Abstract: Mainstream tourist volume forecast methods generally rely on historical market data, not suitable for newly developed tourism area due to insufficient data accumulation. In view of this issue, this paper proposes an objective forecast model for newly developed rural tourism areas. It takes the characteristics of rural tourism areas into account and designs a novel modified gravity model-based method featured with destination tourism attraction variable, effective tourist origin set and parameter quantification methods, to forecast tourist volume indirectly. The empirical study results could well fit with actual tourist volume data, verifying the effectiveness of the proposed method.

Keywords: tourist volume forecast, forecast model, tourist market, modified gravity model, rural tourism, newly developed rural tourism area, destination tourism attraction, tourist origin, shortest road distance, AHP

 1. Introduction

Tourist volume is a key planning index for the development of tourism areas and tourist volume forecast constitutes a fundamental task in the planning, development and management of tourism areas (Liu et al., 2019). At the administrative level, an accurate forecast of tourist volume helps to improve the predictability of tourism decision-making (Huang et al., 2016), and provides a basis for local governments in planning tourist development strategies. At the economic level, it helps to effectively avoid resource waste and repeated construction on the tourism market due to oversupply, better clarify the investment and financing status, and the future development of tourism areas, as well as to more accurately estimate their economic benefits and timely give an early warning.

Mainstream tourist volume forecast methods adopt the historical tourist volume data of a tourism area as the basis to quantitatively analyse the development trend of tourist volume using theoretical tools(Jiao and Chen, 2019). However, given that most rural tourism areas in China are newly developed(Xinhua, 2017; Wang et al., 2013), the lack of historical market data makes it difficult to accurately forecast tourist volume, which may result in blindness in the development and investment promotion of rural tourism areas, restricting their continuous development. Some other methods usually adopt the expert forecast method, the subjectivity of which, however, limits its effectiveness and popularity. In view of this issue, this paper studies how to forecast tourist volume for newly developed rural tourism areas considering their characteristics and takes the newly developed Jiufengshan rural tourism area in China as a case study.

To this end, this paper proposes an objective tourist volume forecast model applicable to newly developed rural tourism areas based on a modified gravity model. This method does not rely on historical data, and gives full consideration to the characteristics of rural tourism areas in conducting an objective tourist volume forecast. In the rest of the paper, firstly, related literature is reviewed; then, a modified gravity model based method that is able to forecast without historical data is proposed in detail; finally, a case study in Jiufengshan rural tourism area in China is presented, followed by the conclusion section.

2. Literature Review

In this section, related works are reviewed from three aspects: tourist volume forecasting model, data in tourist volume forecast and gravity model. In addition, motivation of this paper and advantages of proposed model and approaches compared with existing works are analysed.

2.1 Tourist Volume Forecasting Models

Studies on tourist volume forecast can be traced back to the 1960s. Generally, there are both qualitative and quantitative methods. The Delphi method is a typical qualitative method, which is restricted by its disadvantages of strong subjectivity, low forecast accuracy and low reliability. Quantitative methods mainly include time series based and causal models, such as regression analysis, moving average model, grey correlation degree analysis, artificial nerve network analysis, and network attention-based forecast analysis and so on (Jiao and Chen, 2019).

Autoregressive (AR) model and its variations have been widely used in the literature. Bangwayo-Skeete and Skeete et al. use AR mixed-data sampling to tackle a common problem of prediction with data sampled at different frequencies in (Bangwayo-Skeete and Skeete, 2015). Factor augmented autoregressive and bridge models with social network and semantic variables are employed to forecast tourism demand(Fronzetti Colladon et al., 2019). Automated Neural Network Autoregressive (NNAR) algorithm with denoising is used in forecasting monthly tourism demand for ten European countries in (Silva et al., 2019). The logistic growth regression forecasting model shows superiority when compared to the benchmark seasonal autoregressive integrated moving average according to an empirical study in Las Vegas in (Chu, 2014). The multivariate singular spectrum analysis is used to forecast tourist arrivals from five continents to South Africa(Saayman and de Klerk, 2019). Recently, advanced machine learning algorithm, the deep learning(Law et al., 2019) is used in forecasting monthly Macau tourist arrival volumes. Specifically, the complex and nonlinear relationship between potentially tourism related search intensity indices and tourist arrival data series is implicitly built via a long short term memory (LSTM) recurrent neural network. Their empirical results demonstrated that the deep learning approach significantly outperforms traditional regression approaches.

With the progress of studies on tourist volume forecast, an increasing number of scholars are adopting combination models to optimize existing tourist volume forecast methods. To improve the prediction accuracy, The BP neural network and the ARIMA hybrid model is used to analyse China’s inbound tourism market(Lei and Chen, 2007). A new hybrid intelligent model called modular genetic-fuzzy forecasting system based on a combination of genetic fuzzy expert systems and data pre-processing is proposed in (Shahrabi et al., 2013). The AB@G integral model is proposed to analyse changes in tourist volume to the Jiuzhaigou Valley tourism area in (Liao and Ge, 2013). A combination of the linear and non-linear features of component models is introduced in (Wen et al., 2019). Differently, model selection scheme is introduced in (Akın, 2015) to predict monthly tourist arrival data to Turkey. Specifically, the author identifies the components of the given time series using structural time series modelling and constructs a decision tree for model selection from seasonal autoregressive integrated moving average, v-support vector regression and multi-layer perception type neural network models.

