Abstract

The Brahmaputra River of South Asia is the fourth largest river in the world in terms of annual discharge. The lower Brahmaputra River basin is susceptible to catastrophic flooding with major social, economic, and public health impacts. There is relatively little rainfall and snowpack information for the watershed, and the system is poorly understood hydrologically. Using a combination of available remotely sensed and gauge data, this study analyzes snow cover, rainfall, and monsoon period discharge for a 14-yr time period (1986–99). It is found that interannual rainfall variability is low and is a weak predictor of monsoon discharge volumes. Strong evidence is found, however, that maximum spring snow cover in the upper Brahmaputra basin is a good predictor of the monsoon flood volume. Despite the temporal and spatial limitations of the data, this study’s analysis demonstrates the potential for developing an empirical tool for predicting large flood events that may allow an annual early window for mitigating flood damages in the lower Brahmaputra basin, home to 300 million people.

1. Introduction

The roles and interactions of climate factors in causing flooding in the lower Brahmaputra basin are not well understood. Improved predictions of monsoon flooding could reduce loss of life and economic damage. For example, the 1998 Brahmaputra–Ganges–Meghna flood inundated 69% of Bangladesh (Mirza 2003), displacing over 30 million persons and causing over 1000 deaths (DMB 1998). Of the three major rivers that contribute to flooding in Bangladesh, the Brahmaputra is the largest, with a peak discharge of up to 100 000 m3 s−1 (Brammer 1990). The discharge peaks of the Brahmaputra strongly influence the timing and patterns of countrywide flooding in Bangladesh (Hofer and Messerli 1997).

The relative contribution of rain and snow to Brahmaputra discharge is not well understood. Snow-fed Himalayan tributaries reportedly supply about 63% of the annual discharge at the Bahadurabad gauge (Bruijnzeel and Bremmer 1989), although the actual contribution of snowmelt to the discharge is not known. Studies of Himalayan tributaries of neighboring watersheds in the region have demonstrated that snowmelt can make significant contributions to Himalayan rivers (Seidel and Martinec 2001; Singh and Kumar 1996; Singh and Kumar 1997). Moreover, recent work in the central Nepal Himalaya finds that winter snow accounts for more than 40% of precipitation at higher elevations (>3000 m), with equivalent water amounts of more than 100 cm and high interannual variability (Lang and Barros 2004). Similar snowfall amounts in the Himalayan portion of the Brahmaputra basin could have a significant impact on discharge variability.

In this paper we investigate relationships among Brahmaputra basin seasonal discharge volume, rainfall volume, and snow cover over the 14-yr period 1986–99 for which concurrent data are available. We focus on Brahmaputra River discharge as recorded in northern Bangladesh rather than flooded area or similar metrics because the flooded area is strongly influenced by factors such as local precipitation, preexisting water table variations, and local topography. Furthermore, accurate estimates of the flooded area in Bangladesh are not available for many years (Rashid and Pramanik 1993). The location of our gauge site in northern Bangladesh makes it immune to backwater effects from the coast, which may play a role in other parts of the country. Other studies have explored sea surface temperature in the Bay of Bengal as a predictor of flood-affected area (Chowdhury and Ward 2007). By using more direct predictive parameters, such as snowmelt and precipitation, our analysis attempts to increase our quantitative understanding of flood mechanisms and their relative importance.

2. Datasets

We restrict our analysis to the annual monsoon period (March–September), the time during which 80% of the total annual discharge occurs and destructive floods have been experienced (Chowdhury and Ward 2004). We have defined the monsoon period to include the premonsoon months of March–May and any rainfall during these months that may influence initial soil moisture conditions prior to the monsoon flooding season.

We use three types of data: (i) daily discharge measurements taken at the Bahadurabad gauge in northern Bangladesh (Figure 1) generated by the Surface Water Modeling Center (SWMC) based in Dhaka, Bangladesh, from daily water level observations and a specified rating curve; (ii) regional 1° × 1° gridded monthly precipitation totals from the Global Precipitation Climatology Centre (GPCC) archive for 1986–99 (Rudolf 1993) and the Tropical Rainfall Measuring Mission (TRMM) data product 3B-43 for 1998–2000 (Adler et al. 2000); and (iii) snow cover data (approximately 1° × 1.5° grid cells) from the National Oceanic and Atmospheric Administration/National Environmental Satellite, Data, and Information Service (NOAA/NESDIS) Northern Hemisphere digital archive of weekly observations based on visual interpretation of photographic copies of shortwave remote sensing imagery available in a polar stereographic projection (Dewey and Heim 1982). The length of our time series is limited by the availability of concurrent discharge, precipitation, and snow cover data.

