2.1 Developing two IPCC future scenarios for the study site

In developing the CC scenarios, we considered both climatic variables and climate-induced biophysical variables. We chose air temperature, rainfall and sea level rise (SLR) as direct climatic variables as they are on the list of most widely applied direct climatic variables in the CC research, in addition to being relevant to our case studyFootnote ¹ (Maunder, Reference Maunder1962; McConnell, Reference McConnell1977; Mieczkowski, Reference Mieczkowski1985; Richardson and Loomis, Reference Richardson and Loomis2004; Whitehead et al., Reference Whitehead, Poulter, Dumas and Bin2009; Becken, Reference Becken2013).

Rainfall and air temperature climatic variables for the Rekawa baseline scenario and future scenarios are taken from the national wetland directory of Sri Lanka and Wijesekara (Reference Wijesekaraundated), respectively. Future temperature and rainfall values for the Rekawa site were adopted from the Hambantota district and maximum values were taken from the HadCM3 (Hadley Centre General Circulation Model- Version 3) model. Attribute levels for SLR for Rekawa were taken from prediction values for Sri Lanka for 2025 and 2050 (UNDP, 2012).

Given the touristic attractions of our study area, we selected three climate-induced biophysical variables: the number of turtle nesting sites, mangrove cover and area of beach inundation. The Rekawa coast in southern Sri Lanka is the prime nesting habitat of turtles in Sri Lanka (IUCN, 2005). The Rekawa beaches provide an ideal environment for turtle nesting (Ganewatta et al., Reference Ganewatta, Samaranayake, Samarakoon, White and Haywood1995) and every year thousands of turtles reach Rekawa beach to lay eggs (Rathnayake, Reference Rathnayake2016). The Rekawa mangroves support ecotourism by providing habitats for local and migratory birds and for many species of mammals and reptiles (Ganewatta et al., Reference Ganewatta, Samaranayake, Samarakoon, White and Haywood1995). The Rekawa lagoon-mangrove ecosystem stretches into a sandy beach, which is approximately 10 km in length (Gunawardena and Rowan, Reference Gunawarden and Rowan2005). Focus group discussions revealed that foreign tourists select Rekawa beach for relaxation due to its serene, beautiful and salubrious nature. While changes in direct climatic variables are already adopted for Rekawa based on prediction values for Sri Lanka, data do not exist for the climate-induced biophysical variables. Hence, we constructed estimates for the climate-induced biophysical variables in the two scenarios. We did this by taking into consideration the impact of SLR. First, we applied a standard survey and levelling technique to measure the present beach elevation. Next, using predicted levels of SLR for Rekawa, we predicted beach inundation for the two scenarios using Geographical Information System ArcView software. These predictions were made assuming that the current beach profile and topography would not change.

Finally, nesting site locations were marked using Global Positioning System. At the time of the preliminary investigation, we found 342 turtle nesting sites along the Rekawa coastal belt. Superimposing Google Earth satellite maps for the Rekawa coastal belt for several years, 55 nesting sites were identified as vulnerable to sea-wave actions. Based on this, a direct potential risk factor was developed, i.e., 16.08 per cent. Then, the indirect risk of losing turtle nesting sites was adopted from predictions based on beach inundation, i.e., 13 and 17 per cent for 2025 and 2050 respectively. Total potential risk of losing nesting sites for 2025 and 2050 was found to be 29 per cent and 33 per cent respectively, by summing up both direct and indirect risks. Hence, of a total of 342 nesting sites, there is a potential risk of losing approximately 100 and 113 nesting sites along Rekawa beach, leaving 242 and 229 sites for 2025 and 2050 respectively. From the pre-testing of the survey, we learned that respondents preferred round numbers. Thus, we rounded these numbers in a subjective manner, with the intention of making a clear difference in changing nesting sites in two scenarios. The number of nesting sites in 2025 and 2050 are rounded up to 250 and rounded down to 200 respectively, considering the adverse CC effects in the long term compared to the short term.

