3.1 Selection of the study area: The study area is Shyamnagar Upazila in the Sathkhira district of Bangladesh. Natural disasters such as cyclones, storm surges, tidal floods, saline water intrusion and water logging are prominent features of this area. In addition, land-use and land-cover change is another major concern. The human-induced shrimp farming initiates salinity that seriously affects agricultural production and makes the region vulnerable to unsafe drinking water.
3.2 Description of the study area Shyamnagar (Figure 3.1) is the largest Upazila, of Satkhira district. This Upazila occupies an area of 1968.24 km2 , including 1,622.65 km2 offorest area. It consists of 13 unions, and the total population of this Upazila is 313,789 (BBS 2001). Household income predominantly depends on agriculture. Almost 65 percent of the total population is involved in agriculture and 38 percent depend on cropping, livestock, forestry and fisheries as their main source of income, whilst 27 percent derive their income from selling agricultural labor. The total cultivable land is 38,552 hectares, with 6,258 hectares of fallow land. Among the total population, 19 percent are landless, 30 percent landed but small, 28 percent are marginal, 16.5 percent intermediate and 6.5 percent are wealthy (BBS 2001).
3.3 Satellite Image analysis Satellite images have been analyzed to answer the research questions and fulfill the objectives of the study. Images have been obtained from the United States Geological Survey (USGS). Two sets of satellite images, dated1999 and 2012, have been used for the analysis. Images obtained in 2012 had a gap mask due to the Scan Line Connector of Landsat ETM+ being offline (SLC off). The mosaicking method was applied to fill those gaps using ENVI\IDL TM. For minimizing the differences in the Digital Number (DN) value of each pixel, a radiometric correction procedure was applied; then Ground Control Points were collected to register these two images. The root mean square error (RMSE) was 0.3. Supervised classification with a maximum likelihood algorithm was applied for image classification. Later, two classified images were used for change detection.
3.4 Data collection and analysis A questionnaire survey was conducted to fulfill the objectives and prove the study’s hypothesis. The questionnaire survey was completed between February and March 2012. A semi-structured household-level questionnaire was developed to investigate the effects of natural disasters, which was pre-tested before conducting the survey. To clarify the purpose of the survey, informed consent was agreed upon with the respondents. A total of 144 household-level questionnaire surveys were conducted in three unions, namely Shyamnagar, Koikhali and Nildumur, with 55, 45 and 44 responses respectively. These three unions were selected because natural disasters are common in them, while the impact level of disasters is different. Non-probability random sampling procedures were followed for the household surveys (Bryman 2004). All the households were selected as the researcher walked through the roads of those areas. Three focus group discussions (FGD) were conducted to investigate the reasons behind the land use and land-cover changes that had occurred in the study area. In each union, one FGD has been conducted where both agricultural farmers and shrimp farmers were the main focus. The numbers of participants in the FGD were 9, 8 and 9 for each of the three unions respectively. Four individual informal interviews were conducted through a semi-structured questionnaire; these comprised two government and two non-government officials. Some secondary data were collected in order to obtain an overview of the occurrence of disasters and their impact on the study area. These data were collected from the Internet, the Local Government Engineering Department, the Upazila Agriculture Extension Office, the Upazila Fisheries Office and the Meteorological Department of Dhaka, Bangladesh. Frequency tests were utilized in order to compare categorical data, such as age, sex and occupation, across the three unions. Probit regression analysis was performed to prove the hypothesis of the study because all the variables collected from the household survey showed normal distribution. The level of significance for the analysis was set to p<0.08.