In view of various prediction models, it is necessary to make a comprehensive study on performance comparison and in-depth understanding of these typical models. By evaluating the performance of autoregressive moving average (ARIMA), exponential smoothing, neural networks, trigonometric Box-Cox ARMA trend seasonal, fractionalized ARIMA, singular spectrum analysis algorithms, moving average and weighted moving average on predicting tourism demand in selected European countries, Hassani et al. found that no single model can provide the best forecasts for any of the countries in the short-, medium- and long-run(Hassani et al., 2017). More recently, after reviewing 211 key papers on tourism demand forecasting published between 1968 and 2018, Song et al. revealed that i) forecasting models have grown more diversified and combined; ii) the accuracy of forecasting results has been improved and iii) there is no single method that performs well for all situations.

The above works confirm that in order to accurately predict tourism demand of newly developed rural tourism areas, we have to customize appropriate model and optimize model parameters, which is just the motivation of this paper. This paper proposes to use the gravity model for tourist volume forecasting. Importantly, different from existing gravity model based works, the gravity is modified according to the market characteristics of rural tourism areas.

2.2 Data for Tourist Volume Forecast

The input data is another key concern of tourist volume prediction. Most existing methods, such as regression analysis and moving average mentioned above, use the tourist market data of previous years as input and perform a trend analysis on future tourist volumes. For examples, monthly tourist arrival data(Akın, 2015) or tourism demand series(Silva et al., 2019; Chu, 2014) is widely used.

Driven by the fast development of Internet, an increasing number of studies are integrating network search data, such as Google trend and Baidu index, to forecast tourist volumes. Sun et al. analyzed the daily tourist volumes to tourism area based on tourists’ network attention(Sun et al., 2017). The keyword search information provided in Google Trends (such as food, housing, transportation, tourism, shopping, and entertainment) is useful for analyzing tourism demand in countries and cities(Önder, 2017). The effect of Google Trends on tourism demand forecasting in Germany is analyzed in (Björn and Stefan, 2019). A new indicator constructed from Google Trends’ search query time series data is proposed in (Bangwayo-Skeete and Skeete, 2015; Björn and Stefan, 2019). Similarly, user-generated content retrieved from online communities on the TripAdvisor travel forum is used in (Fronzetti Colladon et al., 2019).

Due to the lack of sufficient historical market data, it is difficult to apply the above methods that commonly rely on tourism demand series to newly developed tourism area. In addition, the limited web influence of newly developed rural tourism area also restricts the effectiveness of network search data based methods. This is the motivation of using gravity model: it requires relatively less data in this paper. The modified gravity model constructs the intensity of the spatial interaction between effective tourist origin and destination tourist origin set, which enables us to forecast tourism volume just using population, income level and other data that is easy to obtain.

2.3 Gravity Model

To study the impact of tourism on trade flows, Marrocu et al. provided an empirical and theoretical research to verify that tourism matters for international trade(Marrocu and Paci, 2013).

Based on a panel data set of bilateral tourism flows among 28 countries over the decade 1990-2000, a gravity framework is proposed to evaluate the importance of transport infrastructure in determining the tourism attractiveness of destinations(Khadaroo and Seetanah, 2008). The authors in (Galli et al., 2016) applied a gravity model to assess the relevance of transport infrastructure on the ability to attract tourist flows to destinations by air and tourism in Brazil. The gravitation model was used to calculate the number of multi-day inland travels and then to study what connection can be discovered between the theoretical tourism accessibility and actual domestic flows, revealing the characteristics of domestic tourist flows in Hungary (Tóth and Dávid, 2010). The gravity model was used to analyse the factors affecting the numbers of international tourists visiting Turkey from the top 11 in (Eryığıt et al., 2010). An eight-factor model (GDP per capita, tourism climate index, population of the originating country, tourism price index, distance, earthquake, neighbouring country, September 11 terrorist attacks) was effectively derived. A theoretical background to the gravity model in tourism demand is summarized in (Clive et al., 2014).

One of the main concern about gravity model is the lack of theoretical background, which has hindered its application in tourism. While gravity specification for international trade can be supported by Heckscher-Ohlin models (for example, gravity models are used to explain bilateral trade flows), there is no general understanding about the patterns of international tourism flows. In particular, since there is inherent gap between tourism flow and international trade flow, we need new theory to interpret the estimated parameters in the model for bilateral tourism flows(Morley et al., 2014). Therefore, researchers try to modify the original model or introduce more complex factors and mechanisms. In addition to distance, the set of explanatory variables including pull and push characteristics to assess their relative roles in determining the attractiveness of the destinations to tourists, in terms of income, density, accessibility, and natural, cultural and recreational attractions, is considered in gravity model in (Marrocu and Paci, 2013). Their results indicate the importance of spatial dependency of both origins and destinations for trips was commonly overlooked in previous gravity specification based works. Morley et al. proposed a theoretical background to the gravity model for bilateral tourism flows derived from the individual utility theory, showing how it is possible to get a gravity equation from the individual utility theory(Morley et al., 2014). Considering the fact that international tourism activity is examined without the aid of economic theory and gravity model estimations are not done via the appropriate estimation methods, the authors in (Kaplan, 2016) introduced the Poisson Pseudo Maximum Likelihood method in the gravity model, which accounts for the heteroscedasticity problem.

