*2.5. Quality of Available Habitat (QAH) Within Space-Time Cluster Area*

In order to characterize the land cover vegetation within two significant space-time clusters, the clusters were overlaid with a QAH map. The results showed different patterns between cluster 1 and cluster 2 (Figure 3). In cluster 1, 50.6% of CSF notifications were reported in areas at QAH 1, while 31.3% were reported in areas at QAH 1.5, and 18.1% were reported in areas at QAH 2 (Table 2). In cluster 2, 22.7% of CSF notifications were reported in areas at QAH 1, 52.5% were reported in areas at QAH 1.5, 2.5% were reported in areas at QAH 1.75, and 22.2% were reported in areas at QAH 2 (Table 2).

**Figure 3.** Locations of the significant space-time clusters of CSF. Notifications: (*P* < 0.05) in Gifu Prefecture overlaid on a map of the quality of available habitat (QAH) levels for wild boar. Graduated colors indicate the quality of habitat availability from darker colors (areas with better quality of habitat availability) to lighter colors (areas with worse quality of habitat availability).

The CSF notifications within clusters 1 and 2 occurred within habitats that included rainfed croplands (QAH 1), a closed (>40%) needle-leaved evergreen forest (>5 m) (QAH 1.5), a mosaic of cropland (50%–70%) and vegetation (grassland/shrubland/forest) (20%–50%) (QAH 1.75), a mosaic of vegetation (grassland/shrubland/forest) (50%–70%) and cropland (20%–50%) (QAH 2), closed (>40%) broadleaved deciduous forest (>5 m) (QAH 2), and closed to open (>15%) mixed broadleaved and needle-leaved forest (>5 m) (QAH 2).

Although different patterns of land cover vegetation were observed between clusters 1 and 2, nearly 50% of CSF notifications within cluster 1 and more than 75% within cluster 2 were notified in QAH 1.5–2, which provides the greatest opportunities for food and shelter for wild boar.


**Table 2.** Quality of availability habitats (QAH) of CSF notifications within the two identified space-time clusters.

DP: domestic pig. WB: wild boar. *n*: the number of notifications.

#### **3. Discussion**

From 2018 until August 2019, all notifications of CSF outbreaks in Japan have been made in Gifu Prefecture as well as in the surrounding four prefectures. A total of 1110 notifications had been reported so far, with 1071 affecting wild boar and 39 affecting domestic pig farms. The continuous notification of CSF in the area might have been attributed to wide spread of the virus within wild boar populations favored by free animal movements, as well as to the emergence of epidemiologically related domestic pig farms. To prevent the disease spreading in wild boar, control measures including (i) fencing to restrict animal movements, (ii) hunting activities for active monitoring and to reduce susceptible populations, and (iii) disseminating baits for oral immunization, were implemented. However, the efficacy of these strategies has not been confirmed. Therefore, we conducted a spatio-temporal analysis to obtain epidemiological information of the spread of CSF in Gifu Prefecture. Results from this analysis could help to increase our understanding of the current CSF epidemic and to contribute strategies for the containment of the disease in domestic pigs and wild boar.

Japan is an island country that has achieved the status of freedom from several contagious animal diseases by implementing adequate control measures that take advantage of the country's geography. Nevertheless, Japan has imported outbreaks of contagious animal diseases from neighboring countries. In 2010, there was an outbreak of foot-and-mouth disease (FMD) in Miyazaki Prefecture in the southern part of Japan, which caused extensive losses in animal husbandry. According to the high degree of sequence homology between an original virus isolated in Japan and viruses that were circulating widely in East Asia, it was suspected that the FMD virus might have been introduced via movement of people or commodities from East Asia [28]. The high homology of genetic sequences between the CSF virus isolated in Japan and viruses prevailing in China suggests that the infectious CSF virus may have been introduced from China. Potential factors that could have contributed to disease introduction include easy access from the international airport to the affected area, which has regular and direct flights from China, and the relatively high population density of Chinese people in the affected area.

In the present study, standard deviational ellipse analysis was conducted to measure the standard distance of CSF notifications. Shifting the centroids of identified ellipses indicated that the disease notification has spread in a northeast direction. Overlaying the three identified ellipses with a map of snowfall area in Gifu Prefecture revealed that the disease spread along the border of the snowfall area. In the south of Gifu Prefecture, there is a widespread area of flat land with field crops or animal farms, residential areas, and forests surrounded by mountains to the north. As suggested by other authors [29,30], wild boar do not move to the snowfall or high mountain areas. Therefore, mountains could have acted as an effective geographical barrier to limit wild boar movements and guide the direction of the spread of CSF.