Changes in mangrove cover in Rekawa were predicted by adapting the mixed mangrove area reduction rate worked out by Jayatissa et al. (Reference Jayatissa, Guero, Hettiarachchi and Koedam2002) for the nearby Kalametiya Lagoon. It is reasonable to assume that this mangrove reduction is merely due to CCs, as felling mangrove trees around the lagoon is prohibited.

2.2 Developing a climate change induced environmental index

Climate change may have profound implications for tourists' visitation behavior due to the potential effects on comfort level. For example, CC may increase the average air temperature to a level above what most people perceive as comfortable. In order to capture the multidimensional nature of CC effects on human comfort level, changes in visitation behavior may be derived using a climate index, which is based on existing scenarios of CC (Rosselló-Nadal, Reference Rosselló-Nadal2014). Mieczkowski (Reference Mieczkowski1985) was one of the pioneers in the development of a climate index for tourism.

The development of Mieczkowski's Tourism Climate Index (TCI) originally encompassed twelve climatic variables, but five variables were omitted due to lack of meteorological data for some developing countries. The remaining seven variables – maximum daily temperature, mean daily temperature, minimum daily relative humidity, daily relative humidity, precipitation, daily duration of sunshine and wind speed – were used to create sub-indices of thermal comfort, daytime comfort, and daily comfort. These variables were allocated weights assuming their relative importance for tourists' wellbeing, to formulate the TCI. A common criticism of the existing climate indices for tourism, including Mieczkowski's TCI, is that assigning weights to various climate factors within the index is based on the researcher's subjective opinion and are not tested empirically against the preferences of tourists (de Freitas, Reference de Freitas2003; Gómez-Martin, Reference Gómez-Martín2006).

Mieczkowski's TCI was originally designed to evaluate the global climate with respect to tourism, but not to assess impacts on tourism from CC. However, it was later used to evaluate the attraction of tourism destinations under anticipated CC scenarios (Rosselló-Nadal, Reference Rosselló-Nadal2014). Morgan et al. (Reference Morgan, Gatell, Junyent, Micallef, Özhan and Williams2000) used Mieczkowski's TCI with a slight modification, to use it in a beach environment. Studies using the TCI to evaluate the attraction of tourism destinations under anticipated CC scenarios have been made for Europe (Moreno and Amelung, Reference Moreno and Amelung2009), North America (Scott et al., Reference Scott, McBoyle and Schwartzentruber2004), and some Mediterranean countries (Amelung and Viner, Reference Amelung and Viner2006).

So far, climate indices applied in tourism studies have been based solely on direct climatic variables. We propose an extension to capture both direct climatic and climate-induced biophysical variables. A main reason for using such a CCIEI in our study is the presence of strong correlations between climatic and climate-induced variables in the econometric model. By converting climatic and climate-induced variables into a common CC index in a systematic way, we avoid such correlations. The climatic and climate-induced variables were converted into the CCIEI by using weights given by participants in focus groups when asked which factors were more or less important when deciding to visit Rekawa.

In composing the CCIEI, we excluded the SLR variable, as this variable was used to estimate the indirect variables ‘changes in turtles’ nesting sites’ and ‘beach inundation’. For the remaining five direct and indirect climatic variables, we estimated the percentage change from the present situation to the scenario level. As all changes were of the same sign, i.e., a deterioration, we added up these changes to give a composite CC index. In order to test for the robustness of the CCIEI as a predictor of visitation behavior, we made various assumptions about the weights of each variable when constructing the CCIEI. First, we calculated the CCIEI without using weights, and next we calculated the CCIEI allocating equal weights (0.2) to each variable. These calculations were repeated allocating various weights to the variables. The weights were based on results from focus group interviews with tourists and expert opinion, but are ultimately subjective, as was also the case in Mieczkowski (Reference Mieczkowski1985). As focus groups revealed that foreign tourists especially come to Rekawa mainly for turtle watching, with beach recreation as the second most important activity, we allocated the highest weight to the turtle nesting site variable. We allocated the second highest weight to the air temperature and beach inundation variables. The remaining variables were allocated equal weights. Table 1 yields information of the CCIEI number for the various weight allocations.Footnote ²

Table 1.

Development of CCIEI

*Value used for CCIEI in data analysis.