Nevertheless, the gravity model is still important for future tourism demand research. Morley et al. believe that fuelled by the success in international trade exercises, gravity models have re-emerged as a way for modelling tourism demand (Morley et al., 2014). According to the results in (Galli et al., 2016), it is noted that although the gravity models was neglected on tourist demand in the past decades, it is receiving attentions by the good empirical results of such model recently.

Note that the mentioned works above mainly focus on country, city or region level tourist volume forecast, but pay little attention on small scale target, i.e., rural tourism areas. Although rural tourism (Bernard and Elisabeth, 2015; Wilson et al., 2001) has developed fast in recent years, tourist volume forecast methods for rural tourism areas, especially newly developed ones, are rare. On the contrary, this paper introduces the gravity model to predict tourist volume on the tourism level. In addition, characteristics of newly developed rural tourism areas of China are incorporated by optimizing the model and using specified parameters settings.

3. Objective Tourist Volume Forecast Model

With regard to the difficulties encountered in forecasting tourist volumes for newly developed rural tourism areas, this section first adopts a modified gravity model to represent the attractiveness degree of tourism area to tourists, laying the foundation for establishing a forecast model that does not rely on massive historical data. After that, it gives full consideration to the characteristics of rural tourism areas summarized in the previous section, and proposes a complete objective tourist volume forecast model based on the attractiveness model. Finally, it puts forward a series of theoretical algorithms and offers some methods for the determination and quantification of the parameters in the model.

3.1 Basic Gravity Model

Since the 1940s, gravity models have been gradually applied in studies on tourism geography, and mainly used to measure the magnitude of the spatial interaction between two areas, and estimate the intensity of connections between tourist origin and destination. So far, they have seen mature application and are mainly adopted to estimate tourist volumes and international tourism economy and trade(Keum, 2010). The application of gravity models by scholars can mainly be summarized from three aspects: the analysis of the important factors influencing tourism economy(Wang et al., 2017); the exploration of the connections between two areas in terms of tourism economy; and the forecast of tourist volumes (Zhao and Yin, 2004; Huang et al., 2017). However, there are few studies on tourist volume forecast in China, and the majority of these take large-scale tourism areas (such as counties and provinces) as research target; in fact, they rarely adopt rural tourism areas as case study target.

The motivation of using gravity model to forecast tourist volume for newly developed tourism areas is that the tourist visitor number from an origin city can be represented by attractiveness index in the model, which can be calculated by analyzing social and economy factors rather than relying on historical tourist volume data. In 1946, Zipf and Stewart took the lead to propose a basic model:    

where  denotes the attractiveness index;  denote the population size of two cities (tourist origin and destination, respectively); and D denotes the distance between them(Zhang, 1989).

The above model only considers the population size of tourist origin and destination and the spatial distance between them; this compromises its reflection of complicated tourism phenomena, and greatly restricts its applications. At the beginning of the 1970s, based on gravity and potential models, British geographer Wilson deduced a new regional spatial interaction model following the principle of entropy maximization(Wilson, 2010). This model can effectively solve the breakpoint paradox in the Newton’s gravity model, and has received close attention from scholars. On the basis of this model, in 2012, Li et al(Li et al., 2012) introduced the spatial damping coefficient, the income elasticity coefficient and the attractiveness of destination parameter to construct a revised tourist attractiveness model:    

where  denotes the tourist origin;  denotes the destination;  denotes the intensity of the spatial interaction between tourist origin  and destination , defined in this paper as person-times;  denotes the attractiveness intensity of destination k;  denotes the population size of tourist origin;  denotes the income level of tourist origin ;  denotes the income elasticity of tourist origin ; β denotes the damping coefficient;  denotes the distance between tourist origin  and destination ; and  denotes the normalization factor.

3.2 Tourist Volume Forecast Model

Through a comprehensive analysis on existing gravity models, it can be found that there are three factors determining the establishment of a tourist volume forecast model: the attractiveness of the destination, the emissiveness of tourist origin and the spatial damping between tourist origin and destination. Based on the characteristics of rural tourism areas, this paper establishes an optimized gravity model from the following two aspects, that is, the attractiveness of the destination and the emissiveness of the tourist origin.

(1) Introduction of the system attractiveness index for rural tourism areas. The attractiveness level of a tourist destination is the key of its success (Charles R. Goeldner. , 2014). Existing attractiveness model introduced the attractiveness intensity of destination as an index, but neither specified the measurement method of this parameter, nor conducted any efficient tourist volume forecast. Unlike the case of traditional sightseeing tourism, most tourists visit rural tourism areas for leisure, relaxation and to have a cultural experience. Considering such travel motivations of tourists to rural tourism areas, tourism resources, tourism products, tourism facilities, tourism enterprises, and tourism policies and so on are important indices that determine the supply level of rural tourism areas. In this context, the traditional approach of substituting resource quality for destination attractiveness is undesirable. On that account, this paper takes into account the demand characteristics of tourists to rural tourism areas in defining the system attractiveness index.

