Quantifying the impacts of 166 years of land cover change on lowland bird communities

Abstract

Land cover change for agriculture is thought to be a major threat to global biodiversity. However, its ecological impact has rarely been quantified in the Northern Hemisphere, as broad-scale conversion to farmland mainly occurred until the 1400s–1700s in the region, limiting the availability of sufficient data. The Ishikari Lowland in Hokkaido, Japan, offers an excellent opportunity to address this issue, as hunter–gatherer lifestyles dominated this region until the mid-nineteenth century and land cover maps are available for the period of land cover changes (i.e. 1850–2016). Using these maps and a hierarchical community model of relationships between breeding bird abundance and land cover types, we estimated that broad-scale land cover change over a 166-year period was associated with more than 70% decline in both potential species richness and abundance of avian communities. We estimated that the abundance of wetland and forest species declined by greater than 88%, whereas that of bare-ground/farmland species increased by more than 50%. Our results suggest that broad-scale land cover change for agriculture has led to drastic reductions in wetland and forest species and promoted changes in community composition in large parts of the Northern Hemisphere. This study provides potential baseline information that could inform future conservation policies.

1. Introduction

Biodiversity loss is accelerating across every ecosystem on Earth [1]. Land cover changes, particularly the conversion of wildland to farmland, are assumed to be the major drivers of these losses [2,3]. Farmland occupies one-third of the Earth's land surface, and approximately 90% of past global land cover changes were related to agricultural activity [3,4]. It is therefore critical to understand the environmental consequences of broad-scale land cover changes for agriculture for informing terrestrial biodiversity conservation actions [5].

Broad-scale land cover changes for agriculture are currently ongoing in tropical regions and are suggested to be the major threat to biodiversity in those regions [6,7]. On the contrary, these changes occurred until 1400–1700s in large parts of the Northern Hemisphere [8,9], and environmental consequences of ongoing agricultural intensification are of concern [8,10–12]. Therefore, few studies have examined the degree to which past broad-scale land cover change impacted biodiversity in temperate Northern Hemisphere before the agricultural intensification (but see [5,13]).

Past land cover change may have had a comparable, or even greater, impact on species richness and abundance of biological communities than recent land-cover changes such as agricultural intensification currently occurring in the Northern Hemisphere [14]. Furthermore, we would expect high variation in the timing and rate of reduction among functional groups, given that responses to land cover change vary among these groups [7,16]. These knowledge gaps could mask the detailed impacts of land cover change on biodiversity and underestimate its impact on vulnerable species or groups. Filling these gaps is key to informing current and future conservation actions [17], assessing the long-term population trends of species of conservation concern and providing quantitative conservation targets from species (e.g. IUCN Red and Green Lists) [18,19] to ecosystem levels [20,21].

Long-term biodiversity monitoring data are suitable for quantifying the impacts of human disturbance, including land cover change [22–24]. However, time-series biodiversity data are rarely available for the pre-clearance period (i.e. the period prior to land cover conversion to farmland), which undermines our ability to quantify the consequences of broad-scale land cover change for agriculture [13,22,25,26]. In fact, available time-series datasets that span more than 40 years are very rare worldwide [23]. Furthermore, time-series data are usually collected in already modified landscapes or protected areas [27]. Therefore, an alternative approach is required to estimate the spatial distribution of species richness and abundance during the pre-clearance period [13].

Combining digitization of historical land cover maps with species distribution modelling provides an alternative approach to quantifying the potential impacts of land cover change on biodiversity [28]. For example, potential species richness and abundance in the pre-clearance period can be estimated by modelling current abundance in various land cover types and projecting the results across historical land cover maps [13]. Though it is unfeasible in many regions to obtain land cover maps covering periods of more than 100 years, fine-scale land cover maps from the 1850s are available throughout Japan. In particular, one of the largest alluvial plains in northern Japan, the Ishikari Lowland (140 × 60 km) in central Hokkaido, was not converted to farmland until the 1860s [29]. This is because hunter–gatherer lifestyles dominated from the Paleolithic to the mid-nineteenth century in Hokkaido [30].

By making use of this socio-economic history and the land cover maps, we aimed to quantify the impacts of broad-scale land cover change for agriculture on breeding bird communities in the Ishikari Lowland over the past 166 years. Assessing habitat-based functional group-level responses is key to understand and derive community-level responses to broad-scale land cover change since each land cover type has different types of key resources such as food and nest sites [31,32], and hence the abundance of each species or species groups is associated with a certain type of land cover [33]. We thus examined the timing and extent of functional group-(forest, wetland, grassland and bare-ground/farmland species) and community-level species richness and abundance changes.

