This study employed agricultural statistical data from the CAFF dataset, which the Ministry of Agriculture, Forestry and Fisheries of Japan compiles every 5 years. We used the eight datasets obtained from 1970 to 2005. The CAFF data were derived from “e-Stat,” which is the website portal used for Japanese government statistical information (Ministry of Internal Affairs and Communications). We obtained the CAFF datasets from the Ministry of Agriculture, Forestry and Fisheries website because this website integrated several raw datasets into prefectural units (CAFF 2005).
The CAFF dataset includes statistical data related to agricultural land area, e.g., total farmland and a breakdown of farmland into categories such as paddy fields and dry farmland, as well as some more detailed categories such as bean and wheat fields (Ministry of Agriculture, Forestry and Fisheries). These records were summarized using hamlet units, which included old municipality units, i.e., one old municipality unit consists of several hamlet units. We aggregated the hamlet units into municipality units using the hamlet records for 2000. The number of hamlet units has gradually decreased as farmers retire, with the greatest number of hamlets existing in 1970 when the period from 1970 to 2005 is considered. The hamlets that have vanished could not be attached to municipalities because the records for old hamlet units were never digitized. Consequently, the area of farmland from the records for 1970 to 1995 was underestimated because we used the hamlet units existing in 2000. In this data paper, we could not use the more detailed crop categories such as beans and wheat, as in Kohyama et al. (2003), because of the masking of locations. If a hamlet had fewer than four farmers, the records were masked to protect the farmers’ personal information (Ministry of Agriculture, Forestry and Fisheries); thus, several values were underestimated. Kohyama et al. (2003) could only use data for 1970–1995 and could not include data from 2000 for the same reason. Use of the larger categories, such as dry farmland, which pools the more detailed categories, allowed us to alleviate much of the underestimation.
To convert the municipality units into Second-order and 5-km mesh units, we divided the municipality units into 100-m (i.e., 1 ha) mesh units, and reconstructed the Second-order and 5-km meshes, which contained 10,000 and 2,500 100-m meshes, respectively. When we divided the municipality units into 100-m mesh units, we partitioned both the total farmland and abandoned agricultural areas equally. If one mesh unit included two or more municipality units, we assigned the mesh unit to the dominant municipality, i.e., the municipality that represented the largest portion of the mesh area. We assumed that the proportions of total farmland and abandoned areas were distributed equally in the 100-m mesh grid within the same municipality. This is the concise conversion technique used by Kohyama et al. (2003). Subsequently, we reconstructed Second-order and 5-km mesh units using 100 × 100-m meshes and summed the areas for all agricultural land use areas. We ignored the numerical error that might occur by dividing the municipality units into the 100-m mesh units because the Second-order and 5-km mesh sizes were sufficiently larger than the 100-m mesh units. We used small 100-m mesh units rather than 1-km grids as the minimum area for reconstructing the Second-order Mesh units because the smallest municipality unit was 1.27 km2 (Ministry of Agriculture, Forestry and Fisheries), and the 1-km mesh did not meet our accuracy requirement. We did not create 1-km mesh data for the same reason. Creating 1-km mesh data using the same methods as the Second-order and 5-km meshes, would have resulted in many neighboring meshes having identical values without a more detailed source dataset, which is only available for purchase from the Association of Agriculture and Forestry Statistics (http://www.aafs.or.jp) (Kohyama et al. 2003). We reconstructed the Second-order and 5-km meshes and calculated the areas of total farmland, paddy fields, other farmland, and orchards in each Second-order Mesh unit. For 2005, we also calculated the abandoned agricultural area because these areas were clearly reported in 2005 (Ministry of Agriculture, Forestry and Fisheries).
The records for 1970 to 2000 were based on the Tokyo Datum. A difference of about 400 m exists between the Tokyo Datum and the Japanese Geodetic Datum 2000 (JGD2000), which is the current standard datum in Japan (Matsumura et al. 2004; Geospatial Information Authority). To make the conversion, we divided the Tokyo Datum mesh according to JGD2000 mesh using the “Intersect” function in a geographic information system (GIS ; ArcGIS, ESRI, California, USA), and then dissolved the resulting mesh into the JGD2000 mesh using the “Dissolve” function in a GIS. Although several methods were available for converting the Datum, we used this method because of its simplicity. Additionally, we believe that any errors in the Second-order and 5-km meshes caused by this conversion were quite small.