1. Introduction
Bangladesh is noteworthy for, among other things, its low elevation (most of Bangladesh is less than 10 m above sea level), high population density, and location at the confluence of three major rivers: the Ganges, Brahmaputra, and Meghna. During the Northern Hemisphere (NH) summer (June–August, hereafter JJA), Bangladesh receives most of its precipitation; it experiences some of the highest rainfall rates in the NH (see section 3). Because of this combination of factors, summer flooding events in which 20% or more of the total land area is covered with water are not uncommon (Rasid and Paul 1987; Chowdhury 2000, 2003; Koelle et al. 2005). The extent and severity of the flooding depends in part on the intensity, duration, and geographic extent of the summer rainfall; although monsoon river flooding is also influenced by snowmelt and rainfall across the entire catchment basin (Khalil 1990; Madsen and Jakobsen 2004).
Past floods in Bangladesh have led to hundreds of thousands of deaths and have been linked to outbreaks of disease (e.g., Koelle et al. 2005). Thus, the annual and interannual variability of the monsoon rainfall in Bangladesh is an interesting research question with enormous societal implications. Rainfall over Bangladesh is also tied to the larger Indian monsoon system, which drives rainfall across the Indian Ocean region and directly impacts not only the global climate but also the livelihood of more than one billion people.
Numerous rainfall products provide temporal and spatial coverage of rainfall in the Indian Ocean region, including satellite observations after 1979. In this paper, we examine representations of monthly-mean rainfall over Bangladesh and the surrounding Indian Ocean region in five research-quality rainfall products. We find there are fundamental differences between these products and that these differences are of sufficient magnitude as to affect qualitative and quantitative conclusions about rainfall across the region. Agreement between the products is lowest over Burma and Bangladesh; however, we also find significant differences over even the relatively well-observed Indian subcontinent. These discrepancies have obvious and important implications for any study making use of these rainfall products to examine the summer monsoon over the Indian Ocean region in general and Bangladesh and Burma in particular. Of particular concern is that the apparent relationship between rainfall and ENSO can depend on the choice of data product.
2. Data and methodology
The five rainfall products considered here include two that are based on merged satellite–gauge observations and three that are based on rain gauge observations. The merged satellite products considered are the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP; Xie and Arkin 1996, 1997) and the Global Precipitation Climatology Project (GPCP; Huffman et al. 1997). Both of these products combine satellite observations of infrared radiation, microwave emission, microwave scattering, and rain gauge data to produce global monthly-mean rainfall estimates on a 2.5° × 2.5° grid. Although broadly similar, the choice of included products and merging algorithms differs for each product. A detailed comparison of the methodologies used in the two products is presented in Xie and Arkin (1997).
The remaining three products are derived from rain gauge observations alone: the Chen et al. (2002, hereafter CHEN02) product; the Indian Meteorological Department (IMD; Rajeevan et al. 2005) product; and a new, objectively analyzed product based on the Climate Anomaly Monitoring System (CAMS) dataset (Ropelewski et al. 1985), which we refer to here as the CAMS objective analysis (OA) and describe below. Each of these three products consists of monthly rain gauge data interpolated to a regular grid: the CHEN02 product is available at 0.5° × 0.5°, 1.0° × 1.0°, and 2.5° × 2.5° resolution (see Fig. 1 for the distribution of stations for our region of interest); the IMD product at 1.0° × 1.0° resolution; and the CAMS OA product at 2.5° × 2.5° resolution. The interpolation algorithms and the gauge networks are different for each product; thus, as with the satellite products, there is ample opportunity for these gridded products to yield different results, despite their general similarities.
a. CAMS objective analysis
The CAMS OA product presented here consists of monthly-mean precipitation for 1979–2003, objectively analyzed from the CAMS dataset at roughly 6000 stations around the world (see Fig. 1a for station distribution within our region of interest) to create a 2.5° × 2.5° gridded dataset of precipitation. To create the gridded product, we then calculate anomalies at individual stations to avoid the biases that can occur when the anomalies are calculated from gridded data. In a grid box with large station-to-station variability, for example, the absence of a single station’s data at a given time can result in a highly anomalous gridded value simply because of the absence of that station’s report. This effect could also be avoided by requiring that all the stations being used have data at every time; however, in a long time series, that would result in the loss of much of the available data. Thus, we first calculate the anomalies at the individual stations relative to the 1979–2003 monthly station climatologies and then grid the resulting anomalies. We calculate a climatology for a given station and month only if five or more years of data were reported during 1979–2003.