Figure 1.

The U.S. Geological Survey GTOPO30 (∼1-km resolution) DEM indicating the Brahmaputra watershed (650 300 km2) upstream of Bahadurabad flow gauge in northern Bangladesh. Areas used to calculate basinwide rainfall volume and snow coverage are indicated.

Figure 1.

The U.S. Geological Survey GTOPO30 (∼1-km resolution) DEM indicating the Brahmaputra watershed (650 300 km2) upstream of Bahadurabad flow gauge in northern Bangladesh. Areas used to calculate basinwide rainfall volume and snow coverage are indicated.

3. Methodology

The Global 30 Arc-Second Elevation Data Set (GTOPO30, 1-km resolution) digital elevation model (DEM) was used to delineate the Brahmaputra watershed (650 300 km2) upstream of the discharge gauge (Figure 1). The DEM-derived watershed was then used to define appropriate regions of interest (ROIs) within which to examine precipitation and snow cover records. Monthly rain totals at the 1° grid cell resolution of the rainfall data were extracted for the ROI, and spatially and temporally integrated to estimate basinwide monsoon period rain volumes. The daily discharge data were summed over the monsoon period to calculate monsoon period flow volumes. Seasonal changes in snow cover were determined for the Brahmaputra basin (27°–31.5°N, 85°–97.5°E; Figure 1), excluding the southern, lower-elevation cells because they never have any snow cover. The large ROI (n = 25 cells) with snow cover cell resolution of approximately 100–150 km on a side was chosen to avoid poor statistics caused by low pixel counts. The NESDIS snow cover data are binary for each cell (snow/no snow), where “snow covered” is “at least 50% snow cover on the last date of each observing period.” (Robinson et al. 1993) For each month, we compute a snow cover value for the entire ROI (n = 25 cells) by counting the number of snow-covered cells within the ROI for that month, and dividing by the maximum total possible (i.e., all ROI cells covered for all observation periods in that month). In this way, a quantitative index sensitive to the variation in snow-covered area from year to year of the selected ROI is developed.

4. Results and discussion

The overall monsoon period discharge for the Brahmaputra from 1986 to 1999 has a coefficient of variation (CV) of 18% (Figure 2). A test for persistence shows no significant contribution from the previous year’s discharge history. Seasonal average integrated monsoon period rainfall volumes from the 3-yr TRMM data (1998–2000) and the 14-yr GPCC data (1986–99) are near or below the computed Brahmaputra River discharge volume over the same time period. Because there must be large losses to evapotranspiration, failure to record precipitation volume well in excess of the discharge volume suggests chronic underestimation of rainfall. Precipitation estimates in the region that are based on ground station data can be biased downward because of low station density in mountainous zones, where recent work has shown that the greatest precipitation may occur (Barros et al. 2000). Chronic underestimation does not prevent estimated rainfall from serving as a predictor of discharge because a correction factor can be developed.

Figure 2.

Basinwide monsoon Brahmaputra discharge volume for 1986–99 computed by temporally integrating daily discharge values. The dashed line indicates the mean monsoon flow volume for the 14-yr period.

Figure 2.

Basinwide monsoon Brahmaputra discharge volume for 1986–99 computed by temporally integrating daily discharge values. The dashed line indicates the mean monsoon flow volume for the 14-yr period.

