Evaluating the Public’s Preferences toward Sustainable Planning under Climate and Land Use Change in Forest Parks

Abstract

Balancing the goals of sustainable planning under climate and land use change (CLUC) with ecosystem service functions is a huge challenge for the management and programming of protected areas today. We construct a new evaluation framework towards the perspectives of sustainable land management based on the choice experiment (CE) model, and apply it to investigate the public’s preferences for the forest parks in Taiwan. This study found that implementing organic farming, increasing species populations, increasing the acreage of secondary forest area, and developing an integrated framework for ecotourism would best satisfy the public’s preferences for sustainable land use management. Second, we identify that the heterogeneity of the public’s preferences for forest park management varies depending on whether individuals are (1) members of environmental groups, (2) agricultural landowners, and (3) residents of the municipality. Third, we find integrated land use programs generate the highest welfare values among scenarios comprising different financial attributes.

1. Introduction

Extreme climate conditions and rapid changes in land use and development have created a crisis for land resources around the globe [1,2]. Land resources are key elements for human society and development [3]. However, as land use change alters landscapes and vegetation, there are knock-on effects for ecosystem functions and ecosystem services (ES) [3]. Moreover, climate system changes occur in response to greenhouse gas (GHG) fluxes, which relate closely to land use decisions [4,5]. Bateman et al. [4] indicated that land use change not only affect agricultural production, but also are simultaneously linked to greenhouse gas emissions, sustainable tourism development, preservation of green space and biodiversity. In other words, land use change is intimately related to both economic development and the natural environment [3,6,7]. Moreover, land use change has been shown to play a major role in studies on global climate and environmental change [3,8,9,10]. Thus, land use management can be conceptualized as a human social issue that is closely linked to natural, ecological, and environmental phenomena. It therefore plays a major role in protecting ES [3,4].

Based on the concept of sustainability, sustainable land management (SLM) is often a preferable strategy for reducing conflicts or negative impacts in the context of sustainable regional development [11,12,13]. Embodying comprehensive perspectives, SLM integrates the attributes of resources and the environment, economics, biodiversity, and social welfare [11,14]. SLM has been defined as the adaptation to land use systems that promotes the maximization of economic and social welfare benefits for land users, based on the functions of ES from land resources [11]. However, SLM also morphs with changes in economic, societal, technological, and environmental conditions, as well as changes in environmental governance, especially when such changes are climate- and land use-driven [2]. Accordingly, land use policy involving changes in patterns of land use, and the associated welfare evaluations of development and conservation, have become issues of urgent public concern [15]. Moreover, Bateman et al. [4] pointed out that land use policy ought to integrate all possible ecosystem service functions in considering the economic value derived by reducing the potential impacts from CLUC. To sum up, the economic evaluation of SLM in forest parks is an avenue through which we can assess the extent to which land use serves as a foundation for SLM and policy making.

A review of the literature on sustainable land use programs includes studies of public perspectives on wetland ES in a Greek case study [16], estimation of the economic value of environmental goods [17], comparison of locals’ preferences for different land use scenarios in Spain [18], testing of preference heterogeneity for biodiversity in the United Kingdom [19], estimation of perspectives on the impact of climate change for evaluating the welfare effects of under land use change in a protected area [4], capturing the public’s preferences for conservation in a National Reserve in Chile [20], evaluation of the public’s perspectives on land use changes in Belgium [15], nonmarket valuation for hypothetical scenarios related to ES [21], estimation of welfare values of alternative land management schemes [22], estimation of welfare values associated with ES in a forest park [23], and the landscape transformations in landscape Parks [24]. Following the line of research and discussions above, we identified a shortage of related research that (1) focuses on the development of a conceptual framework for SLM with integrated attributes under climate and land use change in forest parks and (2) estimates the public’s preferences regarding integrative perspectives and ES under SLM for forest parks.

To fill this gap, our study establishes a conceptual framework for SLM under CLUC, based on the relevant theoretical considerations, which informs the choice of attributes and corresponding levels of nature conservation and social welfare. Then, we test the heterogeneity of the public’s preferences across demographically differentiated groups by examining respondents’ perspectives on sustainable land use management. Next, this study evaluates the welfare effects under sustainable land use management scenarios in the forest parks. We then provide the backgrounds of three forest parks in Taiwan, as well as a summary overview of the theoretical thinking on SLM. Furthermore, we build-up the design attributes and levels, construct the preference model for evaluating the residents’ preferences based on the CE methodology, and devise the hypothetical scenarios. In the fourth section, we summarize the results from our estimation of the public’s preferences, estimate the heterogeneity of these preferences across demographic groups, and explore the welfare effects under different scenarios for SLM. Finally, this study concludes with our explication of the main research contribution and our summary the implications for management programs and policy-making, with specific reference to the conceptual framework of sustainable land use management proposed herein.

2. Study Area

2.1. Forest Parks in Taiwan

In 2002, the Taiwanese Ministry of Agriculture initiated the policy known as the “Plain Landscape Afforestation Program” (PLAP) with the afforestation of agricultural land [25] to meet the goals of the World Trade Organization (WTO) regarding the functions of ES [26]. In the same period, Taiwan’s government established afforestation parks with the express functions of natural resource preservation, reduction of carbon emissions, and promotion of sustainable tourism to mitigate CLUC [26]. In 2007, Taiwan’s government established the “Extension of the Afforestation Policy” as part of a scheme known as the “i-Taiwan 12 projects”, in order to develop three forest parks [26,27] and to promote sustainable land use management and community-based tourism (CBT) in these forest parks [23,25,26]. The forest management authorities were aiming to transition toward adaptive forest management [23,26], which meant understanding resident’s perspectives and striving to meet the public’s changing expectations associated with a growing level of environmental awareness, under sustainable forest management in the forest parks [23,25]. The established forest parks brought tangible benefits to the public, such as environmental education and the opportunity to discover local communities and participate in recreational and experiential activities under the functional rubric of CBT [27,28].