(1) Determination of the tourist origin set and measurement scale of rural tourism areas. Attractiveness models are mainly used to measure the interaction between a destination and a single tourist origin. The tourist volume to a rural tourism area is the sum of tourists to all of its tourist markets. Li’s revised attractiveness model (Li et al., 2012) mainly measures the interaction between tourist origin and destination, but neither offers a comprehensive estimation of the tourist volume to a destination, nor defines the analysis method of tourist market structure. Ideally, forecasting the tourist volume based on a tourist market structure has to cover data from all related provinces and cities nationwide, which inevitably complicates the forecast model and introduces uncertain errors. Since rural tourism areas are mainly destinations for family-based, short-distance leisure travels on weekends and their visitors are mainly nearby urban residents, we argue that screening out real effective tourist origins through approximate simplification could help to strike a compromise between complexity and accuracy of forecast methods, as illustrated in Fig. 4. However, the adopted criterion for determining an effective tourist origin (or equivalently, strong and weak attractiveness) may depends on specific scenarios.

When introducing new explanatory variables into a model, the correlation among variables must be analyzed. In other words, upon introducing destination system attractiveness as a variable, this model must give due consideration to the close correlation between system attractiveness and spatial damping. In this case, distance is not only a key variable of spatial damping, but also a general factor deciding destination system attractiveness. With the increase of distance between tourist origin and destination, the attractiveness of a destination to tourists is gradually weakened, and the continuously increasing tourist flow to a destination can be guaranteed only when the destination system attractiveness outweighs the resistance caused by the increase of distanc(Wu and Yu, 2010). Thus, we do not consider the influence of distance in the evaluation of destination system attractiveness.

After introducing the destination system attractiveness index and determining the tourist origin set and measurement scale, this paper establishes a tourist volume forecast model based on the modified gravity model. The tourist volume to the rural tourism area denoted as , can be forecast by the following formula:

where  denotes the intensity of the spatial interaction between tourist origin  and destination , which is defined as person-times in this paper;  denotes the effective tourist origin set of a tourism area;  denotes the destination system attractiveness k;  denotes the population size of the tourist origin;  denotes the income level of the tourist origin ;  denotes the income elasticity of the tourist origin ; β denotes the damping coefficient;  denotes the passage distance between tourist origin  and destination ; and  denotes the normalization factor. According to (Liu, 2017), the attractiveness of tourism resources attenuates with the increase of distance, a fact that cannot be ignored in tourism studies. However, we argue that the degree of attenuation is also closely related to aspects such as the intensity of attractiveness of the tourist destination, the size of the tourist destination, and the means of transportation. Hence, the spatial damping coefficient is an index to measure the attenuation rate of a regional acting force, which is jointed determined by all the above-mentioned factors.

3.3 Quantification Method for Model Parameters

The tourist volume forecast model for rural tourism areas involves four variables: the population size of tourist origin , the income level of tourist origin, , the passage distance between tourist origin and destination,  and destination system attractiveness, .

(1) Population size  and income level of tourist orign

Given that the main visitors to a rural tourism area are nearby urban residents and the consideration for newly developed rural tourism area, it is inaccurate to adopt only the population size of the tourist origin to measure the tourist volume from a tourist origin. This is because the emissiveness of the tourist origin is related not only to population size, but also to the local economy level and so forth. In this paper, tourist origin information involves two aspects, i.e., population size and income level. More into detail, city scale is replaced by total population, and income level by per capita GDP.

(2) Passage distance  between tourist origin and destination

Considering that tourist origins are measured at prefecture-level city or administrative district scale, the administration center of a tourist origin is selected as the starting point to measure its distance away from a destination (in km). The accessibility of a rural tourism area for tourists from a specific tourist origin is related to distance and to the degree of transportation convenience. Another key feature of  tourists to rural tourism areas is that most of them go there by means of self-driving. For this reason, geographical distance alone is insufficient to offer a full explanation of tourist volume. In view of the obvious differences between geographical distance and road distance (as the example shown in Fig. 5), this paper adopts road distance instead. Actually, the accurate road distance could be easily obtained from map APPs, such as “Amap”..

(3) Destination system attractiveness 

Given that tourists travel to rural tourism areas mainly for leisure and relaxation, traditional tourism areas focusing on landscape values can no longer fully satisfy tourist demands. Taking tourist demands into account, this paper establishes a relatively scientific evaluation system for destination system attractiveness. The system attractiveness index of a destination, used to explain its tourism supply level, is a comprehensive variable closely related to tourism resources, tourism attractions, infrastructure, service facilities, marketing promotion, brand image, and other factors. Motivated by previous works (Zhao et al., 2017), this paper defines the attractiveness index of rural tourism areas from several aspects, i.e., resource advantages, tourism environment, facilities and services, and brand influence. Fig. 7 shows the evaluation system of destination system attractiveness, where  is a function of resource attractiveness, tourism environment, facilities and services, and system influence, calculated by AHP.

4. Empirical Analysis

In this section, we conduct a case study in Jiufengshan rural tourism area to validate the proposed tourist volume forecast method.

4.1 Analysis of Market Characteristics

(1) Overview of Jiufengshan rural tourism area

Jiufengshan rural tourism area is located in the north of the Xingdian Sub-District, Pukou District, Nanjing, as shown in Fig. 6. It is also one of ten pearl villages in the beautifule village of Pukou District. 