Another concern regarding the spread of the disease is the potential for it to jump to remote areas. During the epidemic, CSFV infections were confirmed on seven farms that were geographically distant from, but epidemiologically linked, to the farms affected by CSFV (i.e., run by the same owner, supported by the same husbandry company, etc.) [13]. Given the potential for transmission of the virus between pigs on any farms or from wild boar near that farm, the epidemiologically related farms may further expand the spread of disease. This "disordered" spread of disease could affect the accuracy of spatio-temporal analysis by overestimating the maximum distance of significant spatial association between notifications. During the FMD epidemic in Miyazaki, the disease was confirmed 70 km away from the zone of movement restriction, which could have been caused by vehicle transportation [28]. Unexpected occurrences of disease in epidemiologically related farms would require reviewing farm biosecurity measures, as well as disease monitoring protocols.

In the present study, the results of the multi-distance spatial cluster analysis revealed that the maximum distance of relationship between CSF notifications was 23 km. Because of the small number of CSF outbreaks on domestic pig farms, we estimated the maximum distance of the relationship between notifications of domestic pigs and wild boar. This assumption could have influenced our estimated distance resulting in overestimation due to long distance spread observed on domestic pig farms. Nevertheless, similar approaches have studied another transboundary animal disease, African swine fever (ASF), which shares hosts and most of the transmission mechanisms with CSF [31–33]. When comparing our results with other studies, the estimated distance (23 km) was similar to that obtained for notifications of ASF in domestic pigs (15 km) and wild boar (25 km) in Sardinia [32]. This finding may be useful for setting the range of effective surveillance and control zones in the affected area.

The application of cluster analysis to identify areas with significant spatio-temporal aggregation of the ASF outbreaks in Sardinia from 2004 to 2013 indicated four clusters, the largest of which had a radius of 30 km [33]. This does not correspond with the results of another report that identified one cluster with a radius of 3 km in the same area [32]. As discussed in Iglesias et al., methodological differences could have led to the discrepancy [32]. In present study, because of the small number of CSF outbreaks in pig farms, we could not identify the maximum distance for the relationship between notifications of CSF in pigs alone, but we were able to do it by considering pigs and wild boar together. The discordance between the findings of the two spatio-temporal analyses in Sardinia may suggest that by using mixed data for two species in the present study, we may have overestimated the distance of the spread of disease compared to true distance of transmission in each of the two species. However, we believe that this uncertainty would be acceptable for setting the monitoring area with high efficacy. Thus, these findings may be useful for setting the range of an effective surveillance and control zone.

Data on wild boar cases consisted of animals found dead and/or captured during surveillance activities. Many of wild boar were captured during active surveillance activities by setting traps and conducting hunting activities. Considering that most of the reported wild boar cases were located close to human habitats, the wild boar capture area may have been biased. Therefore, the disease could be wider spread in the area than what has been reported in official notifications, and the identified clusters could have had a shorter radius. Ideally, active virologic surveys should be intensively implemented to decrease the reporting biases by providing more samples to detect low levels of prevalence [34,35]. The Gifu Animal Health Administration has authorized hunting activities to reduce the number of susceptible, as well as potentially infected, individuals. Hunters are a critical group for implementing population control and proper disposal of wild boar carcasses.

According to the investigative report of the affected farms, there were some factors that might have increased the risk of CSFV introduction into affected farms, including (i) improper preparedness against invasion of wild or small animals into farms; (ii) imperfect clothing and boot changes in farms and pig pens, or disinfection of those materials; and (iii) inadequate vehicle disinfection [13]. To prevent contact among each of the hosts, in addition to raising awareness of disease among farmers and hunters, it is important to improve biosecurity measures in pig farms against CSFV as well as other infectious diseases.

Finally, we analyzed the QAH level of areas within the two identified clusters to characterize land cover vegetation in areas of disease aggregation. According to Bosch et al., a QAH 1 level corresponded to suitable areas for food or shelter for wild boar (mainly agricultural landscapes) [36]. In cluster 1, 50.6% of CSF notifications were reported in areas at QAH 1, whereas in cluster 2, 22.7% of CSF notifications were reported in areas at QAH 1. Considering that frequent direct and indirect contact is likely to occur between both hosts, contagious viruses in wild boar could be transmitted to pigs in the farms due to insufficient biosecurity in the affected farms since wild boar was the suspected source of infection on 80% of affected domestic pig farms in Gifu Prefecture during the studied epidemic [13,35]. On the other hand, almost 50% of CSF notifications within cluster 1 and over 75% within cluster 2 were associated with QAH 1.5–2, which mainly corresponded to natural landscapes. These natural areas provided the greatest opportunities for food and shelter for wild boar. In the case of ASF, it has been reported that wild boar can transmit the disease efficiently at local levels within their own population [32,36]. Furthermore, De la Torre et al. suggested that the spread of ASF in Europe was driven by contact between animals from different populations that moved short distances [37]. Although ASF is caused by another virus, given that wild boar play an important role in both diseases, it is plausible to assume that CSF also could have expanded through contact between individual wild boar. Therefore, it would be critical to control wild boar populations and manage wild boar carcasses adequately from the environment to reduce habitat contamination.