This study chose these forest parks as its research scope. From 2011 to 2014, three large afforestation forest parks were established in Taiwan as part of the “i-Taiwan 12 projects” plan (Figure 1) [23,28]. The first was the Danongdafu Forest Park (DFP), located in Hualien County and covering 1250 hectares. In addition to maintaining the forest landscape, developing traditional aboriginal or organic agriculture and environmental remediation were the identified management priorities for DFP. The second was the Aogu Wetland Forest Park (AWFP), located in Chiayi County and covering 1470 hectares. Due to land subsidence and salinization, the AWFP is not suitable for agriculture. Therefore, maintaining the coastal wetland landscape, which is a rich habitat for winter migratory birds, as well as developing environmental education programs that cover the dynamic processes of the land system are the main purposes of AWFP management. The third was Linhousilin Forest Park (LFP), located in Pingtung County and covering 1005 hectares. The LFP not only features afforestation and agricultural landscapes, but also contains an old railroad line with remnants of tracks and stations, as well as the Er-Feng irrigation system that makes it suitable for water activities. It is also close to the Neishe trailhead of the “Kunlun Pass Trail”, and therefore developing a low-altitude forest to reconstitute the natural environment of the Dawu Mountains and the plains below them was the stated goal of LFP management. As noted previously, land use in the study areas has changed substantially over the 1995 to 2015 period due to the “Afforestation Policy in the Plain Area” (Figure 1). Therefore, a total of 2562.6 hectares (77.7%) of agricultural land was converted to forestry land in the period from 1995 to 2015, among which afforestation accounted for 98.9%. The largest converted areas, in decreasing order, were the DFP covering 1032 hectares (Figure 1), the AWFP (772 hectares), and the LFP (758 hectares). In the next section, we will outline the conceptual framework for SLM based on the background of these three forest parks.

2.2. Conducting the Theoretical Thinking on SLM

The SLM perspective on protected areas includes both “nature conservation” and “social welfare” [15,16,18,20,23,27,29,30,31,32,33]. The concept of biodiversity, which refers to the variety of plant and animal species [34], has been identified not only as an important attribute of nature conservation [23], but as one which reflects the value derived from natural land use in protected areas [15], including wetland management [16] and forest parks [23]. Thus, managers consider biodiversity as a core element of sustainable land use management. From the standpoint of nature conservation, land use type focuses on the various modes of land use in a protected area [15]. The concept has figured prominently in studies of the welfare evaluation of nature conservation under climate change [33], land use scenarios for a semiwatershed [18], lowland plantation land use with ecosystem modeling [25], and local’s preferences towards different land use types [23]. Thus, we identified land use type as an appropriate attribute for SLM in this study.

Compared with nature conservation, social welfare is concerned more with human well-being, and, specifically, with the publics’ preferences and perspectives relating to the management and of provision of both cultural and ES in protected areas [18,23,27]. As the literature shows, social welfare is also closely related to phenomena such as farming methods [29,30,31,35] and modes of ecotourism [18,27,34]. Human society benefitted from agricultural production [30]. Thus, it’s important to know the residents’ perception for natural protection and the food safety [30,36], we can expect to see a commensurate enhancement of agroforestry ES. The aforementioned “farming methods” in this context include both conventional farming and organic farming. The mode of intensive production was the typical agricultural mode in Taiwan [31]. Therefore, this study treats the levels of “conventional farming” and “organic farming” that represent the attribute of agricultural methods for the SLM in the forest parks. Secondly, the “ecotourism mode” gives the alternative recreation program under the aspects of lower impacts, specific customs, and cultural tourism [37]. The tourists would prefer on integrating travel information on website and travel center [34], and would like to have a community-based ecotourism in the protected area [27]. Thus, we integrate the ecotourism mode into the framework of SLM.

Finally, this study uses an environmental trust fund as the financial attribute, and sets two different payment vehicles (i.e., a donation and a tax) for evaluating the public’s preferences toward improving the SLM in the forest parks. First, a donation can indicate people’s preference for conservation of a protected area [23]. Second, an annual tax also treat as a financial attribute to access the public’s perspectives on natural’s conservation [15,18]. The financial attribute is the key factor for dissuading respondents’ preference for valuing public goods, and we evaluate the land use attributes and the marginal effects by using willingness to donate (WTD) and willingness to pay a tax (WTPT). The evidence uncovered by these studies prompted us to integrate this mode of investigation with the theoretical thinking in our study. The following attributes were selected for inclusion and analysis in this research; farming method, biodiversity, land use type, ecotourism mode, WTD, and WTPT.