 This rural tourism area is in the construction period, which started from 2018 and is expected to be completed in 2021. At present, only part tourism projects are put into market. So far, there is no complete market information, thus we try to use the modified gravity model to forecast the number of tourists. At present, the projects that have been put into market are Huhong Village, the Jiufeng Temple, the Shrimp & Crab museum, and the wetland park. At a later stage, to meet the leisure demands of tourists, it is expected to further launch more tourism projects, such as themed aquatic clusters, ecological nurseries, experience orchards, wetland parks, Hong folk customs village, parent-child camps, cultural creation streets, craftsmanship experience zones, aerial cafés, and outdoor libraries. As one of the “Ten Pearls” in the beautiful countryside of the Pukou District, the Jiufengshan rural tourism area will be continuously enhancing its tourism attractiveness, and becoming an important destination of leisure travels for the residents of Nanjing on weekends.

(4) Analysis of tourist markets

To collect tourist information, a questionnaire survey was conducted on the rural tourism areas adjacent to the Jiufengshan rural tourism area in the Pukou District, and on the relatively mature rural tourism areas in the Pukou District. Considering that weekends and holidays are peak periods for rural tourism, the survey covered as many weekends as possible. It was carried out in two periods, from June 8 to June 13, 2018, and from June 14 to June 24, 2018. The first period targeted two relatively mature rural tourism areas, that is, the Pearl Well Tourism area and the Yufa Ecological Garden; the second one focused on four rural tourism areas adjacent to the Jiufengshan rural tourism area, including the Lasting Appeal of Chu, the Xigeng Home of Lotuses, the Ink-water Ridge, and the Hometown of Educated Youths in Houchong. The survey distributed a total of 1,600 copies, 1,566 of which were valid, with a validity rate of 97.87%.

According to the questionnaire survey results, the visitors to this rural tourism area had two significant characteristics. Through the descriptive statistical analysis on questionnaire data using the SPSS software, it was found that the Jiangsu and Anhui Provinces contributed with the majority of visitors to the rural tourism areas of the Pukou District, Nanjing. The visitors from the Jiangsu Province accounted for 61.56% of the total number of visitors, while those from the Anhui Province accounted for 12.52%; moreover, each province had a market share of above 10%. In contrast, the Provinces of Zhejiang, Henna, Jiangxi, Sichuan, Guizhou, Shanghai, and other provinces (and municipalities directly under the central government) were minor tourist markets for the rural tourism areas of the Pukou District, each with a market share of less than 3%. More into detail, the Jiangsu and Anhui Provinces were major tourist markets for the Jiufengshan rural tourism area, claiming market shares of 83.10% and 16.90%, respectively. The cities involved in the Jiangsu Province (Fig. 8) included Nanjing, Xuzhou, Wuxi, Suzhou, Yancheng, Suqian, Lianyungang, Changzhou, Zhenjiang, Nantong, Yangzhou, and Taizhou. The cities involved in the Anhui Province (Fig. 9) included Chuzhou, Huai’an, Hefei, Xuancheng, Wuhu, Bengbu, Anqing, Ma’anshan, Fuyang, Huangshan, Chizhou, Huaibei, and Bozhou.

More results are summarized in the following, which is coincident with the general features of tourists to rural tourism areas mentioned in Section 3.

l The questionnaire survey inquired about the time taken for tourists to arrive at the rural tourism area investigated. Results show that more than 30.2% of tourists were less than one hours’ drive time away from the rural tourism area, while those less than two hours’ drive time away and those less than four hours’ ride away accounted for ＞52.0% and 72.7%, respectively. Clearly, the majority of tourists to rural tourism areas are on short-distance leisure travels, and tourist volume drops abruptly with the increase of travel distance. That is, the spatial damping coefficient is relatively large.

l Most tourists to rural tourism areas go there by means of self-driving. According to the questionnaire survey, among the tourists to the rural tourism area investigated, self-driving tourists accounted for 53.58% of the total tourist volume, while those traveling by package tour, by public transportation, and by other means of transportation accounted for 12.53%, 16.8%, and 4.3%, respectively. The transportation accessibility of a rural tourism area is a key factor influencing its tourist volume; it is closely related not only to the length of travel distance, but also to the degree of transportation convenience.

l Questionnaire survey results indicate that 73.3% of tourists traveled to rural tourism areas for stress management, while those wishing to enjoy the rural scenery, those intending to experience the local culture, and those expecting a relaxing vocation accounted for 56.4%, 27.9%, and 20.9%, respectively. Thus, in terms of tourism purposes, the attractiveness of a destination is a key factor determining the tourism market of a rural tourism area, and a critical index to forecast the tourist volume to the rural tourism area.

 4.2 Variable Determination and Parameter Estimation

The forecast model for the Jiufengshan rural tourism area mainly involves five variables: tourist origin and its social and economic information, attractiveness intensity of destination, spatial damping coefficient, income elasticity and distance between tourist origin and destination.

(1) Social and economic information of important tourist origins

Adopting the market comparison approach and the questionnaire method, this paper analyzes the characteristics of visitors to rural tourism areas from different tourist origins. According to the analysis results, the tourist supplies of the Jiangsu and Anhui Provinces both accounted for more than 10% of all tourists, while the tourist supply of each of the other provinces accounted for less than 3%. Hence, the Jiangsu and Anhui Provinces were effective tourist markets. In both the Jiangsu and Anhui Provinces, the tourist supply of each prefecture-level city was lower than 10%, except in the case of Nanjing, the tourist supply of which exceeded 50%. For the sake of the accuracy, we measured tourist supplies at administrative district scale in the case of Nanjing, and at prefecture-level city scale in the case of other provinces, and used the Statistical Yearbook of Nanjing (2015-2017) and the Statistical Communique on National Economic and Social Development (2015-2017) of various regions and cities as data basis. To better forecast tourist supplies, we also adopted basic data for the period 2015-2017 as support (Table 1 and Table 2), and acquired data for the period 2018-2025 through trend extrapolation. Considering the characteristics of rural tourism (i.e., short travel distance and short travel time) and the status of Nanjing as the primary tourist origin for the Jiufengshan rural tourism area, the social and economic statistics on Nanjing were made using district as unit, with the aim of ensuring the accuracy of tourist volume forecast.

