

Wednesday, 17 January 2007: 8:45 AM
Generating river discharge estimates for the Bay of Bengal using NASA'S Land Information System
213A (Henry B. Gonzalez Convention Center)
John E. M. Brown, Univ. of Miami/RSMAS, Miami, FL
Poster PDF (1.1 MB)
Although the land drainage area of the Indian Ocean is rather small, the influence of the Asian rivers is amplified by the monsoonal climate. The summer floodwaters of the Ganges and Brahmaputra rivers discharging into Bay of Bengal plus the Irrawaddy and Salween rivers emptying into the Andaman Sea combine to influence the salinity of the surface waters over thousands of kilometers offshore.

In the hydrological research community the Bay of Bengal and Andaman Sea have not been studied as much as other regions. While global climate models have produced continental scale estimates for this region, there has been little, if any, basin scale modeling studies conducted. The Bay of Bengal region has also not been studied well from an ocean modeling point of view. The role of river influx, especially that from the Ganges/Brahmauptra (GB) river, in the dynamics of coastal currents in the Bay of Bengal is not well known.

Land surface modeling seeks to predict the terrestrial water, energy, and biogeochemical processes by solving the governing equations of the terrestrial medium. The land surface and atmosphere are coupled to each other over a variety of time scales through the exchanges of water, energy, and carbon. An accurate representation of land surface processes is critical for improving models of the boundary layer and land-atmosphere coupling at all scales and domains. Long term descriptions of land use and fluxes also enable the accurate assessments of climate characteristics. In addition to the impact on the atmosphere, predicting land surface processes is also critical for many real-world applications such as water resource management.

Here, we use NASA's Land Information System (LIS) and the Noah LSM to generate river discharge estimates. The model was forced using NOAA's merged satellite/gauge precipitation product and a NASA finite-volume numerical weather model. The University of Washington's VIC routing model was used to channel the LIS surface and sub-surface runoff. Flow direction was along the University of New Hampshire's STN-30 0.5° artificial river network. Velocity information was derived for each network point utilizing 2-minute global relief data. LIS was run at 0.25° resolution, on hourly time steps with daily output for the year 2002.

Model performance is analyzed against observations from Georgia Institute of Technology (GT) and climatology data from the Global River Data Center (GRDC) looking at the GB basin as a first case. We test the hypothesis on whether coupling a river routing scheme to a LSM is beneficial. Basic calculations show that travel times for most river basins in the study area are less than 30 days, supporting the argument that routing is not that necessary on monthly time scales. Totaling the surface (Qs) and sub-surface (Qsb) runoff could be used as a substitute parameter for river discharge (Qtot).

Values from the non-routed flow look low compared with the range given by the GRDC climatology, finally approaching the GRDC extreme minimum in August and September (within 10%) then staying within range from October to December (within 5%). In the dry winter months the Qs makes up only a small portion of the total river discharge, then as the summer monsoon begins in June the impact of the rainfall can be seen making up almost 50% of the total discharge. By July Qsb starts to jump dramatically as the soil moisture retains memory of the earlier rainfall. It is interesting to see that there is little difference between the August and September LIS generated Qtot, but that the contribution from Qs and Qsb has changed. We find that using the combination of both Qs and Qsb generated over a basin-wide area and averaged for each month could give an acceptable estimate for river discharge, especially for un-gauged areas where no real-time observations are available.

Results using the VIC routing model show that the dry season months are somewhat lower than the observed GRDC extreme minimums, but that the peak of the monsoon is close to these GRDC minimums. Without knowing more about the monsoon for 2002 and without having access to observations from these GRDC stations it would be difficult to make an immediate assessment on the impact of river routing. According to the Indian Institute for Tropical Meteorology, 2002 was a drought year with rainfall being ~35% less than normal by mid-August recovering somewhat by the end of September. If we multiply the GRDC mean values for the summer monsoon months by 0.65, we end up with values that resemble the VIC routing values and the GRDC observed climatological minimum.

In comparison to the GT observations, the VIC routing results follow the observations in their pattern, but are much lower for January - April, are only somewhat lower in May and June, miss the peaks of July and August and then are fairly close for September - December. Results for the Godavari, Krishna, Irrawaddy, and Salween river basins are compared only to GRDC climatology since no discharge observations for these rivers are available. These results may have been improved if the experiment was run with more optimal initial conditions and more knowledgeable operator experience.

One possible explanation for the lower than expected VIC routing values could be a problem encountered when using a post-processor river routing model. Because the LIS model state file contains only information about the state variables that are internal to the LIS model, it does not store information about the routing process. As a result, the routed runoff from a LIS model run that is restarted from a state file can be under-predicted compared to a model run that has been “properly” spun-up. A “start-up” effect is observed that is expected to disappear after a few months when routing arrays have been “filled”.

Overall, the results are encouraging and show promise in providing critical river discharge values in areas where observational data are not readily available.

Supplementary URL: