Frequency Analysis of Water Levels
The fitted PDFs and the corresponding values of PPCC for annual maximum water levels of the selected gauge stations and for maximum discharge for Kushiyara River at Sherpur (SW_175.5) station are shown in Tables 2, 3, 4, 5, and 6. The probability plots along with 90% confidence interval for the annual maximum water levels of the selected gauge stations are shown in Fig. 4. It is seen that the observed values fall well within the 90% confidence interval of the fitted distributions for annual maximum water level.
Model Calibration and Validation
After simulating the hydrodynamic model Delft3D, the first step was to calibrate and validate the model’s output with observed data. Water level (WL) data for the month April to May 2007 (pre-monsoon) of Markuli (SW 270) station were used for calibration purpose. The selected Markuli (SW_270) station belongs to the Surma-Meghna River system (Fig. 1). The calibrated parameter was the Manning’s roughness coefficient ‘n’ in the river. Similar to the study by Dutta and Nakayama (2009), land use types were the basis for estimating the roughness coefficients for rivers and surface in this present study. From the calibration, a close agreement between the observed and calculated water level (Fig. 5) using the roughness coefficients between 0.019 and 0.023 was found.
After calibration, the model was validated against the water levels (WL) for the month April to May 2007 of Azmiriganj (SW 271) station at Surma-Meghna river. The validation shows that the observed and computed water levels are close (Fig. 6). The computed water level at this station was found to vary within − 0.17 and the coefficient of determination (R2) is 0.72. Further validation was carried out between the observed flood extent of 1998, a 100-year flood event (Islam and Chowdhury 2002), and the modelled 100-year flood extent (Fig. 7). The extent of 1998 flood was maximum from 27th July 1998 to 07th September 1998 in the study site as per BWDB (2010). RADARSAT image covering the 1998 inundation extent was found available for 26th August, which is within the specified range of maximum flood extent. Therefore, this RADARSAT image was used for delineating the observed inundation extent of 1998 flood. It was found 76.5% accuracy between the modelled and the observed inundation extents when a GIS overlay operation was performed between these two extents.
Simulation by Delft3D model yielded floodplain inundation depths at different return periods, as presented in Fig. 8 and Table 7. It was found that overall with the increase in return period the inundated area increases substantially for the flood class of “Low (0–2 m)” and “Medium (2–4 m)”. However, exception is 20-year return period of flood, where inundated area for the flood class of “Low (0–2 m)” decreased as compared to the flood class of “Low” at 10-year return period. Noticeably, at 2.33-year flood event, only the north side of the study area was inundated, but with greater return period of flood the western and the southern sides of the study area were inundated gradually. There was no inundated area for “High (4–6 m)” class of flood at 2.33-year return period; however, at greater reoccurrence of flood (greater than 10-year) inundated area of “High” class of flood was present with slight extent (~ 1 km2). The percentage area of inundation increased from 28.77 to 80.28% corresponding to 2.33-year return period to 100-year return period, respectively, for flood class low. Whereas for medium and high class floods, the area of inundation increased from 0.31 to 10.23% and 0 to 0.28% for return period 2.33 years to 100 years, respectively.
At low reoccurrence interval of flood, the major river systems are not always seen to play major role in flooding the adjacent floodplains by their overbank flows. Rather, floodplains are inundated from the flows coming from local floodplain channels connected with the major river systems (Jeb and Aggarwal 2008; Luo et al. 2018; Tanaka et al. 2017). This is why the distribution of floodplain inundation can be sporadic at low reoccurrence interval of flood. However, at later stages of flood as water levels continue to rise, floodplains get closer to major river and meet flows directly coming from the river (Fantin-Cruz et al. 2011; Karim et al. 2016; Yin et al. 2013; Zin et al. 2018). Then, a vast expanse of the floodplain is inundated. The pattern of inundation in the study site showed similar behaviour at different return periods of flood. Except few areas, the whole study site was inundated (Fig. 8) by the overbank flows coming from the major rivers (Fig. 1) (the Kushiyara, the Khowai, the Surma-Meghna and the Barak) at recurrence interval of flood greater than 20. At relatively low recurrence intervals of flood, a small portion of the study site, northern portion mainly, was inundated possibly by local floodplain channels only. The floods with a relatively low return period have a large influence on the annual risk. At the same time as these floods may cause relatively low economic damage per event, their relatively frequent occurrence means that they should be fully considered in flood risk assessments (Ward et al. 2011).
The inundation extent in the study area simulated by Delft3D model is comparable to other studies. The percentage of flooded area in Baniachong Upazila was 86.63 as on June 13, 1998 (BWDB 2010). Study conducted by Bhuiyan et al. (2010) termed the 1998 flood is a return period of 75 to 100 years. In this study, for 100-year return period, 80.28% (399.34 km2) area was found under flooding (Table 7), which is quite close to the BWDB study.
Inundation of Different Land Use Categories
Supervised classification of LANDSAT image with ArcGIS yielded different land cover existing in Baniachong. The land cover map was assessed against Google Earth image. The assessment was 30 m apart sampling basis, since the LANDSAT image used in this study is approximately 30 m in resolution. Bai et al. (2015) also used Google Earth images when assessed land use map of China. Overall accuracy in the Bai et al. (2015) study varies from 48.6 to 68.9%. In the present study, the assessment shows 70%, 100%, 60%, 62% and 60% accuracy in water bodies, urban settlement, rural settlement, cropping land and bare land categories, respectively, in the derived land cover map.
The land use map of Baniachong is shown in Fig. 9. About ~ 4%, ~ 18%, ~ 27%, ~ 37% and ~ 15% are covered by water bodies, urban settlement, rural settlement, cropping land and bare land, respectively. This distribution of land use classes is representative to a typical rural area of Bangladesh (e.g. Khan et al. (2015), Parvin et al. (2017)).
Figure 10 presents the percentage of inundation area for each land use class at different return periods of flood. It is found from Fig. 10 that the affected area increases with the increase of return period and flood depth for all land use classes. It is noticeable that, with the increase in return periods from 2.33 to 100 inundated areas become more than tripled for land use classes rural settlement (~ 37 to ~ 126 km2), urban settlement (~ 13 to ~ 63 km2) and bare land (~ 14 to ~ 57 km2), and more than doubled for land use class cropping land (~ 76 to ~ 188 km2). The percentage inundation of urban settlement (~ 16 to ~ 85) is higher than that of rural settlement (~ 15 to ~ 69) with the increase in return period. Water bodies were inundated much more (~ 29% at 2.33-year flood and ~ 90% at 100-year flood) than that of any other classes of land use. It reflects the loss of capture fisheries during flooding. The rising trend of inundated area for cropping land, which is the most dominant land use type, decreases with the increase of return periods. Noticeably, in case of higher reoccurrence interval of flood (e.g. 50 and 10-year floods), inundated area for cropping land remains the same with the increase of return period.
The settlement areas (both urban and rural) in the study site are slightly elevated locally, as like as other settlement areas in Bangladesh (Choudhury 1973). This is why these areas usually do not receive flood flow from major rivers nearby or even from local floodplain channels at low reoccurrence interval of flood. Furthermore, man-made structures, such as road network, obstruct lateral and longitudinal connectivity of flood water fluxes (Kumar et al. 2014), thereby reducing the chance of inundation in the settlement areas of Baniachong Upazila. As a result, only a negligible portion of the settlement areas (~ 28% for rural settlement and ~ 16% for urban settlement at 2.33-year flood) were inundated at low reoccurrence interval of flood as found from the simulation. However, at greater reoccurring interval of flood, the settlement areas received flood water as the flood flows defeat the settlement elevation and or overtop the man-made structures. This led to abrupt inundation of the settlement areas, thereby increasing the inundated area in percentage (~ 28% at 2.33-year flood to ~ 94% at 100-year flood) (Fig. 10). Cropping or agriculture land areas, which are typically as depressions or low elevated zone in active and older floodplain, often receive flood water from major rivers as well as from local floodplain channels (Charlton 2008). The cropping land area in Baniachong Upazila most probably has similar characteristics in terms of flood water connectivity with the major rivers and the local floodplain channels. Therefore, most of the area of cropping land was inundated at low reoccurrence interval of flood (~ 40% and ~ 80% at 2.33 and 10-year floods, respectively) and unlike settlement areas, the inundated area did not increase abruptly and even did not change at greater reoccurrence interval of flood (from 50- year to 100-year flood) (Fig. 10).
Floodplain Damage Vulnerability
Damage Function and Damage Vulnerability Mapping
Figure 11 shows the depth–damage curves for two elements of risks: cropping lands and rural settlements for the study site. Depth–damage functions for these two elements were constructed with the help of hazard maps shown in Fig. 8. It is noted here that the damage function shown here for rural settlement refers to an average for four dominant types of settlements usually found in the study site (as discussed in Sect. 2.4.2). This damage function was used to represent the physical vulnerability of the rural settlements since it was not possible to distinguish the four different types either in the satellite image processing or through field survey.
Using the hazard map (Fig. 8) and the stage-damage curve (Fig. 11), crops vulnerability maps for different return periods of flood were constructed, as shown in Fig. 12. In the crop vulnerability mapping, the flood depths were divided into five scales (0–0.5 m, 0.5–1 m, 1–1.5 m, 1.5–2 m and 2 m and above) and their respective vulnerability is: very low vulnerable (0–0.25), low vulnerable (0.25–0.45), medium high vulnerable (0.45–0.65), high vulnerable (0.65–0.84) and very high vulnerable (0.84–1) for different return periods of flood.
From Fig. 12, it is found that crop vulnerability increases with the increase in return period. Almost 20% of the total cropping land area is “high” and or “very high” vulnerable to 100-year flood. These areas are close to the rivers (Fig. 12), and further generally laying at low elevations (Elevation map—Fig. 1). On the other hand, ~ 42% of the total cropping land area is “low” to “medium high” vulnerable to 100-year flood. Most of these areas tended to be further away from the high drainage density areas. Significantly, the results in Fig. 12 depict the fact that the cropping land in northern portion of the study site is much more vulnerable to flood than any other area of the study site. This is due to the fact that the northern portion is very close to the Suriya-Kalakhai River system (Fig. 12), one of the major rivers in the study site. Furthermore, the area is a depression zone identified in the elevation map in Fig. 1. Consequently, the extent of flood damage would be higher in norther portion than any other portion of the study site. Note that, this proposition is true for same flood, in terms of intensity and exceedance probability. However, one might argue with different varieties of crop as crop vulnerability can differ from one variety to another variety (Cutter 1996). This is probably a scope of further study in future.
Rural settlement vulnerability maps for different return periods of flood were constructed, as shown in Fig. 13. In the rural settlement vulnerability map, the flood depths were divided into six scales (0–0.5 m, 0.5–1 m, 1–1.5 m, 1.5–2 m, 2–2.5 m and 2.5 m and above) and their respective vulnerability is: very low vulnerable (0–0.15), low vulnerable (0.15–0.35), medium low vulnerable (0.35–0.6), medium high vulnerable (0.6–0.75), high vulnerable (0.75–0.9) and very high vulnerable (0.9–1) for different return periods.
The result of vulnerability map for rural settlement showing in Fig. 13 depicts that rural settlement vulnerability increases with the increase in return period. Overall, irrespective of any class of vulnerability, ~ 28% of the total rural settlement area is vulnerable to flood at 2.33-year return period. This percentage becomes ~ 95% at 100-year flood. Now, if looking on the basis of vulnerability class, only 0.2% of the total rural settlement area is “high” and or “very high” vulnerable to flood at 2.33-year return period. However, at 100-year flood this percentage increases to ~ 8. This increasing trend is probably due to the nearby Kalni River (Fig. 13). Noticeably, rural settlement area with “low” and or “very low” vulnerable to flood increases sharply with the increase in return period (~ 15% of the total rural settlement area at 2.33-year flood increase to ~ 30% of the total rural settlement area at 100-year flood). These areas where rural settlement vulnerability with “low” and or “very low” increases are tended to be either close to the Suti River and the Barak River (Fig. 13) or laying at relatively low elevations (Elevation map—Fig. 1). As a whole, distance from river to settlement determines the settlement vulnerability in the study site. However, type of settlement is obviously a factor in this regard, but it is out of scope in the present study.
Damage estimation and risk mapping
The expected damage of the inundated land use types was estimated using equation outlined in Sect. 2.5. The value of P (property value) was found to be Tk. 9.4 (Tk. 2820 per cell) for agriculture and Tk. 590 for settlement (Tk. 177,000 per cell). Thus, raster-based damage maps for various return of floods were produced.
Figure 14 shows the expected damage or risk map for cropping land at different return periods of flood. Figure 15 presents the percentage of cropping land area with different levels of risk at different return periods of flood. From Figs. 14 and 15, it is found that overall the cropping land area with different classes of flood risk increases with the increase in return period. However, cropping land areas with “Low” and “High” flood risk decrease (~ 30 km2 to ~ 28 km2 for “Low” and ~ 28 km2 to ~ 12 km2 for “High”) with the increase in return period from 10 to 100-years and 50 to 100-years, respectively (Fig. 14). The cropping land in southern portion of the study area is not under risk at 2.33-year flood, but in northern portion it is always under risk at all level of floods (Fig. 14).
The reason why cropping land areas with “Low” and “High” flood risk decrease with the increase in return period is probably because these areas are shifted to either risk class of “Medium” or “Very high” when affected by greater reoccurrence interval of flood. Overall, existing topography and river-channel network performing as vulnerability element and flood depth as hazard element determine the spatial distribution of different levels of flood risk for the cropping land in the study site. The northern portion of the study area has depression (Elevation map—Fig. 1), and this is probably the main reason for which the area is under flood risk at all level of floods. Furthermore, the Kalni River at north is also responsible for the area to be flooded at all level of floods. The drainage density in southern portion of the study area is relatively higher (Fig. 14), thus increasing the flood risk level for cropping land in respective area.
Figure 16 shows the risk maps for rural settlement at different return periods of flood. Figure 17 presents the percentage of rural settlement area with different levels of risk at different return periods of flood. Figures 16 and 17 show that except the area covered by the risk classes of “Low” and “Medium high” all other classes increase with the increase in return period. Significantly, the area covered by the risk class of “Very low” increases much more (~ 14 km2 at 2.33-year flood to ~ 67 km2 at 100-year flood) than any other classes. The rural settlement areas located south and south-east of the Suti River (Fig. 16) are not at risk to 2.33-year flood. However, at greater reoccurrence interval of floods, these areas, particularly the areas close to the Suti River, fall under the risk classes from “Medium low” to “Very high”. The reason could be the existence of intricated network of river-channels close to the area as shown in Fig. 16.
Overall, the study finds that areas with different levels of flood risk do not correspond to the areas of inundation at different return periods of flood. While the areas of inundation increase with increase in return period of flood, the areas with different levels of flood risk are determined by depth of inundation and depth–damage function (i.e. vulnerability index).