2.3 Contingent behavior method

The contingent behavior method (CBM) is a survey-based methodology involving the construction of hypothetical changes in the environment and asking respondents about their intended behavior contingent on such changes. CBM has been used in valuing environmental resources for recreational purposes (Chase et al., Reference Chase, Lee, Schulze and Anderson1998; Eiswerth et al., Reference Eiswerth, Englin, Fadali and Shaw2000; Richardson and Loomis, Reference Richardson and Loomis2004; Richardson et al., Reference Richardson, Loomis and Weiler2006; Scott et al., Reference Scott, Jones and Konopek2007). Still, there is a growing demand for estimation of changes in consumer welfare from recreation due to changing environmental quality or management of natural resources (Grijalva et al., Reference Grijalva, Berrens, Bohara and Shaw2002). The seminal paper of Loomis (Reference Loomis1993) verified the reliability of CBM by implementing a test-retest procedure. Grijalva et al. (Reference Grijalva, Berrens, Bohara and Shaw2002) also tested the validity of CBM for outdoor rock-climbing demand and suggested that when a future project has implications beyond current and historical range, CBM data may be a useful supplement to revealed preferences data.

Two types of format have been used in CBM studies: reassessed contingent behavior and intended contingent behavior. When the reassessed contingent behavior format is used, respondents are asked to reassess their visitation behavior: that is, how they would have behaved in the past (i.e., number of visits they would have made) had hypothetical changes taken place (Simões et al., Reference Simões, Barata and Cruz2013). In the intended contingent behavior format, respondents are requested to indicate their intended future visitation behavior for proposed hypothetical changes. In this format, instead of reassessing their former behavior, respondents are asked to predict how they will behave under future proposed conditions (Christie et al., Reference Christie, Hanley and Hynes2007). According to Simões et al. (Reference Simões, Barata and Cruz2013), these two formats vary in terms of reference period considered for the contingent behavior question. In this study, the intended contingent visitation behavior format is used for two reasons. First, during the preliminary investigations we found that a majority of the foreign tourists are visiting this destination for the first time in their life, and thus it is difficult for them to reassess how they would have behaved regarding previous visits. Second, CC scenarios are predictions into the future, which in turn demands the use of forward-looking survey methods. We developed two climate scenarios and asked respondents whether the changes described by the scenarios would increase, decrease or not change their future visitation to Rekawa.

2.4 Data collection

Using two future climate scenarios, one based on IPCC predicted values for Sri Lanka for the year 2025 and one for the year 2050, we designed a survey asking people to answer a hypothetical question about destination visitation in the future under the two scenarios. Figure 2 shows how the scenarios were presented and the framing of the future visitation question.

Figure 2.

Choice card and contingent visitation question as used in the survey.

The timeframe of the scenarios was not explicitly mentioned as it may extend beyond the life span of many tourists to be interviewed. In addition, the survey encompassed demographic questions and questions on their concern about CC when selecting a destination and personal view on CC and its consequences for Rekawa.

The survey was pre-tested with tourists, tour guides and Rekawa community inhabitants, including staff members of a turtle conservation project. Based on their comments, modifications were made. The final survey was translated into the local language Sinhala, to be used for domestic tourists, whereas an English version of the survey was used for foreign tourists. We trained a team of five Sri Lankan graduate students in data collection. The training included how to approach potential respondents, briefly explain the objectives of the study and obtain their consent to participate. Face-to-face field interviews were made during December 2016 – February 2017, covering one of the peak seasons of tourists to this site. Weekends were selected for domestic tourists, while weekdays were used to reach foreign tourists.

Every second tourist or tourist group at each chosen sampling site was randomly selected for the interview. After the introduction, the respondent received the survey, and the future climate scenarios for Rekawa coastal wetland were explained. Next, the respondents were asked to answer the contingent visitation behavior question. Preliminary investigations revealed that this destination is visited once in a lifetime or every few years by the tourists, especially foreign tourists. Therefore, the contingent visitation behavior question was asked for a period of 5 years instead of number of annual trips. The data collection resulted in 365 completed questionnaires for data analysis, including 213 foreign and 152 domestic tourists. A summary of descriptive statistics for the respondents is provided in table 2. As can be seen, there are distinct differences in socio-demographic characteristics of domestic and foreign tourists. Each respondent answered two contingent visitation behavior questions, and thus the number of observations is 730.