We used a hierarchical community model (HCM) [34] to estimate habitat–abundance relationships for breeding birds. Birds were suitable for our objective as they can respond well to environmental change and exhibit a diverse array of ecological functions [35,36]. We georeferenced historical land cover maps from five periods (the pre-clearance period, 1900, 1950, 1985 and 2016). We then projected abundance estimates across those land cover maps to quantify the degree to which land cover change has reduced bird abundance and species richness at the community and functional group levels.

2. Material and methods

(a) . Study area

Hokkaido, northern Japan, spans a transitional area between temperate and boreal zones. The mean annual temperature and precipitation in Sapporo, located in the center of the Ishikari Plain, are 9.2°C and 1146 mm, respectively (Japan Meteorological Agency, http://www.jma.go.jp). The Ishikari Lowland (comprising the Ishikari and Yufutsu Plains; 42°32′–43°50′ N, 141°08′–142°10′ E; figure 1) is now dominated by rice paddy and cropland, but the majority of the land base was not converted into farmland until the 1860 s [29,37]. This is because the Ainu, the Indigenous people of northern Japan, relied mainly on salmon fishing, deer and bear hunting, and gathering edible plants [29,37] until the Japanese government established a colonial government in Sapporo in 1869. An explorer from southern Japan investigated the Ishikari Lowland in the 1850 s and reported that the landscape was dominated by dense forests (e.g. alders [Alnus spp.], or Japanese elm [Ulmus japonica]) and large wetlands (dominated by common reed [Phragmites australis] or sedges [Carex spp.]) [29,38]. In 1869, the Japanese government began systematic forest clearing and wetland drainage to produce large tracts of arable land and expanded economic control of the region.

Figure 1.

Extent of mapped areas (red grids = 2 × 2 km resolution). (Online version in colour.)

(b) . Field sampling and functional group

We used bird data from our published studies [33,39], and details of the bird surveys are available in the studies and electronic supplementary material, S1. Bird surveys were conducted throughout the breeding season (May–July) in 2015 and 2017 in 79 sample plots (2 ha, 100 × 200 m) located throughout the Ishikari Lowland. We visited each plot three times and plotted the putative territories of individual species in each plot on a map based on field observations and estimates of territory size [39]. We used the summed number of territories for each species within each plot as a metric of abundance.

We focused on habitat-based functional groups (electronic supplementary material, S1) and classified all observed species into one of four functional groups based on habitat preferences and nesting habitats listed in the JAVIAN database [40]: forest species, wetland species, grassland species and bare-ground/farmland species. We defined bare-ground/farmland species as those that prefer frequently disturbed habitats, including agricultural habitats, such as cropland or rice paddy and short grassland [33]. Although some of those species nest on trees or shrubs, we focused on foraging habitat because the availability of foraging habitat could be more important for those species than that of nesting habitat in Japan (electronic supplementary material, S1).

(c) . Explanatory variables and land cover maps

We ranged the proportion of seven land cover types [wetland, grassland, wet abandoned farmland, dry abandoned farmland, cropland, rice paddy and forest] (representing the major land cover types currently present in the lowland) within sample plots, and calculated the proportion of each land cover type within each sample plot (0–1) using a vegetation map provided by the Nature Conservation Bureau of the Ministry of Environment. We then generated 400-m-wide buffers around each plot and measured the proportion of surrounding open land (0–1; hereafter referred to as ‘landscape openness’; open land is defined as summed area of wetland, grassland, wet abandoned farmland, dry abandoned farmland and cropland) in the buffer area [39]. We used this variable because our previous studies showed that this landscape variable was the most strongly associated with bird abundance in Hokkaido [39,41]. We tested for collinearity between variables using variance inflation factors (VIF), and the values were less than or equal to 2.

We analysed land cover maps from five different periods between 1850 (prior to the land cover change) and 2016 (present) (electronic supplementary material S2). The maps from 1900, 1950 and 1985 were based on 1 : 50 000 topographical maps and were provided by Himiyama et al. [42]. The 1850 map was produced by Himiyama et al. [42] and was partly based on a literature review; therefore, we assume that this map includes a higher level of uncertainty than the other maps, which are based on survey data. In particular, this map might underestimate the extent of wetlands, as explorers noted that ‘there are no trees in sight, only reeds, sedges and silver grass, and rarely a few alder trees' in some parts of the Ishikari Lowland [38].