The 14-yr record of monsoon period rainfall from the GPCC dataset has an interannual CV of 10%, which is relatively low compared to the interannual discharge variability of 18%. The low interannual rainfall variability recorded by the GPCC data is supported by both the more recent TRMM data and by studies using long-term records (century) of Indian rainfall. For example, although the 1998 monsoon period Brahmaputra river discharge was 41% larger than in 1999, both the TRMM 3B-43 and GPCC datasets indicate that total monsoon precipitation was about 1% lower in 1998 than in 1999. Second, analysis of the spatial and temporal variations in rainfall in the 113 yr of record available from the Indian Meteorological Survey shows a strong trend of decreasing interannual variability in monsoon season rainfall over India as one moves to the northeast (Figure 3). These long-term studies indicate that the variability of precipitation on an annual basis ranges between 10% and 12% for the Indian portion of the Brahmaputra basin (Rao 1984; Parathasarathy 1984; Mooley and Shukla 1987), similar to what we find in the more recent GPCC dataset. We find no indication of interannual variability in rainfall from any of the three precipitation data sources that is as large as that observed in the discharge records.

Figure 3.

The interannual variability in rainfall for the time period 1871–1984 over regions in India shows that the CV increases in a westerly direction and is high over areas of low rainfall and low over areas of high rainfall. The Brahmaputra basin lowlands (boxed area) exhibit the lowest variability (∼11%). This map was produced using CV values for each state from Mooley and Shukla (Mooley and Shukla 1987).

Figure 3.

The interannual variability in rainfall for the time period 1871–1984 over regions in India shows that the CV increases in a westerly direction and is high over areas of low rainfall and low over areas of high rainfall. The Brahmaputra basin lowlands (boxed area) exhibit the lowest variability (∼11%). This map was produced using CV values for each state from Mooley and Shukla (Mooley and Shukla 1987).

Regressing total monsoon flow volume on total monsoon rain volume yields an R2 of just 38%; the regression coefficient is statistically significant at the 1% level. Underestimation aside, monsoon rainfall is a weak predictor of monsoon discharge, and it particularly fails to explain the largest discharge years (1991 and 1998) during the study period of 1986–99. An additional source of variability is required to explain the higher interannual year-to-year variability in seasonal discharge volume. We note that a previous analysis (Chowdhury and Ward 2004) of precipitation records and Brahmaputra basinwide seasonal (June–September) discharge from 1962 to 2000 found a more significant correlation of R2 = 0.66. The study used the climatology of Hulme (Hulme 1994), which is derived from interpolated sparse ground station data, available at a significantly lower spatial resolution (3.75° × 2.5°).

Snow in the Himalayas is the other source of water in this region that could account for variability in flow. To explore possible relationships between snow cover and discharge we generated two indices of snow cover using the NESDIS data: maximum spring snow cover (MSC) and a snowmelt runoff index (SRI). The first index, maximum spring snow cover, simply uses the observed annual snow cover maxima as a predictive tool. The maximum fraction of cells covered with snow usually occurs in early spring during March or April of each year while the minimum fraction of cells covered with snow occurs in late fall (September or October). The interannual CV of MSC is similar to discharge (15%) and, relative to seasonal rainfall, the MSC is a much better predictor for discharge later that year with an R2 of 0.65 (Figure 4) and a p value of 0.5%. When monsoon period precipitation and the MSC are both included in a multiple regression model, R2 rises to 0.74; the regression coefficient for MSC is statistically significant at the 0.1% level and the coefficient for monsoon period precipitation is significant at the 2.5% level. The MSC successfully predicts the two largest flood events in the record (1991 and 1998, respectively). Neither of these events is predicted by rainfall alone.

Figure 4.

Graph showing correlation between MSC and basinwide monsoon flow volume (1986–99).

Figure 4.

Graph showing correlation between MSC and basinwide monsoon flow volume (1986–99).

MSC is a measure of snow cover area and does not provide complete information on the volume of water in the snowpack, which is a function of snow depth, density, and area. To better account for snow water volume, we defined a snowmelt runoff index: SRI = Maxnsnow – Minnsnow, which uses the observed differences in snow cover between spring (maximum) and fall (minimum). See Figure 5. As we have no data on snow depth and density, we applied an exponent n to explore the likely relationship between a snow-covered area and snow water volume. Because a snow-covered area has units of L2, where L represents units of length, and discharge has units of L3, some adjustment is appropriate before a linear relationship is sought between the two variables. A linear relationship between changes in a snow-covered area and water volume (equivalent to uniform snow depth and density over the region) would require n = 1, while if increased snow cover is associated with thicker snowpack and greater water content in some regions (a very likely situation), n must be larger. We found that the regression analyses employing SRI improved from n = 1 to n = 2 but were relatively insensitive to a choice of exponent for n ≥ 2, so n was assigned a value of 2 (Figure 6). The SRI has a CV of 42% for the 14-yr record, more than enough to provide a source of water that can explain a discharge variability of 18%. An examination of snow cover change during 1986–2000 shows that the snowpack that melted during 1998 is anomalously high compared to other years. During the summer of 1997 the snowpack did not melt completely. We suggest that the persistence of snowpack through the 1997 summer contributed subsequently to the extreme flood in summer 1998, when melting was more complete. Regression of monsoon period flow volume on the SRI yields an R2 of 0.60, statistically significant at the 1% level. When monsoon period precipitation and the SRI are both included in the model, R2 increases to 0.64, though monsoon precipitation is not statistically significant.

