2.1. Study Scope
The selected target species is the red-crowned crane (Grus japonensis). It is the most highly endangered of the genus. First, we reviewed earlier studies on red-crowned cranes and methods of evaluating habitat suitability. We then gathered red-crowned crane presence data and selected and optimized the model for the Cheorwon-gun area. Next, we mapped and analyzed Cheorwon-gun habitats. We also assessed other areas near the DMZs as potential alternative red-crowned crane habitats. Finally, we proposed a conceptual diagram to support a conservation plan and prevent the negative effects of development on the red-crowned crane.
This study was conducted from October 2009 to March 2010 and covered the wintering season of the red-crowned crane. Since Cheorwon-gun has large CCZs and DMZs, the ROK government granted permission to run field surveys. Two areas were selected: one was Cheorwon-gun, Gangwon-do (ROK), an important wintering area for red-crowned cranes (Figure 1
). A habitat suitability model was established and habitat characteristics were analyzed for red-crowned cranes wintering in this area, which was also used for crane presence data collection. For the purpose of finding alternative crane wintering habitats, a second study area was selected within an 8 km radius of the DMZ (Figure 1
). Owing to the strict military security regulations there, this area was seldom affected by development.
2.2. Field Surveys for Collecting Presence Data
Crane presence data were collected during the wintering season. Field surveys were conducted six times (2009: 30 October, 14 November and 19 December 2010: 9 January, 19 February and 13 March) in Cheorwon-gun. The survey schedule was selected according to weather conditions and military authority regulations. The field survey area was divided into four sub-areas by location: (1) Jungyeon-ri, Igil-ri, and Hagal-ri; (2) Sapseulbong; (3) Daema-ri and Sammyeong-ri; and (4) Hantang-river and Dongsong-eup. These sub-areas were surveyed simultaneously to collect reliable presence data.
Field surveys were conducted by four teams, one in each sub-area. Each team consisted of two people who used a motor vehicle to move through the area. Data were collected around noon, which is the crane feeding time. When a group of cranes was observed, the survey team stopped the vehicle far away from them, counted them, and noted their location within rectangular rice paddies. Cranes usually made a group that consisted of three to five members, and thus we recorded a group of cranes as one location. Crane locations were marked on a topographical map with a corresponding symbol and coordinate. Symbols and coordinates were then collected and marked on a geographical information systems (GIS) map.
For all subareas, the maximum and minimum numbers of cranes sighted were 863 (December 2009) and 11 (October 2009) respectively (Table 1
). The number of red-crowned cranes sighted was highest in Sapseulbong and lowest in Hantan-river and Dongsong-eup. All data from October 2009 to March 2010 were used for further analysis.
Presence data collected from the field survey was fed into a digital map using the coordinates of each point (Figure 2
). All incidents of crane landing and taking-off were recorded and compared with other researcher’s maps to avoid double counting. We used these data as model inputs. The survey team also collected information and photographs of food traces and droppings to confirm location data.
2.3. Habitat Suitability Analysis
Most studies on the habitat characteristics of a single species focus on presence/absence data in a sampling area [15
]. Target species absence data, however, tend to be poor compared to species presence data [16
]. Hence, in order to perform a proper presence/absence data analysis, areas where no animals were spotted, and which do not have suitable habitat conditions, should be selected, as well as appropriate habitats where animals have been seen [17
]. Nevertheless, the precision of this method may be compromised since it relies upon random site selection from areas where the species has not been found. Phillips et al
] suggest that, when presence alone has been surveyed, species distribution should be plotted using a maximum entropy model rather than a presence/absence model. Thus, we used a maximum entropy model to analyze crane habitat suitability.
Crane habitat preference is related to food availability [19
], disturbance agents [20
], and sleeping places [22
]. Research on crane sleeping sites is limited since nocturnal access to this area is restricted. On the other hand, there are many studies on food resources in the area. Cranes prefer rice paddies with abundant grain and are affected by disturbance factors such as automobile traffic.
Based on a literature review, nine variables that affect crane habitats were selected. Land cover type was used as categorical variable. Variables related to distance included the distance from a farmland, a farm road (unpaved road), a paved road, a residential area, a river, and a forest. Farmland area was included as an area-related variable. The land cover type enabled the identification of the land cover preferred by cranes, whereas distance-related factors helped to identify the preferred environmental characteristics. The distance to farmland may also help to identify crane preferences in relation to environmental change of the area. Each variable was created as a 30-m2 raster data.
A detailed method for setting the presence data was needed to make an appropriate model. Since model performance may differ between data sets, we assessed 10 random data subsets [18
]. The model was based on data from 3222 locations. Three-quarters (2417) of the points were randomly selected, and 10,000 random background pixels were used as negative instances and training data. One-quarter (805) of the points were used for model testing. During the model run, a five-fold cross-validation was applied to minimize errors that occurred in the process of making crane location training data. Meanwhile, we obtained very similar habitat properties among modeling results of each observation period. Thus, we decided to use all period data for modeling habitat suitability.
The response curve function provided information on the relationship between presence points and environmental characteristics. Linear, quadratic, product, hinge, and categorical functions were selected as variables of the model. The maximum number of iterations was set to 1000 and the convergence threshold was set to 10−5 in the advanced tab. The regularization multiplier was set to one, and the maximum number of background points was set to 10,000 in the basic tab. To assess model performance, a receiver operating characteristic (ROC) analysis was undertaken. The ROC analysis compared Maxent model results with random predictions. Model performance can also be evaluated by using the Area Under the Curve (AUC) value.
After building an optimal model for Cheorwon-gun (a study area), it was applied to the zone within an 8 km radius of the DMZ (another study area) to find alternative potential habitats for red-crowned cranes. A projection function was used to assess the potential habitats. This helps analyze the habitat suitability of an area lacking presence data by using a model for another area. Data from the same variables used in Cheorwon-gun were used in the adjacent area.