Study area and analysing unit

The study was conducted across all of Japan (Figure 1). A grid size of approximately 10 km was used for the analyses. This is the Japanese Standard Second Mesh (hereafter, 2nd mesh), which corresponds to 1:25 000 map grids (Figure 1). The locations of the 2nd mesh grids were determined arbitrarily by the Japanese government for comparing several statistics, including population, infrastructure and prevention of disasters at this scale of geographical resolution [28]. We obtained 2nd mesh information as polygon data from the Japan Integrated Biodiversity Information System available at the Biodiversity Centre of Japan [29]. In the past 15 years, the rate of agricultural abandonment in Japan has approximately doubled [25]. This is a critical issue in Japanese agricultural planning, not only for biodiversity but also for food production [25].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Study area and units selected for this study.

Locations and sizes of grids were defined by the Japanese government.

https://doi.org/10.1371/journal.pone.0079978.g001

PT species selection and occurrence records

In Japan, Japanese Prefectural Red Data Books (PRDBs) are created by local authorities in all 47 prefectures in the country. Listed taxa occur locally and are subject to decline within the prefecture, which is determined using prefecture-specific criteria. Using all of the PRDBs, it is possible to identify species that had formerly broad distributions but have declining populations, namely PT species, across wide areas of Japan. Species with limited habitat and/or small distribution ranges, such as specific islands or high montane areas, were ruled out of consideration for our purposes. We defined the PT species, as listed in the PRDBs, for at least 35 prefectures that occupy 75% of the national surface area. To assign PT status to species, we collected the 2010 PRDBs from all 47 prefectures (Table S1). Subsequently, we listed all plant species records from PRDBs and counted the numbers of each. In this procedure, the threatened category assignment in each PRDB was ignored because the prefectural definitions were variable.

Distribution records of PT species were obtained from J-IBIS [29]. We used occurrence records of threatened plants in 2nd meshes in J-IBIS that were compiled from Japanese National Red Lists (NRLs) in 2001 and 2007. These data were corrected using observations and information from regional experts [30]. Distribution records were refined by conducting additional surveys over a 5-year period based on previous records. Thus, these data are cumulative occurrence records from past to present, include extinctions and cover all of Japan. We took PT species records from this data set and selected those with at least 50 records. Thus, we obtained distribution records for 23 PT plant species per 2nd mesh grid unit for the whole of Japan (Table 1). These records were used in our analysis.

In addition to the PRDBs, a NRL exists, and species therein have well-known ecological traits, distributions and habitat requirements because conservation research and/or activities have tended to focus on PRDBs and/or NRL-listed taxa [31,32]. Indeed, the NRL publication itself contains considerable relevant information about each of the species [26]. For each PT species, we used the primary habitat and the threatened category assignment in the NRL as reference information. We referred to both the current NRL [26] and the book of endangered plants by Yahara and Nagata [33].

Agricultural land-use data set

We obtained agricultural land-use data from the Census for Agriculture, Forestry and Fisheries (CAFF) data set [34]. This census is conducted every 5 years by the ministry. We used data released in 2005, which matched years when PT species records were released (2001 and 2007).

The CAFF data set includes statistics on agricultural land use, e.g. total farmland and breakdown of farmland into categories such as abandoned areas and pasture [34]. These records were summarised using old municipality units. The average area of a municipality unit was 148.32 km² [35]. To convert the municipality units into 2nd mesh units, we divided the municipality units into 100-m (i.e. 1 ha) mesh units, and reconstructed the 2nd mesh. The 2nd meshes could contain 10 000 m² that was divided into 100-m meshes. First, we divided the municipality units into 100-m mesh units and partitioned both total farmland and agriculturally abandoned areas equally. If one of these mesh units included two or more municipality units, we assigned the mesh unit to the dominant municipality; i.e. we assigned the mesh to the single municipality that represented the largest portion of the mesh area. Thus, we assumed that the proportions of total farmland and abandoned areas were equally distributed in the 100-m grid within the same municipality. Subsequently, we reconstructed 2nd mesh units using 100 × 100-m (i.e. 10 000 m²) meshes and summed area values for both total farmland and abandoned areas. We ignored the numerical error that might occur by dividing the municipality units into the 100-m mesh because the 2nd mesh size was sufficiently larger than the 100-m mesh. We used 100-m small mesh units rather than 1-km grids as the minimum area for reconstructing the 2nd mesh because the smallest municipality unit was 1.27 km² [34,35], and the 1-km mesh did not meet our accuracy requirement. We reconstructed 2nd meshes and calculated the total farmland and abandonment area in each 2nd mesh unit.

Analysing relationships between PT species records and areas of farmland

The 23 PT species records per 2nd mesh unit were merged with the data for both total farmland area and abandonment area coverage in the same units. We used Mann–Whitney U-tests to compare average abandoned area between the ranges of PT species’ distributions and all of Japan per 2nd mesh unit for both 2001 and 2007. Mann–Whitney U-tests are robust for relatively small samples [36] such as the PT occurrences in this study. We did not use total farmland area in this analysis. In this procedure, we excluded 2nd meshes that had no abandoned area per 2nd mesh unit. If species occurrences overlapped areas where abandonment is increasing, the average abandoned area within the PT species’ distribution would be significantly larger than the abandoned area across the whole country.

We also used generalised linear models (GLMs) with logistic error distribution (log-link) and Wald tests to examine the effects of abandoned area and total farmland area for each PT species occurrence. We used total farmland as an explanatory variable in GLM analyses to determine if PT species depend on farmland and to separate the effects of abandoned areas and farmland. The correlation value between abandoned area and total farmland was 0.0553. We deemed these results “meshes”, which have PT species records as spatial presence units. The other results did not have PT species records as absence units to apply to the GLMs. Although our absence units include pseudo-absences, which means that the species was not detected but does actually occur, we found this analysis sufficient to provide general trend information on the preparation of farmland type requirements for each PT species.

