2.1 Location of the study Bangladesh (latitude: 200 34’N-260 38’N; longitude 880 01’E-920 41’E) has a tropical monsoon climate distinguished by heavy seasonal rainfall, high temperatures, and high humidity. The average annual rainfall fluctuates from 1,500 mm in the west-central to over 3000 mm in the southeast and northeastern parts of the country. More than 70 per cent of total rainfall in Bangladesh occurs during monsoon (July to September). The mean summer and winter temperature ranges from 30 to 40° C and 18 to 22° C respectively, whereas April is the hottest and January is the coldest month [Bangladesh Meteorological Department (BMD), 2016]. Based on climatic conditions, the land of Bangladesh has been divided into 7 climatic zones, as shown in Figure 1. There are two distinct cropping seasons: Rabi (mid-November to mid-March) and Kharif (mid-March to mid-November). The Kharif season is further subdivided into two parts namely Kharif-I (mid-March to mid-July) and Kharif-II (mid-July to mid-November). Common Kharif crops include Aus and Aman rice, jute, and summer vegetables, while the Rabi crops include Boro rice, wheat, potato, pulses, oilseeds, and winter vegetables.
2.2.1 Primary data. The primary data used in this study were collected using a disproportionate stratified random sampling technique to have a representative sample across all 7 climatic zones in Bangladesh (Arshad et al., 2016). The climatic zones shown in Figure 1 were used as a basis for stratification. Each climatic zone was treated as one stratum. Three administrative districts were selected from each stratum randomly. From each selected district, two Upazilas (lower administrative units) were randomly selected whereas one village from each Upazila and ten farm households from each village were subsequently selected randomly. This resulted in 60 farm households from each climatic zone, totaling 420 respondents across all climatic zones. The selected districts represent a broad degree of agroclimatic, socio-economic and geographic features of Bangladesh.
The data were collected through field surveys conducted between January 2017 and April 2017 considering the cropping seasons of 2015-16. The survey questionnaire was designed to collect detailed information on the socio-economic characteristics of the sampled households including basic household information (gender, farmer’s age, education, household size, etc.), and the farm characteristics which include farming experience, farm area, soil types, access to bank credit, distance to market, access to extension services, and irrigation facilities. We also collected information on farmers’ perceptions about climate change and changes in the local climatic patterns observed by farmers over the past decades. In addition, we collected information on current adaptation measures undertaken by the farmers in the study area (Table III). Net crop income per hectare is a core variable of Ricardian regression analysis. To estimate net crop income, farmers were asked in detail about their agriculture management practices, input and output costs. These include crop types, growing seasons of different crops, transportation and miscellaneous cost, cost and amount of inputs such as labor, irrigation, seeds, pesticide, fertilizer etc.
2.2.2 Secondary data. We collected 46 years (1971-2016) climate data on monthly average temperature and monthly average rainfall from the Bangladesh Meteorological Department, Dhaka. These climate data were interpolated from the respective meteorological stations located in the surveyed districts and climatic zones under the study. After testing a number of alternative definitions of seasons, dry (November-March) and wet (April-October) seasons were considered for the analysis. Due to the country’s geographical location and small areal extent, average winter (21.16°C) and summer (28.11°C) temperatures of Bangladesh do not vary significantly. The dry (Rabi) and wet (Kharif) seasons are also the coldest and the hottest seasons respectively.
2.3 Ricardian methodology The relationship between climate and agriculture is typically modeled using three approaches: crop growth simulation models, agro-economic models, and integrated assessment models such as computable general equilibrium and whole farm models (Mishra et al., 2015). The basis of these models is on climate-crop physiology and development. The first two modeling approaches can include some adaptation and crop management practices such as change of planting dates, variety selection, and fertilizer use. However, as both the modeling approaches are crop-specific, they are unable to account for other adaptation measures such as responses to economic stimuli including input replacement, price variations, crop shifting, and multi-cropping. Without accounting for such behavioral responses, those approaches can lead to an overestimation or underestimation of the climate impacts.
For the development of an unbiased estimate of climate impacts, there is a need for a whole farm approach that allows for adaptation responses. The model that is currently being widely used across different countries to measure the economic impacts of climate change in the agriculture sector is the “Ricardian model”, named after the work of economist David Ricardo (1772-1823). Mendelsohn et al. (1994) was the pioneer to develop this cross-sectional model. The basis of the model is on the observation that land rents capture the long-term productivity of the farm. The Ricardian model assesses the performance of farms across landscapes, capturing impacts of spatial variations in climate attributes and other factors including input prices, soils, and socio-economic factors where the value of the lands reflects the present value of the future stream of net farm income.
The main advantage of this model is that it accounts for adaptation because farmers tend to adapt to the climate where they live to maximize the outputs and farm incomes. Other important advantages of the model are: it does not require data over time which is very difficult and costly to collect, rather it needs data across geographic space where climate attributes vary; it is a flexible model as it allows to consider all major enterprise activities; and it is very ease to implement.
However, the Ricardian model has some limitations. The first drawback is the possibility of omitted variable bias, which is present in all cross-sectional analysis. Another concern is the inefficiencies of the land and labor prices/markets that may distort prices. It does not account for CO2 fertilization effects. Furthermore, the model cannot capture transition costs of an instantaneous adaptation to a new technology against climate. However, sudden adaptation to new technology is never experienced. Another controversial drawback is concerned with irrigation. It is also not an issue in the study area as maximum farms rely on irrigation.