3. Research Framework

3.1. The Attribute Design For Measuring the Publics’ Preferences

We selected the attributes and determined the corresponding levels based on a literature review encompassing protected areas, land use management, forest park management, and CLUC impact. Second, we conducted on-site interviews and focus group discussions (FGDs) with members of non-governmental organizations, managers of forest parks, and economists specializing in nonmarket evaluation [18,23,38]. Based on suggestions and feedback from the FGDs and the literature review, we chose the following attributes of this study; farming method [29,30,31,35], biodiversity [15,16,20,39], land use type [15,18,32,33,39,40,41], ecotourism mode [18,27,34,37], WTD, and WTPT [15,18,23,39,42]. The key factors of SLM are summarized in

Table 1. Variables for measuring public preferences for sustainable land management (SLM) in forest parks.

Attributes		Levels	Variable Name
Farming Method (FARM)		a. Current situation (conventional farming)	FARM±
		b. Organic farming	FARM+
Biodiversity (BIO)		a. Current situation	BIO±
		b. Increase in species populations	BIO+
Land Use Type (LAND)		a. Status quo (artificial & mixed forest)	LAND±
		b. Increasing secondary forest area	LAND1
		c. Increasing ethno-botany area	LAND2
		d. Increasing wetland area	LAND3
Ecotourism Mode (TOUR)		a. Current situation (individual tourism)	TOUR±
		b. Integrated framework for ecotourism	TOUR+
Welfare foundation	(a) Willingness to donate (WTD)	a. Current situation (no payment)	FUND
		b. NTD 1000/family/year
		c. NTD 2000/family/year
		d. NTD 3000/family/year
		e. NTD 4000/family/year
	(b) Willingness to pay a tax (WTPT)	a. Status quo (no payment)	TAX
		b. 1%/family/year
		c. 2%/family/year
		d. 3%/family/year
		e. 4%/family/year

* FARM: farming method; FARM+: organic farming; BIO: biodiversity; BIO+: increase in species populations; LAND: land use type; LAND1–3: each alternative level of land use type; TOUR: ecotourism mode; WTD: willingness to donate; WTPT: willingness to pay a tax; NTD: New Taiwan Dollars; the unit is NTD per family and US$1 equals NTD32.

.

3.2. The Questionnaire Design of SLM

Following the established procedures of the CE method [23,34], the first part of the questionnaire pertains to the attributes and behavior of respondents in the forest park, while the third part focuses on the respondents’ demographics [38]. From a standpoint of gleaning data on respondents’ perspectives, the second part of the questionnaire is the most important, as it directly relates to the publics’ preferences for potential sustainable land use management scenarios by presenting options in the form of choice sets composed of attributes and levels (Table 1).

The number of attributes and corresponding levels gives rise to 160 potential profiles (2 × 2 × 4 × 2 × 5 = 160). To develop the CE questions to probe the public’s preferences for SLM, we employed the orthogonal main effect design (OMED), which is an approach advocated in past empirical studies [23,34,38,43]. The OMED allowed us to reduce the number of profiles to 25 alternatives [44]; after we deleted unreasonable combinations, we arrived at a total of 13 choices when the procedure had been repeated three times [34]. To illustrate our design, we provide an example of the CE questions based on a SLM (see Figure 2 and Figure 3). Each alternative is composed of the same SLM attributes, albeit with the specific levels and conditions for each potential SLM (i.e., attributes levels) varying according to the descriptions.

3.3. The Preference Model

The CE model can reduce the hypothetical and yes-saying bias relative to the contingent valuation method (CVM) [23,38,45], and focuses on capturing individuals’ preferences with a utility function based on attributes and levels [45]. The random parameter logit (RPL) can help researchers to understand respondent’s preference for a policy change, and evaluates the welfare values with level changes [46]. The RPL model can test the individual’s preference of heterogeneity for nonmarket goods [47,48]. Relatively, the latent class model (LCM) can the capture the individual’s preference regarding to their number of groups of individuals with the attributes and levels. These segments, however, differ substantially in their preference structures [48], and the heterogeneity of preference analyzed for different groups with their demographic simultaneously [23,27,34,38]. Thus, preferences are assumed not to be homogenous across groups. The groups can be used to estimate the membership probability. It is suitable to clarify systematic causes of taste variation in a single framework, rather than via the RPL model alone [34]. Thus, we can test the public’s preferences for SLM with the RPL model, and use the LCM to reveal the public’s preference heterogeneity across respondents’ demographics with respect to the selected attributes and levels [38,48].

Based on the experiment design [23,27,38,43] encompassing the attributes and levels for the sustainable land use management (Table 1), the preference function of our study can be represented with the following equation.

$$V_{ij}=ASC+{\beta }_{1}FAR{M}_j+{\beta }_{2}BI{O}_j+{\beta }_{3}LAND{1}_j+{\beta }_{4}LAND{2}_j+{\beta }_{5}LAND{3}_j+{\beta }_{6}TOU{R}_j+{\beta }_{7}FUN{D}_j$$

(1)

$$V_{ij}=ASC+{\beta }_{8}FAR{M}_j+{\beta }_{9}BI{O}_j+{\beta }_{10}LAND{1}_j+{\beta }_{11}LAND{2}_j+{\beta }_{12}LAND{3}_j+{\beta }_{13}TOU{R}_j+{\beta }_{14}TA{X}_j$$

(2)

In Equations (1) and (2), $V_{ij}$ is a utility function for SLM, related to alternative j and individual i. The alternative specific constant ($ASC$) reveals the preference for alternative program or not, and can be treated as a dummy variable. Where ${\beta }_1$ and ${\beta }_8$ are the coefficients for the type of agricultural farming; $FAR{M}_j$ is the attribute for type of agricultural farming with 2 levels; ${\beta }_2$ and ${\beta }_9$ are the coefficients of biodiversity; $BI{O}_j$ is the biodiversity at level 2; ${\beta }_{3–5}$ and ${\beta }_{10–12}$ are the coefficients for land use type; $LAND{1}_j$, $LAND{2}_j$, and $LAND{3}_j$ at levels 2, 3, and 4, respectively; ${\beta }_6$ and ${\beta }_{13}$ are the coefficients of the mode of ecotourism; $TOU{R}_j$ represents ecotourism mode at level 2; ${\beta }_7$ and ${\beta }_{14}$ are the coefficients of the fund; and the variables $FUN{D}_j$ and $TA{X}_j$ represent the cost attribute.

All of the variables ($FARM$, $BIO$, $LAND1$, $LAND2$, $LAND3$, and $TOUR$) are dummy variables. $FARM$ takes the value 1 if the alternative “implement organic farming” is chosen; $BIO$ takes the value 1 if the alternative “increase in species populations” is chosen; $LAND1$ takes the value 1 if the alternative “increase secondary forest area” is chosen; $LAND2$ takes the value 1 if the alternative “increase ethno-botany area” is chosen; $LAND3$ takes the value 1 if the alternative “increase wetland area” is chosen; and $TOUR$ takes the value 1 if the alternative “integrated framework for the ecotourism” is chosen.

We can evaluate the marginal willingness to donate (MWTD) and marginal willingness to pay a tax (MWTPT) with the preference of SLM [15]. The Equations (3) and (4) evaluated as the ratio of two parameters [49], and then revealed as following equations.

$$MWTD=-{\beta }_{attribute}/{\beta }_{FUND}$$

(3)

$$MWTPT=-{\beta }_{attribute}/{\beta }_{TAX}$$

(4)

where ${\beta }_{attribute}$ is the coefficient of each attribute, ${\beta }_{FUND}$ is the coefficient of the willingness to donate (WTD) [50], and ${\beta }_{TAX}$ is the coefficient of the willingness to pay a tax (WTPT).

3.4. Hypothetical Scenarios

The welfare evaluation of the CE model has been carried out in studies of land use programs in semiwatersheds [18], impact reduction scenarios in island tourism areas [40], multiple land use management in a forest park [23], and sustainable ecotourism scenarios in a national park [38]. We conducted multiple scenarios with the levels and attributes based on the estimation results. With the results of the RPL model, we evaluated the MWTD and MWTPT with Equations (3) and (4) for each SLM scenario, and compared the utility of any alternative option to the reference alternative (the status quo). Three strategies of SLM were identified as outcomes of the implementation of particular land use policies, they are as follows.

• Strategy I—Nature conservation: Devoted to increasing biodiversity and increasing the secondary forest area. For the farming method and mode of ecotourism, it retains its current situation.
• Strategy II—Social welfare: Devoted to implementing organic farming and integrated ecotourism. For biodiversity and land use type, it retains its current situation.
• Strategy III—Integrated land use programs: Set a mode of ecotourism, devoted to creating and increasing healthy environment areas via implementing organic farming, increasing biodiversity, and increasing secondary forest area.

3.5. Sample Design and Survey Method

Taiwan’s population aged 20 years or above was 18,738,629 people at the end of 2017. This study is based on the 95% confidence level, 3.5% estimation bias, and we assume that agreement and disagreement are the same for the land use program. We thus arrived at a total of 1068 samples. The sample also considered differences in administrative area (i.e., city and county), gender, and age (20–29 years old, 30–39 years old, 40–49 years old, 50–59 years old, age 60 and above) to reduce the survey bias.

The formal surveys were conducted from October 2016 to April 2017 in 12 cities and counties: Taipei City, New Taipei City, Taoyuan City, Hsinchu City and County, Miaoli County, Taichung City, Changhua County, Tainan City, Kaohsiung City, Yilan County, Hualien County, and Taitung County. A total of 927 respondents were sampled using the formal survey. The respondents were selected randomly (i.e., out of every ten people, one was randomly chosen to interview). The respondents were contacted by well-trained interviewers and asked to participate in face-to-face interviews at train stations or major crossroads. Explanations were provided by interviewers to give the local residents (respondents) an overview of potential SLM.

The questionnaire contained two categories of explanatory variables. The first comprises the respondent’s demographics, including whether or not the respondent has joined any environmental organization (D1), farmland ownership (D2), the household monthly income (D3), the urban–rural divide index (D4), and annual household income tax (D5). The last category is the respondent’s attitude toward the policy of afforestation and land use planning (D6). All of the explanatory variables are dummy variables. $\mathrm{D}1$ takes the value 1 if the respondent has joined an environmental organization; D2 takes the value 1 if the respondent owns farmland; D3 takes the value 1 if the monthly income of household is more than NTD 80,000; D4 takes the value 1 if respondent lives in the municipality; D5 takes the value 1 if the annual household income tax is more than NTD 30,000; and D6 takes the value 1 if the respondent agrees with continuing afforestation and land use planning. All these dummy variables (D1–D6) can apply to the topics of the publics’ preferences estimation (

Table 2. Preference estimation results for sustainable land use management under the WTD function.