(5) Passage distance between the Jiufengshan rural tourism area and tourist origins

 denotes the distance between tourist origin j and destination k, which is replaced by the actual shortest road distance travelled by tourists from a tourist origin to arrive at a destination. Amap is used to assess the actual road distance between the Jiufengshan rural tourism area and the center of each tourist origin, and acquire the data illustrated in Table 3, which show the means of transportation used by tourists to reach rural tourism areas, and the dominant status of self-driving.

(6) Destination system attractiveness index and data acquisition

In this paper, a three-layer attractiveness evaluation index system is established for rural tourism areas for four aspects: resource advantages; tourism environment; facilities and services; and brand influence. AHP is used to obtain the weights of various indices according to the ranking of various index factors by experts, based on importance degree. In the course of our study, based on planning documents and field survey, experts were invited to evaluate the development status, and assign scores to various index factors, thus eventually obtaining attractiveness intensity indices. From September 10, 2018 to September 30, 2018, ten experts from the tourism industry of Nanjing, five employees from the rural tourism areas of the Pukou District, and five employees from sub-district agricultural offices and tertiary industry offices, were invited to complete the weight assignment and the destination attractiveness scoring. Based on the scores assigned by the experts according to the tourism development status, and the scores assigned in the planning documents of the Jiufengshan rural tourism area, the development status of the Jiufengshan rural tourism area has a score of 5.4. It is estimated that, with the completion of all tourism projects, the continuous enrichment of tourism products, and the further perfection of tourism services and facilities according to the overall planning, its score will reach 8.3.

(7) Estimation of the spatial damping coefficient

Based on the population particle model, Wang (Wang and Jinghong, 1994)adopted a tourism area model method, obtaining a value of β=0.0046 according to the formula , and claimed that the damping coefficient of national population is equal to 0.0046. However, the accuracy of this value is uncertain because 19 years have passed since the proposing of this figure; moreover, visitors of rural tourism areas are mostly nearby urban residents on short-distance leisure travels. Hence, there exist differences between the spatial damping coefficient here, and that of general tourism area. With regard to this problem, based on the tourist volume and tourist origin information obtained from the questionnaire survey, this paper performs a partitioning and integration of the tourist volumes to a given destination K at different distance scales using the “tourist volume integral method”, and further estimates the spatial damping coefficient of the destination(Liu et al., 2016). According to the results in Fig. 11, when taking the whole Pukou District as tourist destination, the tourist volume within the scope of 100 km, 200km and 300km accounted for 48.02%, 54.16% and about 70%, respectively. The final result obtained through the tourist volume integral method is 0.019785.

(8) Income elasticity of demand

The overall “emissiveness” of a tourist origin not only is constrained by population size and per capita income level, but also has an important relation with income elasticity. Since visitors to the rural tourism area of the Jiufengshan rural tourism area mainly come from the Jiangsu and Anhui Provinces, and their income elasticity of demand has no correlation with tourist destinations. Thus, this paper assumes the spatial “homogeneity” of tourist origins, and adopts this figure.

4.3 Results Analysis and Verification

By substituting α=0.64, β=0.019785, and other statistical data into the calculation formula of the revised attractiveness model, the tourist volume to the Jiufengshan rural tourism area is calculated. Based on the calculated tourist volume data of 2016-2018, a trend extrapolation is performed to obtain the tourist volume data of 2025, as shown in Table 4. More into detail, the tourist volume to the Jiufengshan rural tourism area in 2018 was 158,000 person-times.

To verify the accuracy of the forecast data, we obtained the data of the tourist number in 2017 and 2018, that is the number of tourists after the opening of some tourism projects. As shown in Table 5, the rural tourism area received a total of 128,600 person-times in the whole 2017, and a total of 157,800 person-times in the whole year of 2018. Clearly, the forecast results of this paper fit well with the current development trend, thus verifying the reasonableness of the method used in this paper. As a result, we can also get the number of tourists after the completion of all the tourism projects. In 2021, the number of tourists will reach 395,000 and 852,000 in 2025.

5. Conclusion and Prospects

5.1 Conclusion

For newly developed and planned rural tourism areas, the forecast of tourist volumes constitutes the premises of scientific and effective management for tourism areas. However, due to the lack of historical data, tourist volume forecast has always been a challenge in this field. This paper establishes a modified gravity model, which introduces a destination system attractiveness index considering the characteristics of rural tourism areas and takes a market comparison approach to define an efficient tourist market set, providing a new idea to tackle this challenge. The findings are as follows:

(1) Based on the modified gravity model, this paper defines a tourist volume forecast model applicable to rural tourism areas, i.e., , where  denotes the total tourist volume from tourist origins;  denotes the tourist origin set of a tourism area;  denotes the attractiveness intensity of a tourism area, which is a function of resource attractiveness, tourism environment, facilities and services, and brand popularity;  denote the respective scores of the above-mentioned four factors. This paper uses AHP to calculate attractiveness intensity, i.e.,  denote their respective weights;  denotes the road distance between tourist origin ’s administration center and destination ;  denotes the population size of the tourist origin;  denotes the income level of the tourist origin ;  denotes the income elasticity of the tourist origin ; and β denotes the damping coefficient. The attractiveness model has given full consideration to aspects such as the destination system attractiveness, the emissiveness of tourist origin, and the spatial damping, and can comprehensively reflect the interaction between tourist origin and destination.