We grid the resulting CAMS monthly anomaly station time series using an algorithm first developed by B. Doty to grid African rainfall data. Each output grid box is divided into one hundred 0.25° × 0.25° subgrid boxes, which are then filled with the data value from the nearest station. The output grid value is the average of the values from these 100 subgrid boxes. The maximum distance searched (d) to fill any of these 100 subgrid boxes is also saved for each grid box. This value can then be used to screen the data to eliminate points that are deemed to be too far from a reporting station. A typical value used in practice for screening data gridded to a 2.5° × 2.5° grid is 4°. An important feature of this gridding method is that gridded points with nearby stations are not affected by more distant data. Thus, the data value at a given grid point is not altered by applying this filter; filtering simply determines whether the value is present or absent. Although simple, an analysis of a multitude of station data gridded with this method reveals that it is adept at retaining the original structure inherent in the station data. This has proven to be particularly effective for data with a detailed structure, such as surface temperature and precipitation (not shown).
Here we grid the data to the CMAP 2.5° × 2.5° grid (Xie and Arkin 1996) to facilitate the comparison with the merged satellite and the CHEN02 2.5° × 2.5° products. Note that, as with other gridded rain gauge products, the methodology described here will produce values over ocean regions within a given distance of a reporting station, despite the absence of included ocean observations.
3. Results
a. Comparison of means
As perhaps the most fundamental check on the agreement between the five rainfall products, we first compare the 1979–2003 JJA climatology for each product (Fig. 2). Somewhat surprisingly, even in this basic measure we find significant differences. The gauge-based CHEN02 (1.0° version), IMD, and CAMS products (Figs. 2a–c, respectively) report significantly larger values than the CMAP and GPCP products (Figs. 2d and 2e, respectively) over Bangladesh and northeast India. The values in the CHEN02 and CAMS products also exceed those reported by the merged satellite products over the Bay of Bengal (the IMD product does not report values for the Bay of Bengal). We also find higher values in the gauge-based products along the west coast of India in the Western Ghats and the India–Nepal border. Both the Western Ghats and the India–Nepal border are areas with abrupt changes in orography, suggesting a systematic difference between the gauge-based and merged satellite products in these regions. Note that the magnitudes are higher in all three gauge-based products, despite that the CAMS OA product uses the same 2.5° × 2.5° grid as the CMAP product and the IMD product includes substantially more stations for this region and period (1803; Rajeevan et al. 2005) than either the CAMS OA or CHEN02 product (we also obtain similar results for the CHEN02 2.5° product). Thus, these differences cannot be explained by resolution alone.
Although the merged satellite products are more similar to each other than they are to the gauge-based products, they are by no means identical. The rainfall maxima over the Bay of Bengal and Bangladesh have a higher amplitude and a greater northward extent in the CMAP product (Fig. 2d) relative to the GPCP product (Fig. 2e). Rainfall is also higher over the Western Ghats in the CMAP product, and the contrast between the Western Ghats and the interior of the subcontinent is more pronounced.
b. Comparison of variability
In addition to the representations of climatological mean rainfall, the degree to which rainfall variability is consistent across the various products is also an important issue. To summarize the differences in variability between the five products, we first remove each product’s 1979–2003 JJA climatology (we also use the CHEN02 2.5° product to allow for a more direct comparison). We then calculate the rms differences between the CHEN02 and CMAP anomalies (Fig. 3a), the CHEN02 and GPCP anomalies (Fig. 3b), the CMAP and GPCP anomalies (Fig. 3c), and the CHEN02 and CAMS anomalies (Fig. 3d). Because of the absence of data over Bangladesh, the IMD product is not included in this comparison.
The differences between the CHEN02 product and the CMAP and GPCP products (Figs. 3a and 3b, respectively) are anticipated by their differences (Fig. 2). The differences in variability tend to follow the region of high rainfall rates near Bangladesh and the sharp changes in orography near Nepal and the Western Ghats. Outside of these regions, the differences are relatively low, although they can still amount to a significant fraction of the mean over drier areas.