Interestingly, the QAH map could also identify routes of CSF introduction or spread, mediated by wild boar, through vegetation or travel corridors. Travel corridors are either unbroken vegetation corridors or patches of habitat that enable animals to travel securely from one habitat to another [36]. These patches of habitat and vegetation corridors could be used as strategic points of vaccination where oral baits could be placed. In Gifu Prefecture, the vegetation is composed mainly of broadleaved evergreen and broadleaved deciduous forests, which provide suitable habitat for wild boar [38,39]. Given that the composition of the vegetation in Gifu Prefecture is common throughout Japan, it is likely that the disease could spread similarly to other prefectures.

It should be noted that vegetation types and wild boar behavior could vary among geographical features. For example, mountains usually have gentle slopes in Germany, whereas Japanese mountains tend to have precipitous slopes [40]. These topographical differences may require different approaches for control of wild boar populations.

Almost one year has passed since the first notification of the CSF outbreak in Japan, and the spread of the disease has been confirmed mainly in wild boar. Fortunately, CSF outbreaks on domestic pig farms have been limited. Nevertheless, the potential risk of CSF introduction on farms could be high due to limited biosecurity, high number of wild boar cases in the area, and difficulties in implementing disease control measures in wildlife [13]. The results from this study provide information on the current epidemic, which may help improve current approaches for controlling CSF in Japan. Information on the direction and distance of disease spread could help with the implementation of control measures by modifying the area for control and surveillance zones or identifying specific locations for increasing efforts of oral immunization.

Given the potential risk of the ASF introduction from neighboring countries, we should summarize and disseminate the lessons learned from the current CSF outbreak to achieve the protection of ASF invasion or rapid containment of its occurrence even if it occurred.

#### **4. Material and Methods**

#### *4.1. Data and Data Sources*

Epidemiological data for the periods from 9 September 2018 to 25 June 2019 were provided by the Gifu Prefectural Government, which provided the dates and coordinates (latitude and longitude) of the notifications of CSF in domestic pigs and wild boar. A total of 743 CSF notifications, 16 outbreaks on domestic pig farms, and 727 cases in wild boar were confirmed by RT-PCR and/or ELISA tests in the laboratory [13]. As we focused on local transmission of CSFV, notifications of CSF in slaughterhouses or in facilities through which CSF-affected pigs had been transported were removed from the current study. Notifications of CSF in wild boar reported on the same day and location were regarded as one case.

#### *4.2. Standard Deviational Ellipse Analysis*

Standard deviational ellipse (SDE) analysis is a tool that provides the orientation and shape of a distribution, as well as its location, and dispersion or concentration of the data [41]. It requires a single point that is used to define the standard deviational ellipse. The analysis was conducted to describe the trend and spatial characteristics of CSF notifications in the study area in ArcGIS 10.6.1 software (ESRI Inc., Redlands, CA, USA) following an approach similar to Fonseca et al. and Lu et al. [42,43]. The ratio (R) of the long and short axes was used to identify the degree of clustering (R > 1) or dispersion (R = 1) [42,43]. To analyze temporal changes of CSF notifications, the study period was divided into three stages—(i) September to December 2018 (four months), (ii) January to March 2019 (three months), and (iii) April to June 2019 (three months).

#### *4.3. Multi-Distance Spatial Cluster Analysis*

A multi-distance spatial cluster analysis tool in ArcGIS software version 10.6.1 was used to identify the maximum distance of the relationships between CSF notifications according to the guide on the manufacture's website [44]. In brief, the tool uses a common transformation of Ripley's k function, wherein the expected result with a random set of events is equal to the input distance. The transformation L(d) is given by the following formula:

$$\mathcal{L}(\mathbf{d}) = \sqrt{\frac{A\Sigma\_{i=1}^N \Sigma\_{j=1,j\neq 1}^N k(i,j)}{\pi N (N-1)}}$$

where *A* is the area, *N* is the number of events, *d* is the distance, and *k*(*i*, *j*) is the weight, in which it is 1 when the distance between *i* and *j* is less than or equal to d and it is 0 when the distance between *i* and *j* is greater than *d*. To analyze the spatial pattern of CSF notifications, Observed K values were compared to the Expected K values of a completely random spatial distribution of CSF notifications with 999 simulations, which is equal to confidence levels of 99.9%.

The Diff K values contain the Observed K values minus the Expected K values. In the present analysis, the Expected K values that yield the highest Diff K values were applied as the maximum distance for relationships between notifications of CSF outbreaks in Gifu Prefecture.

#### *4.4. Kernel Density Estimation Analysis*

Kernel density estimation is a non-parametric estimator for describing the spatial extent of a series of events [45]. In the current study, the kernel density tool was applied to explore the influence of the CSF notifications in the study area by calculating the density of CSF notifications in ArcGIS 10.6.1. A radius of 23 km based on results obtained from Ripley's k function, was applied as the maximum distance for significant spatial association between CSF notifications. Kernel density estimation was divided into five categories according to the equal interval method.