(2) Defining the tourist origin set constitutes a key point in tourist volume forecast. Unlike the approach taken by Liu to forecast the tourist volume to the Shanghai Disney Resort using residents from various provinces nationwide as main visitors(Liu et al., 2016), in review of the adjacency of tourists to rural tourism areas, this paper applies a market comparison approach and market survey to understand the tourist market structure and identify the tourist origin set.

(3) The measurement scale of tourist origins has a direct bearing on the accuracy of the tourist volume forecast. The Jiufengshan rural tourism area is located in Nanjing, the residents of which constitute the primary tourist origin for this rural tourism area. According to the proportion of tourist supplies of the Jiangsu and Anhui Provinces, this paper measures tourist supplies at administrative district scale in the case of Nanjing (that is, the separate calculation of index factors and parameters), and at prefecture-level city scale in the case of other provinces, thus clearly improving the accuracy of data.

(4) In the calculation of the destination system attractiveness index, referring to existing studies on the competitiveness of rural tourism and other index systems, this paper establishes a three-layer destination system attractiveness index system for rural tourism areas, and uses AHP for the assignment of weights. In the study, industry experts were invited to score the current and future attractiveness of tourism areas according to the present situation and planning documents. On one hand, this approach has fully taken into account the future development potential of rural tourism; on the other hand, it can obtain the attractiveness intensity score in a relatively objective manner.

(5) In calculating the damping coefficient, Zhao et al. (2004) directly adopted the national-scale figure of 0.0046 obtained by Wang using the tourism area model method. However, the assumption of the nationwide spatial “homogeneity” has to be satisfied before applying this figure. Considering that rural tourism areas mainly target nearby urban residents, this assumption does not stand here. Thus, this paper refers to the “travel volume integral method”, and adopts the “tourist volume integral method” to calculate the damping coefficient of 0.019785. Although this figure is clearly above the national average level, it can be reasonably explained when taking into consideration the characteristics of the tourists to rural tourism areas.

(6) Considering the fact that self-driving is the primary mean of transportation for tourists to rural tourism areas, the actual shortest road distance rather than the geographical distance is used as the passage distance index, which could improve the forecast accuracy.

(7) The forecast results in the case study of Jiufengshan rural tourism area fit well with the actual tourist volume data, verifying the effectiveness of the proposed method.

5.2 Prospect

This paper proposes a modified gravity model-based tourist volume forecast method suitable for newly developed rural tourism areas. Although the case study in Jiufengshan rural tourism area verified its effectiveness, the method can be further improved. On one hand, the damping coefficient in the modified gravity model is determined using the tourist volume integral method, which is fitted and calculated according to changes in tourist volume caused by the increase of the road distance from a destination. However, spatial damping is not only related to road distance, but also may have close connections with administrative divisions and transportation convenience. The influence of the latter two factors could be investigated in future works. On the other hand, due to limited resource and cost, this paper only takes the Jiufengshan rural tourism area as a case study. We argue that more case studies are needed to further polish the model and develop more universal index parameter calculation methods.

the proposed model and approaches As most Chinese cities and regions share common culture and have similar economy level, tourism developing stages and so on, both the motivation of rural tourists and the potential interaction between tourists and rural areas are similar. Of course, individual rural tourism areas possess its own characteristics. If these distinguishing characteristics are further considered, more customized models and accordingly better performance could be derived.

References

Akın M. (2015) A novel approach to model selection in tourism demand modeling. Tourism Management 48: 64-72.

Bangwayo-Skeete PF and Skeete RW. (2015) Can Google data improve the forecasting performance of tourist arrivals? Mixed-data sampling approach. Tourism Management 46: 454-464.

Bernard L and Elisabeth K. (2015) Rural tourism: the evolution of practice and research approaches towards a new generation concept? Journal of Sustainable Tourism 23: 1133-1156.

Björn B and Stefan L. (2019) Spurious patterns in Google Trends data: An analysis of the effects on tourism demand forecasting in Germany. Tourism Management 75: 1-12.

Charles R. Goeldner.  M. (2014) Tourism, Beijing, China: China Renmin University Press.

Chu F-L. (2014) Using a logistic growth regression model to forecast the demand for tourism in Las Vegas. Tourism Management Perspectives 12: 62-67.

Clive M, Jaume R and Maria S-G. (2014) Gravity models for tourism demand: theory and use. Annals of Tourism Research 48: 1-10.

Eryığıt M, Kotıl E and Eryığıt R. (2010) Factors affecting international tourism flows to Turkey: a gravity model approach. Tourism Economics 16: 585-595.

Fronzetti Colladon A, Guardabascio B and Innarella R. (2019) Using social network and semantic analysis to analyze online travel forums and forecast tourism demand. Decision Support Systems 123: 113075.