Despite the methodology and data sources being similar, we still find substantial rms differences between the CMAP and GPCP products (Fig. 3c) and between the CHEN02 and CAMS products (Fig. 3d). In the case of the merged satellite products, the differences between the CMAP and GPCP products (Fig. 3c) are concentrated in the regions of highest rainfall, particularly along the coasts. These differences are likely a consequence of different methodologies for treating microwave data in mixed land and ocean pixels (Huffman et al. 1997; Xie and Arkin 1997).
In contrast to the merged satellite products, the differences between the CHEN02 and CAMS products (Fig. 3d) tend to be concentrated over the higher topography near Nepal and eastern Bangladesh/Burma. The CHEN02 product includes more stations near Nepal (cf. Figs. 1a and 1b); thus, it is likely to be the more reliable product in this region. In contrast, there are essentially no stations reporting in Burma, and the values here are a clear function of the gridding algorithm and the closest reporting station.
The differences in comparing two gauge-based products or two merged satellite products emphasize the difficulties involved in determining the exact rainfall rates across the Indian Ocean region. The question naturally arises as to the potential impact of these differences on our understanding of the region’s interannual variability. Taking the Bangladesh area mean summer rainfall anomalies as an example (Fig. 4), we find that differences of as much as 4 mm day−1 are typical between the five rainfall products. The degree of agreement between the individual rainfall products varies over the course of the record, with large discrepancies in the 1980s and relatively close agreement in the early 1990s. In a number of these years, even the sign of the anomaly over Bangladesh is dependent on the choice of rainfall product.
The dependence of the sign of the rainfall anomaly on the choice of data product is particularly troubling because it affects even simple qualitative conclusions about whether a summer was wet or dry. We summarize the extent of this disagreement over the sign of the anomaly between the products by calculating the fraction of summer months from 1979 to 2003 in which all the products agree on the sign (Fig. 5). The comparison is limited to the months reporting anomalies with magnitudes greater than 0.15 mm day−1 (roughly 4.5 mm month−1), to avoid emphasizing insignificant deviations from climatology. Consistent with the distribution of rain gauges (Fig. 1) and rms differences (Fig. 3), we find the region of greatest disagreement in eastern Bangladesh and Burma. The four products considered (CMAP, GPCP, CHEN02, and CAMS) agree on the sign of the anomaly across most of this area less than 50% of the time. Agreement is substantially higher across most of India, although it is important to note that values between 60% and 80% dominate most of the subcontinent. Thus, even in this relatively well-observed region, there is substantial uncertainty about the sign of the monthly-mean anomalies during the monsoon.
The impact of this uncertainty can be clearly seen when considering the reported rainfall values during a moderate El Niño event in the summer of 1982. In the gauge-based products (Figs. 6a–c), the rainfall anomaly pattern over Bangladesh is characterized by enhanced rainfall in the east and dryness immediately to the west in India. These wet anomalies are entirely absent in the CMAP (Fig. 6d) and GPCP (Fig. 6e) products. Instead, both the CMAP and GPCP products indicate significant dry anomalies over all of Bangladesh. These differences are sufficient to alter even qualitative conclusions drawn about the influence of this El Niño event on the Bangladesh region.
4. Summary and discussion
Rainfall in Bangladesh and across the Indian Ocean region is related to questions of tremendous scientific and societal interest. Accurate, consistent observations are crucial to advancing our understanding of the processes influencing rainfall in this region, both in the present and in a changing climate.
In this paper, we compare the representations of rainfall across the Indian Ocean in five research-quality rainfall products. Four products have appeared previously in the literature (CMAP, GPCP, CHEN02, and IMD), whereas the fifth product (CAMS OA) is presented here for the first time. We find significant differences between the five products regarding even such basic properties as the climatological summer mean rainfall. We also find significant differences in the representation of variability across the region. These discrepancies are of sufficient magnitude that the products frequently disagree about the sign of the monthly anomaly. Agreement decreases sharply eastward from 60%–70% over India to less than 40% along the coast of Burma. Researchers using these products to study the Indian Ocean region should be aware that there is a sharp decrease in the reliability of data over Bangladesh and Burma (as seen in Fig. 5) and that there is considerable uncertainty even over India.
The sharp drop in data reliability appears to be primarily driven by the absence of surface observations from Burma and by differences in satellite rainfall estimates in coastal regions. These two issues combine to drastically reduce data reliability along the coast of Burma and eastern Bangladesh. Given the large population of the Indian Ocean region, its vulnerability to flooding, and its dependence on the monsoon rains for agriculture, there is clearly a need for further research to reduce the uncertainties represented by the differences in these products.
Acknowledgments
We gratefully acknowledge support from NSF Grants EF-0429520 and ATM-0332910, NOAA Grants NA04OAR4600194 and NA04OAR4310034, and NASA Grant NNG04GG46G. Xavier Rodó also received funding from the project CIRCE, Climate Change and Impact Research (SUSTDEV-2005-3.L3.1-036961-2) UE, 6th FP.
REFERENCES
Chen, M., Xie P. , and Janowiak J. E. , 2002: Global land precipitation: A 50-yr monthly analysis based on gauge observations. J. Hydrometeor., 3 , 249–266.
Chowdhury, M. R., 2000: An assessment of flood forecasting in Bangladesh: The experience of the 1998 flood. Nat. Hazards, 22 , 139–163.
Chowdhury, M. R., 2003: The El Niño–Southern Oscillation (ENSO) and seasonal flooding—Bangladesh. Theor. Appl. Climatol., 76 , 105–124.
Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78 , 5–20.
Khalil, G. M., 1990: Floods in Bangladesh: A question of disciplining the rivers. Nat. Hazards, 3 , 379–401.
Koelle, K., Rodó X. , Pascual M. , Yunus Md , and Mostafa G. , 2005: Refractory periods and climate forcing in cholera dynamics. Nature, 436 , 696–700.
Madsen, H., and Jakobsen F. , 2004: Cyclone induced storm surge and flood forecasting in the northern Bay of Bengal. Coastal Eng., 51 , 277–296.
Rajeevan, M., Bhate J. , Kale J. D. , and Lai B. , 2005: Development of a High Resolution Daily Gridded Rainfall Data for the Indian Region. Meteor. Monogr. Climatol., Vol. 22/2005, National Climate Centre, India Meteorological Department, 27 pp.
Rasid, H., and Paul B. K. , 1987: Flood problems in Bangladesh: Is there an indigenous solution? Environ. Manage., 11 , 155–173.
Ropelewski, C. F., Janowiak J. E. , and Halpert M. F. , 1985: The analysis and display of real time surface climate data. Mon. Wea. Rev., 113 , 1101–1106.
Xie, P., and Arkin P. , 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions. J. Climate, 9 , 840–858.
Xie, P., and Arkin P. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78 , 2539–2558.
(a) Locations of stations used in creating CAMS OA gridded rainfall product. Circles represent stations with more than 5 yr of observations in the period 1979–2003. Shading denotes elevation (m). (b) Mean JJA gauge density for 1979–2003 for CHEN02 2.5° product.
Citation: Journal of Hydrometeorology 9, 5; 10.1175/2008JHM928.1
Comparison of (a) CHEN02, (b) IMD, (c) CAMS, (d) CMAP, and (e) GPCP products for 1979–2003 JJA seasonal means (mm day−1).
Citation: Journal of Hydrometeorology 9, 5; 10.1175/2008JHM928.1
RMS differences (mm day−1) for JJA rainfall anomalies between (a) CHEN02 and CMAP, (b) CHEN02 and GPCP, (c) CMAP and GPCP, and (d) CHEN02 and CAMS products.
Citation: Journal of Hydrometeorology 9, 5; 10.1175/2008JHM928.1
Bangladesh area mean JJA rainfall anomalies (mm day−1) for 1979–2003 for each of five rainfall products: CHEN02 (green line, filled circles), IMD (red line, filled squares), CAMS OA (black line-empty circles), CMAP (red line, crosses), and GPCC (purple line, empty circles).
Citation: Journal of Hydrometeorology 9, 5; 10.1175/2008JHM928.1
Fraction of summer months 1979–2003 in which all datasets reporting observations agree on sign of anomaly. Only includes anomaly values greater than 0.15 mm day−1.
Citation: Journal of Hydrometeorology 9, 5; 10.1175/2008JHM928.1
Comparison of 1982 JJA seasonal rainfall anomaly, expressed as percentage (mm day−1) of JJA seasonal mean, for (a) CHEN02, (b) IMD, (c) CAMS, (d) CMAP, and (e) GPCP products.
Citation: Journal of Hydrometeorology 9, 5; 10.1175/2008JHM928.1