Figure 5.

Monsoon period rainfall, monsoon period discharge, and snowmelt runoff index SRI with n = 2 (1986–99).

Figure 5.

Monsoon period rainfall, monsoon period discharge, and snowmelt runoff index SRI with n = 2 (1986–99).

Figure 6.

The strength of regression coefficient (R2) of estimates of discharge volume resulting from snow cover change were explored by changing the exponents (n). Snow cover change is computed in terms of the SRI by taking the difference between maximum snow cover raised to an nth power and the minimum snow cover raised to the nth power. As shown, the regression coefficient is insensitive to choice of exponents between n = 2 and n = 3.

Figure 6.

The strength of regression coefficient (R2) of estimates of discharge volume resulting from snow cover change were explored by changing the exponents (n). Snow cover change is computed in terms of the SRI by taking the difference between maximum snow cover raised to an nth power and the minimum snow cover raised to the nth power. As shown, the regression coefficient is insensitive to choice of exponents between n = 2 and n = 3.

Overall, MSC (R2 = 0.65) performs slightly better than SRI (R2 = 0.60) at explaining seasonal discharge variability. When combined with monsoon season precipitation, one obtains R2 = 0.74, which is the best model identified. MSC also has the advantage of being based solely on data that are available prior to the flooding season, and thus is a potentially useful predictive tool with a 3–5-month lead time. However, the SRI highlights the changes in observed snow-covered area and a theoretically more relevant change in the snow water reservoir. The regression results for both indices suggest that snow is an important determinant of seasonal discharge variability and that snow distribution in the mountainous regions of the Brahmaputra watershed is an important area of future research.

5. Conclusions

The two main limitations of the current study are the relatively short time series and the low resolution of the snow cover data. We further recognize that flooding in Bangladesh is not necessarily a simple function of discharge recorded at the Bahadurabad gauge, as local rainfall and other rivers below the gauge also contribute. We do not therefore expect a perfect correlation between any index of snowmelt and flood events. Our analysis demonstrates that despite these limitations, the currently available datasets show significant correlations between spring snow cover and monsoon season discharge in the Brahmaputra basin, and indicate potential for a predictive tool of considerable value. We hope this analysis will point the way to additional studies that can improve the empirical relationships defined here and further our understanding of the mechanistic processes responsible.

Acknowledgments

We thank E. Ahmad of the Institute of Water Modelling for providing discharge data, as well as A. Moore and W. Philpot for valuable comments, and B. Belcher, J. Aliperti, and T. Duffy for technical support. This work was funded partly by the Teresa Heinz Scholars for Environmental Research program.