Attributes	Willingness to Donate				Willingness to Donate
	Coefficient	t-Value	Std. Dev.		Coefficient	t-Value	Std. Dev.
			Coefficient	t-Value			Coefficient	t-Value
ASC	−1.00344	−4.50 ***	3.51399	14.07 ***	−1.35512	−5.20 ***	3.63932	12.33 ***
FARM	0.64339	11.69 ***	0.49549	4.4 ***	0.73794	11.02 ***	0.61775	4.93 ***
BIO	0.12895	2.78 ***	0.0191	0.17	0.15752	2.95 ***	0.03456	0.21
LAND1	0.27470	3.98 ***	0.47735	2.77 ***	0.35417	4.45 ***	0.5273	2.66 ***
LAND2	−0.03258	−0.41	0.0515	0.17	−0.01721	−0.19	0.01067	0.04
LAND3	−0.00045	−0.01	0.60747	3.9 ***	−0.03002	−0.35	0.67067	3.65 ***
TOUR	0.26628	5.28 ***	0.51589	4.79 ***	0.32191	5.41 ***	0.60595	4.91 ***
Interactions between attributes and other variables
FUND $\times$ D1	−	−	−	−	0.00047	2.23 **	0.00067	1.65 *
FUND $\times$ D2	−	−	−	−	0.00019	2.00 **	0.00021	0.99
FUND $\times$ D3	−	−	−	−	−0.00033	−3.33 ***	0.00018	0.58
FUND $\times$ D4	−	−	−	−	−0.00021	−1.94 *	0.0008	7.51 ***
FUND	−0.00051	−12.18 ***			−0.00056	−5.3 ***
Log-likelihood	1179.58 ***				1260.65 ***
Chi Squared	χ2 0.01(15) = 30.5779				χ2 0.01(23) = 41.6384
Inf. Cr. AIC	AIC = 4960.9; AIC/N = 1.784				AIC = 4895.8; AIC/N = 1.76

***, **, *: significance at 1%, 5%, and 10% levels.

and

Table 3. Preference estimate results for sustainable land use management under the WTPT function.

Attributes	Willingness to Pay the Tax				Willingness to Pay the Tax
	Coefficient	t-Value	Std. Dev.		Coefficient	t-Value	Std. Dev.
			Coefficient	t-Value			Coefficient	t-Value
ASC	−1.00575	−3.41 ***	5.53772	12.49 ***	−1.77296	−4.54 ***	5.83395	10.25 ***
FARM	0.62437	8.78 ***	0.84348	6.64 ***	0.782	7.25 ***	1.14563	6.17 ***
BIO	0.08985	1.66 *	0.01752	0.05	0.12173	1.73 *	0.25665	1.13
LAND1	0.17931	2.16 **	0.59655	3.31 ***	0.2821	2.53 **	0.73269	2.76 ***
LAND2	0.11151	1.19	0.18512	0.76	0.17463	1.49	0.26763	1.05
LAND3	−0.04647	−0.54	0.6224	3.31 ***	−0.1228	−1.15	0.66411	2.4 **
TOUR	0.25578	4.15 ***	0.70595	5.72 ***	0.33618	4.18 ***	0.86368	4.99 ***
Interactions between attributes and other variables
TAX $\times$ D1	-	-	-	-	46.2139	1.72 *	95.2335	2.59 ***
TAX $\times$ D2	-	-	-	-	18.5402	1.14	48.627	1.05
TAX $\times$ D4	-	-	-	-	−30.0938	−1.87 *	121.358	7.33 ***
TAX $\times$ D5	-	-	-	-	2.53302	0.17	58.7888	1.88 *
TAX	−0.63174	−11.35 ***			−1.04796	−5.97 ***
Log-likelihood	1496.76 ***				1578.94 ***
Chi Squared	χ2 0.01(15) = 30.5779				χ2 0.01(23) = 41.6384
Inf. Cr. AIC	AIC = 4643.7; AIC/N = 1.670				AIC = 4577.5; AIC/N = 1.646

***, **, *: significance at 1%, 5%, and 10% levels.

) and public’s heterogeneity of preference (

Table 4. Parameter estimates of the public’s preference heterogeneity on the MWTD function.

Attributes	Class 1			Class 2
	Coefficient	t-Value	MWTD (NTD/family/year)	Coefficient	t-Value	MWTD (NTD/family/year)
ASC	−1.16594	−1.86 *	-	−5.75257	−2.33 **	-
FARM	0.49973	8.89 ***	2173	−0.16249	−0.46	-
BIO	0.11356	2.31 **	494	1.69087	2.44 **	500
LAND1	0.31239	3.96 ***	1358	0.22003	0.73	-
LAND2	−0.04586	−0.47	-	−1.16492	−1.65 *	−345
LAND3	0.10884	1.44	-	−2.13392	−1.99 **	−631
TOUR	0.32417	5.79 ***	1409	−0.80559	−1.44	-
FUND	−0.00023	−5.25 ***	-	−0.00338	−2.84 ***	-
Probability	0.604			0.396
Class membership parameters
Constant	−0.05429	−0.17
D1	1.15898	3.54 ***
D2	0.33889	2.58 ***
D3	−0.39015	−3.16 ***
D4	0.33425	2.19 **
D6	0.20812	0.74
Log-likelihood Ratio	562.94 ***
Chi Square	χ2 0.01(22) = 40.2849
Inf. Cr. AIC	AIC = 5591.5; AIC/N = 2.011

***, **, *: significance at 1%, 5%, and 10% levels; MWTD: marginal willingness to donate, NT$/family/year.

and

Table 5. Parameter estimates of the public’s preference heterogeneity on the MWTPT function.