Table 2.

Descriptive statistics for respondents divided on domestic (Sri Lankan) and foreign tourists

2.5 Two-stage regression using instrumental variables

The fact that domestic and foreign tourists differ when it comes to socio-demographic characteristics complicates the statistical analysis of relevant explanatory variables for change in visitation behavior under CC. Using both tourist type and other socio-demographic characteristics as predictors will cause problems in the form of multicollinearity, while excluding either tourist type or the socio-demographic characteristics will cause correlation between remaining predictors and the error term (Angrist and Krueger, Reference Angrist and Krueger2001). One method to solve for these problems simultaneously is to use instrumental variable estimation, which is one type of a two-stage regression procedure (Angrist and Krueger, Reference Angrist and Krueger2001). Applying the instrumental variable technique, the predictors possibly correlated with the error term must have at least one, but can have more, instrument. We use tourist type as a predictor, along with the CCIEI and attitudes toward climate change impacts (CCimp). The latter is strongly correlated with tourist type because a large share of foreign tourists answer ‘don't know’ to the question of whether CC impacts are visible at Rekawa. Eliminating the ‘don't know’ answers, foreign and domestic tourists are relatively similar when it comes to attitudes toward CCimps at Rekawa. Hence, in a reduced model, eliminating the ‘don't know’ answers, we use CCimp as a predictor.

The model we apply is thus the following:

(1)\begin{align}\textrm{VisC}{\textrm{h}_i} = \textrm{AS}{\textrm{C}_1} + \; {b_1}\textrm{CCIE}{\textrm{I}_{ij}} + {b_2}\textrm{CCim}{\textrm{p}_i} + {b_3}\textrm{T}{\textrm{T}_i} + \varepsilon \vert \; \textrm{se}{\textrm{x}_i} + \textrm{ag}{\textrm{e}_i} + \textrm{ed}{\textrm{u}_i} + \textrm{oc}{\textrm{u}_i},\end{align}

where ASC_i, i = 1, 2 are alternative-specific constants, $\varepsilon$ and ∈ are error terms respectively, and ${b_1} - {b_3}$ are parameters to be estimated. Subscript i is a running index for the respondent and subscript j, $j\in 1,2$, indicates the scenario (scenario 1 or 2). The CCIEI is the CC environmental index, as explained above. The predictor CCimp is a measure of how respondents perceive CC effects in Rekawa. It takes the value 0 if the respondent indicates (s)he doesn't know, the value 1 if the respondent doesn't think CC takes place in Rekawa, the value 2 if the respondent thinks CC takes place in Rekawa but that it is not visible, and the value 3 if the respondent thinks that CC takes place in Rekawa and that it is visible. The tourist type variable (TT) takes the value 1 for foreign tourists and 0 for domestic tourists.

To determine the effects of socio-demographic characteristics on changes in visitation behavior, which works through the tourist type variable, we estimate the following model:

(2)\begin{equation}\textrm{T}{\textrm{T}_i} = \textrm{AS}{\textrm{C}_2} + {\beta _1}\textrm{Gende}{\textrm{r}_i} + {\beta _2}\textrm{Ag}{\textrm{e}_i} + {\beta _3}\textrm{Ed}{\textrm{u}_i} + {\beta _4}\textrm{Oc}{\textrm{u}_i} + \in.\end{equation}

The gender variable takes the value 1 if the respondent is a woman and 0 if a man. Education level of the respondent is coded as 1 for no formal education, 2 for primary education, 3 for secondary education, 4 for technical diploma, 5 for bachelor's degree, and 6 for postgraduate degree. The occupation variable takes the value 1 if the respondent has paid work and 0 if not. Only age is a numeric variable indicating the actual age of the respondent in number of years. The rest are categorical and binary (dummy) variables. ASC₂ is an alternative specific constant, and ${\beta _1} - {\beta _4}$ are coefficients to be estimated.