Additionally, we conducted a “bagging GLM” analysis, which combines GLM and bootstrap aggregation, a machine learning technique [32]. The bagging GLM model creates a large number of GLM models using the same composition of presence/absence data derived from all data and then combines these models. This method has high predictive ability with presence data and also has advantages when analysing presence-only records for broad range records, such as the data used in this study [32]. A bagging GLM cannot serve as a significance test but can show coefficients in each explanatory variables [32]. In this study, we randomly selected a fixed number of supposed absence data from all meshes without PT species records as presence data, repeated the process to create 5 000 GLM models, and then combined these models into a final model. The final model was used as a linear predictor based on the average values of the coefficients of the 90% confidence intervals of all 5 000 models [32]. We used the statistical package R ver. 2.14 [37] for all analyses.

Dependency of PT species on farmland

Most of the PT species grew in agricultural habitats such as paddy fields, ponds and ditches (Table 1). The occurrences of 14 species were positively influenced by the total area of farmland, and all species studied were influenced by either abandoned area or total farmland area (GLM analysis; Table 3). In addition, the occurrences of 20 PT species were positively influenced by the total area of farmland (bagging GLM results; Table 4). These results suggest that nearly all PT species are likely to depend on farmland, especially paddy fields. Paddy fields represent a major portion of agricultural land use in Japan. Therefore, our expectation of which PT species depend on farmland for habitat is generally supported.

Farmland likely to be abandoned

We demonstrated unequivocally that areas of increasing agricultural abandonment overlap areas where PT species occur in Japan. Also, at least 18 species occurrences were positively influenced by abandoned area in the results of both the GLM and bagging GLM analyses (Tables 3, 4). From a study of four threatened plants, one of which (V. Pycnostelma) was included in our study, Uematsu et al. [12] showed that threatened plant species are likely to be distributed in paddy fields with a high risk of abandonment. We provide supporting evidence for this idea across the entire Japanese archipelago, thus supporting the generality of this result. This suggests that relationships may exist between farmland likely to be abandoned and habitat for threatened plant species.

Effects of abandonment on PT species

Our results showed an overlap between abandoned areas and the occurrences of PT species. Many previous studies have suggested that agricultural abandonment negatively affects regional biodiversity in several parts of Japan [10,12,19,21,22,38]. Our findings provide supporting evidence and offer a panoramic view of biodiversity losses over large agricultural landscapes. Several mechanisms that drive biodiversity loss when farmlands are abandoned have been proposed in the past. These mechanisms include reduced rates of disturbance [22] and changes in water conditions after abandonment [12,38]. Thus, biodiversity losses are likely driven by multiple factors after land-use shifts. In fact, our 23 PT species included several life-forms, morphologies and diverse habitat requirements (Table 1); declining numbers of such species are highly unlikely to be attributable to a single mechanism. Thus, agricultural abandonment probably affects plant species with diverse ecological traits through multiple mechanisms [12]. Water is likely a key component of the mechanisms because the three habitats most occupied by the 23 PT species were paddies, ponds and ditches (Table 1); furthermore, previous research has shown that water dynamics in agricultural fields strongly influence several plant species [23]. Future investigations should examine in more detail the processes of agricultural abandonment that cause population declines for each species and/or species trait clusters because the influences of agricultural abandonment are determined by species-specific habitat preferences and/or ecological traits [12].

Our results also suggest that abandoned farmlands provide habitats for some PT species. Ikegami et al. [23] indicated that agricultural abandonment may recover natural habitat from human-managed environments. From this viewpoint, habitat recovery should currently be progressing through all of Japan, with positive influences on regional biodiversity. In fact, agricultural abandonment has occurred relatively recently, since 1990 [11,12], whereas biodiversity losses generally occurred before 1970 [39]. Thus, some of the PT species occurrences might have been the result of habitat recovery following agricultural abandonment. Nevertheless, many previous studies have suggested that agricultural abandonment negatively affects regional biodiversity. Conditions might exist under which abandonment has different effects on PT species.

Differences within abandoned areas

Our results suggest that agricultural abandonment has contradictory effects on plant diversity compared to previous conclusions. Here, we propose two interpretations for both positive and negative effects of agricultural abandonment that explain our results and those of previous studies. The first is the potential natural habitat effect. As agricultural enterprise changes natural habitats to human-managed habitats [1,2], the former native habitats might have been more diverse originally. For example, before conversion to single-use agriculture, a given habitat may have supported wetlands, forests and grasslands. Agricultural abandonment might recover some these natural habitats, which may or may not be suitable for PT species. Thus, the influence of abandonment can be positive or negative depending on the suitability of habitats that emerge from the process.

Our second interpretation involves land-use history effects. Specifically, consolidation work, such as heavy land levelling, in farmland might negatively influence habitat conditions for plant species. Ikegami et al. [23] indicated that abandoned farmlands harbour some threatened species under existing conditions, i.e. with no consolidation work. Consolidation work may negatively influence plant species diversity over protracted time periods [12,23]. Recently, agriculture developments have been consolidated in high-yield areas for economic reasons [39,40]. Furthermore, technological developments for agriculture provide increasing yields per unit area. As a result, we can expect increases in the areas of previously consolidated but subsequently abandoned farmlands, and this process may inevitably have negative effects on plant species diversity. With our data, we were not able to demonstrate which of these positive and negative effects is more likely in chosen sections of landscape because we used a macro-scale approach that ignored local fine-scale details [41,42,43]. Therefore, future investigations should also examine fine-scale details of original habitat (pre-agriculture) and land-use history (experience of consolidation work) within regions of increasing farmland abandonment.