Attributes	Class 1			Class 2
	Coefficient	t-Value	MWTPT (%/family/year)	Coefficient	t-Value	MWTPT (%/family/year)
ASC	−5.87112	−2.64 ***	-	−1.40263	−1.53	-
FARM	−0.30254	−1.06	-	0.37933	6.93 ***	1.46
BIO	1.7774	2.81 ***	0.52	0.0213	0.42	-
LAND1	−0.07521	−0.19	-	0.20112	2.60 ***	0.77
LAND2	−1.50172	−2.46 **	−0.44	0.15896	1.63	-
LAND3	−1.81195	−1.90 *	−0.53	0.00342	0.04	-
TOUR	−0.94261	−1.80 *	−0.28	0.27508	4.85 ***	1.06
TAX	−3.39307	−3.23 ***	-	−0.2605	−6.10 ***	-
Probability	0.455			0.545
Class membership parameters
Constant	0.35302	1.08
D1	−0.6243	−2.43 **
D2	−0.37909	−3.02 ***
D4	−0.38417	−2.62 ***
D5	−0.26913	−2.31 **
D6	0.05153	0.20
Log-likelihood Ratio	483.47 ***
Chi Square	χ2 0.01(22) = 40.2849
Inf. Cr. AIC	AIC = 5671.0 AIC/N = 2.03

***, **, *: significance at 1%, 5%, and 10% levels; MWTD: marginal willingness to donate, NT$/family/year.

).

3.6. Characterization of Respondents

In this study, the respondents’ socioeconomic background information shows there was an almost even distribution in terms of the gender and age of respondents. Of the total, 443 were males (47.8%) and 484 were females (52.2%). The age breakdown for the total sample (including both genders) was 20–29 years old: 19.5%; 30–39 years old: 20.4%; 40–49 years old: 23.5%; 50–59 years old: 18.4%; and 60 years and above: 18.1%. In addition, 63.3% of respondents were married, 19% of respondents lived in the municipality, 56% of respondents had college level education, 32.1% of respondents owned farmland, and 93.6% of respondents had not previously joined environmental organizations. Furthermore, 20.3% of respondents’ household monthly incomes were less than NTD 40,000, while 20.7% were NTD 40,000 to NTD 60,000, and only 5.8% of respondents’ household monthly incomes were more than NTD 200,000. However, 58% of respondents’ annual household income tax was less than NTD 30,000. The respondents’ occupations were predominantly in service (23.5%), industry (13.4%), and expertise (10.8%). As for the respondents’ attitudes toward climate change and specific policy, 95% of respondents felt the climate is changing, 94.5% of respondents thought climate change will affect the forest ecosystem, and 95.8% of respondents thought the Taiwan government should formulate policy to reduce greenhouse gases. In addition, 72.9% of respondents had never visited forest parks, 92.9% of respondents were in agreement with establishing a specific agency to protect forest ecosystems, 94.7% of respondents in agreement with protecting the forest park by implementing forestation and land use planning, while 92.8% of respondents thought forest parks can help to reduce climate change and its economic impact on human society.

4. Empirical Results

4.1. The Publics’ Preferences Estimation

The result of the RPL model with WTD is shown in the first column of Table 2. Besides the attributes of LAND2 and LAND3, the other attributes are significant in the choice of a SLM strategy in the forest park. This means respondents prefer those strategies that would result in implementation of organic farming, an increase in populations of species, an increase in secondary forest area, and the development of an integrated framework for ecotourism. The sign of the fund attribute is negative, which reveals respondents’ willingness to donate a lower monetary amount. To sum up, the most important land use planning attributes are farming method, land use type, ecotourism mode, and biodiversity (ranked from most to least important). The negative and significant sign of the ASC coefficient of ASC reveals that the respondents prefer the alternative SLM. However, the fact that the standard deviations of the ASC, FARM, LAND1, LAND3, and TOUR coefficients are significant, confirming there is unobserved heterogeneity across respondents.

Next, the RPL model with interactions for WTD is reported in the second column of Table 2. The model that fits the data best includes the variables D1, D2, D3, and D4, which are included in the model as they have significant interactions with the fund attribute. The variables D1 and D2 have a significant and positive interaction with the fund attribute, indicating that respondents that are members of environmental groups or own farmland are both associated with a smaller disutility from donating compared to the other respondents. However, the variables D3 and D4 have a significant and negative interaction with the funding attribute. This means respondents having lower levels of household monthly income or respondents not living in the municipality are both associated with a smaller disutility from donating compared to the other respondents. Also, the fact that the standard deviation of the ASC, FARM, LAND1, LAND3, TOUR, D1, and D4 coefficients are significant, this revealed that the respondents have the heterogeneity of preference.

Furthermore, for WTPT, the result of the RPL is similar to that for WTD (see the first column of Table 3). This model shows respondents prefer those land use planning strategies that would result in implementation of organic farming, an increase in populations of species, an increase in secondary forest area, and development of an integrated framework for ecotourism, too. The sign of the tax attribute is also negative; this reveals respondents’ willingness to pay a lower level of tax. To sum up, the most important land use planning attributes are farming method, land use type, ecotourism mode, and biodiversity (ranked from most to least important). However, the fact that the standard deviation of the ASC, FARM, LAND1, LAND3, and TOUR coefficients is significant indicates the heterogeneity of preference across our respondents. Next, the RPL model with interactions for WTPT is reported in the second column of Table 3. The model that fits the data best includes the variables D1, D2, D4, and D5, although only D1 and D4 have a significant effect. The D1 is significant and has a positive interaction with the tax attribute, indicating that respondent’ membership in environmental groups is associated with an increase in tax resulting in a smaller disutility compared to the other respondents. However, the coefficient of D4 is significant and has a negative interaction with the tax attribute, which means respondents not living in the municipality are associated with an increase in tax resulting in a smaller disutility compared to the other city residents.