We thus also used a map produced in 1880, created by Hokkaido [38] to aid in the selection of suitable colonial sites. This map is the oldest surveyed map of the Ishikari Lowland. Although the land cover changes for agriculture began in the 1860s, the 1880 map may still represent ecosystems in their natural state as clearance had not yet occurred in the majority of the lowland as of the 1880s. We thus used both the reviewed and surveyed maps to estimate bird population prior to the land cover changes to farmland. Paper maps were digitized and georeferenced for analysis. We also used a categorical land cover map issued in 2016 by the Geospatial Information Authority of Japan [43]. We confined the mapped area to lowlands (i.e. alluvial fans and deltas) to quantify the impact of land cover changes for agriculture on bird communities, because agricultural land is concentrated in lowlands [44], and because lowlands and mountainous areas differ with respect to bird community composition [45,46].

(d) . Estimating current abundance and species richness

We used an abundance-based HCM to estimate the effects of land cover types and surrounding habitat on the abundance of each species and functional group [34]. We did not consider the detection process in the model because openland songbirds have relatively high detectability in Hokkaido (approx. 0.66) [47], and we conducted surveys three times for each plot. We first assumed that the abundance of species i in plot j (Z_ij) followed a Poisson distribution (Z_ij ∼ Poisson[λ_ij]). We then assumed that the expected abundance of species i in plot j (λ_ij) was a function of the proportion of the seven land cover types within the plot, as well as landscape openness. Only landscape openness was standardized prior to analysis. We treated the proportion of the seven land cover types and landscape openness as continuous explanatory variables (0–1), and omitted the intercept of the linear predictor (i.e. we employed the cell means method [48]) using the parameter estimates

$$\begin{array}{rl}\log ({\lambda }_{ij})& =\hspace{0.17em}{\beta }_{i1}\hspace{0.17em}\times x_{\hspace{0.17em}j1}+\hspace{0.17em}{\beta }_{i2}\times \hspace{0.17em}{x}_{\hspace{0.17em}j2}+\hspace{0.17em}{\beta }_{i3}\times \hspace{0.17em}{x}_{\hspace{0.17em}j3}+\hspace{0.17em}{\beta }_{i4}\times \hspace{0.17em}{x}_{\hspace{0.17em}j4}\\ & +{\beta }_{i5}\times \hspace{0.17em}{x}_{\hspace{0.17em}j5}+\hspace{0.17em}{\beta }_{i6}\times \hspace{0.17em}{x}_{\hspace{0.17em}j6}+\hspace{0.17em}{\beta }_{i7}\times \hspace{0.17em}{x}_{\hspace{0.17em}j7}+{\beta }_{i8}\\ & \times \hspace{0.17em}{x}_{\hspace{0.17em}j8},\end{array}$$ 2.1

where x_j1–7 indicates the proportion of each land-cover type (0–1) in plot j and x_j8 indicates landscape openness (0–1) surrounding plot j. ${\beta }_{i1-8}$ indicates the partial regression coefficients of species i for each explanatory variable.

We assumed that species-level parameters followed a functional group-level normal distribution with hyperparameters. This implies that different functional groups can have different means and standard deviations for each coefficient, reflecting our assumption that the effects of land cover type and landscape openness differ among functional groups. We obtained median parameter estimates from the field survey data using a Monte Carlo Markov chain with three chains, a burn-in of 50 000, a thinning interval of 5, and 100 000 post-iterations. We used uninformative normal distributions (0, 100²) for the mean value of ${\beta }_{i1-8}$ and half-Cauchy distributions (0, 5) for any variance of coefficients as weakly informative prior distributions [49]. We performed 10-fold cross-validation to test and evaluate the ability of the HCM to predict bird abundance, and the root mean square error and mean absolute error values were 8.11 and 6.28, respectively. See electronic supplementary material, S1 for details of the HCM. We conducted these analyses using R v. 3.2.0 [50], JAGS v. 4.2.0 [51] and the R package jagsUI v. 1.4.2 [52].