REFERENCES

REFERENCES
Adler
,
R. F.
,
G. J.
Huffman
,
D. T.
Bolvin
,
S.
Curtis
, and
E. J.
Nelkin
.
2000
.
Tropical rainfall distributions determined using TRMM combined with other satellite and rain gauge information.
J. Appl. Meteor.
39
:
2007
2023
.
Barros
,
A. P.
,
M.
Joshi
,
J.
Putkonen
, and
D. W.
Burbank
.
2000
.
A study of the 1999 monsoon rainfall in a mountainous region in central Nepal using TRMM products and rain gauge observations.
Geophys. Res. Lett.
27
:
3683
3686
.
Brammer
,
H.
1990
.
Floods in Bangladesh: Geographical background to the 1987 and 1988 floods.
Geogr. J.
156
:
12
22
.
Bruijnzeel
,
L. A.
and
C. N.
Bremmer
.
1989
.
Highland-lowland interactions in the Ganges Brahmaputra River Basin: A review of published literature. ICIMOD Occasional Paper 11, International Centre for Integrated Mountain Development, Kathmandu, Nepal, 136 pp
.
Chowdhury
,
M. D. R.
and
N.
Ward
.
2004
.
Hydro-meteorological variability in the greater Ganges-Brahmaputra-Meghna Basins.
Int. J. Climatol.
24
:
1495
1508
.
Chowdhury
,
M. D. R.
and
N.
Ward
.
2007
.
Seasonal flooding in Bangladesh— Variability and predictability.
Hydrol. Process.
21
:
335
347
.
Dewey
,
K. F.
and
R.
Heim
Jr.
.
1982
.
A digital archive of Northern Hemisphere snow cover, November 1966 through December 1980.
Bull. Amer. Meteor. Soc.
63
:
1132
1141
.
DMB
1998
.
Report on Bangladesh flood 1998 (chronology, damages and response. Disaster Management Bureau, Dhaka, Bangladesh, 73 pp
.
Hofer
,
T.
and
B.
Messerli
.
1997
.
Floods in Bangladesh: Process understanding and development strategies. Synthesis Paper, Swiss Agency for Development and Cooperation, Institute of Geography, University of Berne, Berne, Switzerland, 32 pp
.
Hulme
,
M.
1994
.
Global Precipitation and Climate Change.
M. Desbois and F. Desalmand, Eds., NATO ASI Series, Springer-Verlag, 387–406
.
Lang
,
T. J.
and
A. P.
Barros
.
2004
.
Winter storms in the central Himalayas.
J. Meteor. Soc. Japan
82
:
829
844
.
Mirza
,
M. M. Q.
2003
.
Three recent extreme floods in Bangladesh: A hydro-meteorological analysis.
Nat. Hazards
28
:
35
64
.
Mooley
,
D. A.
and
J.
Shukla
.
1987
.
Variability and forecasting of the summer monsoon rainfall over India.
Monsoon Meteorology, Oxford Monogr. on Geol. and Geophys., No. 7, Oxford University Press, 26–58
.
Parathasarathy
,
B.
1984
.
Interannual and long-term variability of Indian summer monsoon rainfall.
Proc. Indian Acad. Sci. (Earth Planet. Sci.)
93
:
371
385
.
Rao
,
K. L.
1984
.
River Basins of India.
Vol. 5, Contributions to Indian Geography: Resource Geography, A. Ramesh, Ed., Heritage Publishers, 225 pp
.
Rashid
,
H.
and
M. A. H.
Pramanik
.
1993
.
Areal extent of the 1988 flood in Bangladesh: How much did the satellite imagery show?
Nat. Hazards
8
:
189
200
.
Robinson
,
D. A.
,
K. F.
Dewey
, and
R.
Heim
Jr.
.
1993
.
Global snow cover monitoring: An update.
Bull. Amer. Meteor. Soc.
74
:
1689
1696
.
Rudolf
,
B.
1993
.
Management and analysis of precipitation on a routine basis. Proc. Int. WMO/IAHS/ETH Symp. on Precipitation and Evaporation, Vol. 1, Bratislava, Slovakia, Slovak Hydrometeorological Institute, 69–76
.
Seidel
,
K.
and
J.
Martinec
.
2001
.
Snowmelt contributions to runoff in an extremely wide altitude range from large area satellite imagery.
Proc. Fifth Int. Workshop on Application of Remote Sensing in Hydrology, Montpellier, France, ESA-DUP Project, 1–14
.
Singh
,
P.
and
N.
Kumar
.
1996
.
Determination of snowmelt factor in the Himalayan region.
Hydrol. Sci. J.
41
:
301
310
.
Singh
,
P.
and
N.
Kumar
.
1997
.
Impact assessment of climate change on the hydrological response of a snow and glacier melt runoff dominated Himalayan river.
J. Hydrol.
193
:
316
350
.

Footnotes

* Corresponding author address: Shithi Kamal-Heikman, Department of Geography, 1832 Ellison Hall, University of California, Santa Barbara, Santa Barbara, CA 93106-4060.

kamal@geog.ucsb.edu