4.2. Testing Public’s Heterogeneity of Preference

In our study, the class membership characteristics include the variables D1, D2, D3, D4, and D6. However, the result of the LCM is similar to that of the RPL model with interactions. Only the variables D1, D2, D3, and D4 are significant vis-à-vis the MWTD. Table 4 shows the MWTD values of different groups. Based on the results of the LCM (see Table 4), we see that the two classes have different SLM preferences, since the coefficients of the attributes are not the same. For Class 1, the utility coefficients for increasing the ethno-botany area (LAND2) and wetland area (LAND3) attributes are insignificant determinants of choice; or in other words, the five other attributes are significant, namely ASC, FARM, BIO, LAND1, TOUR, and FUND. Then the class membership coefficients reveal that a respondent having joined an environmental organization, owning farmland, having lower household monthly income (≤$80,000 NTD), and the respondent who living in the municipality belongs to Class 1 (probability = 60.4%). In addition, for Class 1, the MWTD is highest for implementing organic farming (NT$2173/family/year), followed (in decreasing order) by developing an integrated ecotourism package (NTD 1409/family/year), increasing the secondary forest area (NTD 1358/family/year), and increasing the populations of species (NTD 494/family/year). For Class 2, the ASC, BIO, LAND2, LAND3, and FUND attributes are significant, and this will increase the likelihood that respondents in Class 2 will choose a forest park land use planning scenario with higher levels of these attributes (probability = 39.6%). Furthermore, for Class 2, the MWTD is highest for increasing the populations of species (NTD 500/family/year).

Turning to the other payment tool, Table 5 sets out the MWTPT values of the LCM analysis with two latent groups. In this study, the class membership characteristics include the variables D1, D2, D4, D5, and D6. However, only the variables D1, D2, D4, and D5 are significant vis-à-vis MWTPT. Based on the results of the LCM (see

Table 6. Welfare changes from three strategy of SLM under climate and land use change (CLUC).

Attributes	Strategic Scenario I: Nature Conservation Program	Strategic Scenario II: Social Welfare Program	Strategic Scenario III: Integrated Land Use Program
FARM	Current situation	Implement organic farming	Implement organic farming
BIO	Increase	Current situation	Increase
LAND	Increase secondary forest area	Current situation	Increase secondary forest area
TOUR	Current situation	Integrated ecotourism package	Integrated ecotourism package
FUND mean (95% CI)	910.99 (894.3~927.7)	1906.42 (1870.7~1942.2)	2817.41 (2776.5~2858.3)
TAX mean (95% CI)	0.38 (0.37~0.39)	1.08 (1.04~1.12)	1.46 (1.42~1.50)

Note: FUND: NT$/family/year; TAX: %/family/year; CI: confidence interval.

), we see the two classes have different SLM preferences, since the coefficients of the attributes are not the same. For Class 1, the utility coefficients for the implementation of organic farming (FARM) and increasing the secondary forest area (LAND1) attributes are insignificant determinants of choice; or in other words, the other five attributes are significant, namely ASC, BIO, LAND2, LAND3, TOUR, and TAX. The class membership coefficients reveal that a respondent not having joined any environmental organizations, not owning farmland, not living in the municipality, and having a lower annual household income tax (≤$30,000 NTD) belongs to Class 1 (probability = 45.5%). In addition, for Class 1, the MWTPT is highest for increasing the populations of species (0.52%/family/year), followed (in decreasing order) by developing an integrated ecotourism package (−0.28%/family/year), increasing the ethno-botany area (−0.44%/family/year), and increasing the wetland area (−0.53%/family/year). For Class 2, the FARM, LAND1, TOUR, and TAX attributes are significant, which means they will increase the likelihood that respondents in Class 2 will choose a SLM scenario with higher levels of these attributes (probability = 54.5%). For Class 2, the MWTPT is the highest for implementing organic farming (1.46%/family/year), followed (in decreasing order) by developing an integrated ecotourism package (1.06%/family/year) and increasing the secondary forest area (0.77% /family/year). This leads us to conclude that these two latent classes’ characteristics and their preferences of SLM are heterogeneous. We found that differences in these parameter estimates between the latent groups could be explained by reference to respondents’ heterogeneity, which is based, in turn, on their social background characteristics.

4.3. The Welfare Effects of Hypothetical Scenarios

To identify the welfare values of potential scenarios in the forest parks, we established the potential scenarios for SLM with the results of the RPL model (see Table 2 and Table 3). The welfare value results are presented in Table 6. Scenario III generated the highest welfare values for both FUND and TAX (confidence interval (CI) of welfare change at 95% for 2776.5 to 2858.3 NTD/year/resident, and 1.42% to 1.50%/year/household), followed by Scenario II (CI of welfare change at 95% for 1870.7 to 1942.2 NTD/year/resident and 1.04% to 1.12%/year/household). Comparatively, Scenario I (CI of welfare change at 95% for 894.3 to 927.7 NTD/year/resident and 0.37% to 0.39%/year/household) was the lowest welfare value scenario. Therefore, the best scenario would have to integrate implementation of organic farming, increasing biodiversity, increasing the secondary forest area, and an integrated ecotourism package. Our findings indicate that this type of integrated land use program would be the optimum program for future SLM in the forest parks.