Galli P, Fraga C and de Sequeira Santos MP. (2016) Gravitational force exerted by Brazilian tourist destinations on foreign air travelers. Journal of Air Transport Management 55: 76-83.

Hassani H, Silva ES, Antonakakis N, et al. (2017) Forecasting accuracy evaluation of tourist arrivals. Annals of Tourism Research 63: 112-127.

Huang T, Xi J and Ge Q. (2017) Tourism spatial competition pattern of urban agglomeration under the influence of high-speed railway using spatial econometrics. Economic Geography 37: 182-191.

Huang X, Zhang L and Ding Y. (2016) Study on the predictive and relationship between tourist attractions and the baidu index: a case study of the forbidden city. Tourism Tribune 28: 93-100.

Jiao EX and Chen JL. (2019) Tourism forecasting: A review of methodological developments over the last decade. Tourism Economics 25: 469-492.

Kaplan FaA, Ali. (2016) The Turkey Tourism Demand: A Gravity Model.

Keum K. (2010) Tourism flows and trade theory: a panel data analysis with the gravity model. The Annals of Regional Science 44: 541–557.

Khadaroo J and Seetanah B. (2008) The role of transport infrastructure in international tourism development: A gravity model approach. Tourism Management 29: 831-840.

Law R, Li G, Fong DKC, et al. (2019) Tourism demand forecasting: A deep learning approach. Annals of Tourism Research 75: 410-423.

Lei L and Chen Y. (2007) Forecast of inbound tourists to China based on bp neural network and ARIMA combined model. Tourism Tribune 22: 20-25.

Li S, Wang Z and Zhong Z. (2012) Gravity model for tourism spatial interaction: basic form, parameter estimation, and applications. Acta Geographica Sinica 67: 526-544.

Liao Z and Ge P. (2013) Research on prediction of tourists' quantity in Jiuzhai valley based on AB@ G integration model. Tourism Tribune 28: 88-93.

Liu H. (2017) The distance attenuation rules of Guangdong inbound tourism flow's western towards diffuse. Economic Geography 32: 162-265.

Liu H, Liu Y, Wang Y, et al. (2019) Hot topics and emerging trends in tourism forecasting research: A scientometric review. Tourism Economics 25: 448-468.

Liu S, Tian J and Lu L. (2016) A case study of Shanghai Disneyland on spatial structure forecast for proposed scenic spot market: Modification and its application of gravity model. Acta Geographica Sinica 71: 304-321.

Marrocu E and Paci R. (2013) Different tourists to different destinations. Evidence from spatial interaction models. Tourism Management 39: 71-83.

Morley C, Rosselló J and Santana-Gallego M. (2014) Gravity models for tourism demand: theory and use. Annals of Tourism Research 48: 1-10.

Önder I. (2017) Forecasting tourism demand with Google trends: Accuracy comparison of countries versus cities. International Journal of Tourism Research 19: 648-660.

Saayman A and de Klerk J. (2019) Forecasting tourist arrivals using multivariate singular spectrum analysis. Tourism Economics 25: 330-354.

Shahrabi J, Hadavandi E and Asadi S. (2013) Developing a hybrid intelligent model for forecasting problems: Case study of tourism demand time series. Knowledge-Based Systems 43: 112-122.

Silva ES, Hassani H, Heravi S, et al. (2019) Forecasting tourism demand with denoised neural networks. Annals of Tourism Research 74: 134-154.

Sun Y, Zhang H and al e. (2017) Forecast of tourism flow volume of tourist attraction based on degree of tourist attention of travel network: A case study of baidu index of different clients. Human Geography 32: 152-160.

Tóth G and Dávid L. (2010) Tourism and accessibility: An integrated approach. Applied Geography 30: 666-677.

Wang L-e, Cheng S-k and Zhong L-s. (2013) Rural tourism development in China: Principles, models and the future. Journal of Mountain Science 10: 116-129.

Wang X, Zhen F and Wu X. (2017) Evaluation indices and empirical study of smart tourist attractions from the perspective of tourists: a case study of the Temple of Confucius and Qinhuai Scenic Site of Nanjing. Progress in Geography 34: 448-456.

Wang Z and Jinghong D. (1994) Theoretical Geomorphology, Beijing, China: Science Press.

Wen L, Liu C and Song H. (2019) Forecasting tourism demand using search query data: A hybrid modelling approach. Tourism Economics 25: 309-329.

Wilson A. (2010) Entropy in Urban and Regional Modelling: Retrospect and Prospect. Geographical Analysis 42: 364-394.

Wilson S, Fesenmaier DR, Fesenmaier J, et al. (2001) Factors for Success in Rural Tourism Development. Journal of Travel Research 40: 132-138.

Wu B and Yu X. (2010) Principles of tourism planning, Beijing, China: China Travel & Tourism Press.

Xinhua. (2017) China moves to boost rural tourism. Chinadaily.

Zhang L. (1989) Review on the study of tourism Gravity models and its future. Geographical Research 8: 76-87.

Zhao W, Li Z and Li R. (2017) Research on county rural tourism competitive ability in Fujian province 3):. Chinese Journal of Agricultural Resources and Regional Planning 3: 183-190.

Zhao Z and Yin H. (2004) On forecast of tourist number in new development tour district. Human Geography 19: 58-61.