Monitoring of the spatial distribution of vacant houses across broad areas for vacant house measures by Japanese local governments requires substantial labor, time, and money because the main survey method is visual inspection via field surveys. Therefore, we developed the method for estimating the spatial distribution of them using various municipal public data and sample field surveys in the paper of part1. This paper aims to improve our previous method and increase the estimation accuracy. The target area was the whole area of Kagoshima city and Asakura city shown in Fig. 1 and Fig. 2.
In chapter 2, we detected the spatial distribution of vacant houses via field surveys by visual inspection in the sample field survey areas in Fig. 1 and Fig. 2, and developed the vacant house database for developing the estimation method of vacant house distribution. The vacant house database has results of field survey shown in Table1 and attributes shown in Table2 of each detached building. Municipal public data included the database are the basic resident register (BRR), Hydrant consumption amount information (HCI), and the building registration information (BRI).
In chapter 3, we developed the estimation method of vacant house distribution. The method first developed crosstab tables with the attributes of Table2 as explanatory variables, secondly calculated the vacancy rate for each combination, and finally estimated the vacancy rate of each building by allocating the vacancy rate of the table for each building. At first, we developed the crosstab tables by cell splitting of each explanatory variable shown as Table 5. Second, as shown in Table6, because of comparing the number of vacant house every 250 m square grid with the true value while decreasing the number of used variables, it was revealed that the estimation result by the crosstab table using all the variables is the most reliable. In addition, it was also revealed that the HCI has high interpretability shown in Fig. 4. By comparison with estimation result of the method in the part 1, it was confirmed that estimation accuracy improved considerably as shown in Fig. 5 and Fig. 6. Third, as a result of estimating Asakura city by the crosstab table developed from the field survey result of Kagoshima city, there were some grids with large error although the high estimation accuracy was obtained as shown in Fig. 7. Therefore, we estimated using a crosstab table developed using the results of field survey conducted by both cities. As a result, high estimation accuracies were obtained in both cities: the correlation coefficient was about 0.95, and the mean absolute error was about 1.0 as shown Table7. From this result, it was expected that by collecting the field survey results and public data of more municipalities in the future, it is possible to estimate vacant house distributions with higher accuracy in various municipalities.
Finally, we accomplished to estimate the spatial distribution of vacant houses in arbitrary spatial unit in Chapter 4. Fig. 8 to Fig. 11 show the estimated results of the number of vacant houses and rates aggregated into 500m square grids of Kagoshima city and Asakura city. The results estimated that 7, 361 detached buildings were vacant of the 270, 631 detached buildings, i.e. 2.72% of houses were vacant in Kagoshima city, and 2, 299 detached buildings were vacant of the 44, 005 detached buildings, i.e. 5.22% of houses were vacant in Asakura city.