5. Discussion

Nowadays, the public has realized how important ecosystem-based management based is for sustainable land use development under climate and land use change in the protected areas. We found that residents have a higher preference for the alternative mode of SLM in forest parks, compared to maintaining the current situation. This study indicated that residents had the highest preference for implementing organic farming in the forest parks, which is a measure that has been suggested by other case studies of environmental-friendly agriculture and subsidy programs for buffer zones [23,29,35]. The evidence presented herein points to the effectiveness of implementing organic farming in SLM in meeting stakeholders’ preferences. The second most preferred attribute is the development of an integrated framework for ecotourism, and the residents prefer an integrated mode of ecotourism incorporated into a potential SLM [27,34,35]. Thus, managers of protected areas might conduct the integrated ecotourism programs to enhance the recreational, in-depth, and educational activities for visitors to forest parks, under sustainable land use programs. Furthermore, the residents had a higher preference for increasing the acreage of the secondary forest area, which reveals that attribute land use type should be a main attribute of SLM [33,39,40]. While SLM serve to provide integrated ES for the public, the respondents showed a strong preference for increasing the populations of species. This outcome point to the importance of biodiversity conservation programs within (and in areas surrounding) protected areas, which has been discussed in other studies [15,20,39].

Moreover, we uncovered that heterogeneity of preferences exists between two identified latent classes, with regard to the function of WTD in the context of the potential SLM (Table 4). The first class comprised the largest proportion of all residents and exhibited a higher preference for implementing organic farming, increasing the populations of species, increasing the acreage of secondary forest, and developing and integrated framework for ecotourism. The second group comprised forty percent of the total sample and only showed a preference for increasing the populations of species. Comparing these segments, we found that the first one had a larger ratio of individuals who had previously joined environmental organizations, were agricultural landowners, and lived in the municipality. In contrast, the second group was made up of individuals who had not previously joined environmental organizations, were not agricultural landowners, and did not live in the municipality. We found that the public’s heterogeneity of preference is supported by other case studies of conservation values in a marine protected area [51], local’s preferences for landscape conservation [19], land use preferences for a semiwatershed [18], locals’ preferences for national park development [52], locals’ preferences toward land use planning [23], and ecotourism development in a national park [38]. Therefore, these insights promise to be useful for informing both perspectives of market segmentation and environmental education on SLM in the forest parks.

We built up three potential scenarios with integrated attributes of a sustainable land use program, based on the evaluation of welfare effects (Table 6). The best land use management scenario proved to be the integrated land use program, followed (in descending order) by the social welfare scenario and the nature conservation scenario. The optimum scenario for a sustainable land use program in the forest parks would therefore be one that implements organic farming, develops an integrated framework for ecotourism, increases the acreage of secondary forest, and establishes a biodiversity conservation program under the financial attributes of WTD and WTPT. The results of the integrated scenario generated the maximum welfare effects, and this evidence was also suggested by the case study from a land use program in southeastern Spain [18] and eastern Taiwan [23].

6. Conclusions

The ES of forest parks include multiple functions, such as healthy food provision, climate regulation, habitat support, and culture & education. Taiwan’s forest parks are facing both human activity impacts and climate change. Thus, it’s important for policy-makers responsible for protected areas to consider the integrative aspects of farming methods, biodiversity, land use type, integrated modes of ecotourism, and financial attributes (such as trust funds and taxes) under the goals of sustainable land use development. Furthermore, policy-makers should strive to align development with residents’ preferences and raise the maximum welfare values via the balanced integration of nature conservation and social welfare, when choosing among SLM strategies.

In this paper we investigated whether, and to whom, the public would have a high MWTP and MWTPT for the ES, and, in particular, for measures aimed at climate change mitigation, in three large public forests parks in Taiwan. This study focused on the public’s perceptions, for example, regarding the role of managers in forest park management, as well as the level of trust between members of society who could be ES users (i.e., the public) or ES providers (i.e., forest park managers). In summary, agroforestry ES are complex and exhibit strong responses to CLUC. This paper may serve as an input for forest managers to develop strategies for potential SLM programs with respect to the ES in the forest parks. Such programs should contain both social welfare and natural conservation features and be based on the key attributes identified herein and shown below in Figure 4.

Based on the aspect of ecosystem-based management in the protected area, future study can access the strategy of payments for ecosystem services (such as provisioning, supporting, and cultural functions) based on our evaluation framework and main attributes, and compare the perspectives under citizen’s demographics, land ownership, and environmental member ownership (Figure 4). Second, understanding publics’ preference heterogeneity towards the SLM under CLUC would help the managers of protected area to establish suitable public’s conservation and education program in the future. Third, evaluating the welfare effects under social welfare, natural conservation, and integrated land use scenario would benefits for policy making for future’s sustainable land use management and budget allocation in relative protected areas. Finally, rapid urbanization is a major factor in loss of green space and agricultural land in the peri-urban area [53], especially on developing country similar to Taiwan (i.e., Thailand and China), and then apply to the management of a metropolis and the centrality effect [54,55]. The researcher also can evaluate the welfare effects for recreational green spaces for SLM [56] under our framework.