(e) . Predictive mapping

We calculated potential bird abundance and species richness in the Ishikari Plain in past periods, assuming habitat–abundance relationships for the breeding avian communities stay constant over 1850–2016 (electronic supplementary material, S3). We first standardized the spatial resolution of land cover maps. The maps from Himiyama et al. [42] are available at a resolution of 2 km, and we converted the remaining maps to the same resolution. We then generated 2 × 2 km grid cells and overlaid these grids onto the maps. We extracted the dominant land cover type within each grid cell, assuming that grid cells were fully occupied by a single land cover type. To estimate bird abundance in the lowland, we converted the 2 km categorical land cover maps to 2 ha grid maps. We first divided all maps into 2 ha grids (100 × 200 m) so that they corresponded to the same resolution as our sample plots. The proportion of each of the five land cover types (wetland, grassland, cropland, rice paddy and forest) was stored in the grid (electronic supplementary material, S2).

We assumed open water and urban areas as unsuitable habitats for birds and assigned them an abundance of zero individuals because terrestrial bird abundance is low in these land cover types [53,54]. For each grid cell in each land cover map, we calculated the expected abundance (hereafter ‘abundance’) and the occurrence probability of each species based on the median parameter estimates from the HCM [34]. We defined occurrence probability of a species in a plot as the probability of occurring at least one individual. We then obtained community- and functional group-level abundance and species richness within each grid by summing the expected abundances and occurrence probabilities of the constituent species, respectively, and mapped these values across the Ishikari Lowland.

We calculated total bird abundance for each period by summing the abundance values of all grids over the Ishikari Plain. This was done only in the Ishikari Plain because Hokkaido [38] did not include the Yufutsu Plain in the 1880 map. We also calculated median values of species richness (i.e. α-diversity) for all grids in the Ishikari Plain, as well as the total area of each land cover type during each period. We quantified the changes in community composition by dividing functional group-level total abundance by community-level total abundance (hereafter ‘functional group proportion’). Finally, we compared these values among periods. We estimated 90% credible intervals for total bird abundance, species richness (α-diversity) and functional group proportions by projecting abundance and species richness 500 times from randomly resampled parameters from posterior distributions. We note that the calculated abundance and species richness were the estimates of potential values since suitable habitats are not always saturated to their full capacity, and we could not consider all breeding bird species in the Ishikari Lowland.

3. Results

(a) . Land cover changes in the Ishikari plain

Forest was the dominant land cover type during the pre-clearance period, occupying 60% (1880) to 75% (1850) of the Ishikari Plain. However, forest area declined by 88% between 1850 and 1900, and by 95% by 2016 (figure 2a and 3a). Wetland was the second most dominant land cover type and occupied 32% (1880) of the plain during the pre-clearance period, but was nearly eliminated by 1985 (figures 2a and 3a). Grassland cover expanded slightly until 1900 but has since decreased, exhibiting a net decline of 67% since the pre-clearance period (figures 2a and 3a). Cropland has increased substantially, occupying 57% of the plain in 1900, after which time it gradually declined (figures 2a and 3a). Rice paddies, in turn, became the dominant land cover type by 1950 and occupied 52% of the plain in 2016 (figures 2a and 3a). In 2016, urban areas occupied 23% of the plain, and their expansion has mainly occurred during the last 66 years (figures 2a and 3a).

Figure 2.

(a) Land cover maps (2 × 2 km resolution) from each period (1850, 1880, 1900, 1950, 1985 and 2016), and maps predicting the abundance of (b) all species, (c) forest species, (d) wetland species, (e) grassland species and (f) bare-ground/farmland species (100 m × 200 m resolution). The maps ‘1850’ and ‘1880’ are based on the reviewed [42] and surveyed [38] maps, respectively. The black area indicates unsuitable habitats assumed to support no bird individuals. (Online version in colour.)

Figure 3.

Estimated historical changes in (a) the area of each land cover type, (b) bird abundance and (c) functional group proportion. Land cover types are represented by colours in (a), and functional groups are represented by colours in (b) and (c). We categorized black-faced bunting as grassland species when conducted HCM. However, we excluded the species from grassland species when calculating group-level abundance and species richness. This is because the species was fairly abundant in every land cover type in the Ishikari Lowland. The maps ‘1850’ and ‘1880’ are based on the reviewed and surveyed maps, respectively. Bars indicate 90% CI: forest species, wetland species, grassland species and bare-ground/farmland species (from left to right for each period). (Online version in colour.)

(b) . Impacts of land cover changes on bird communities

The results of the HCM indicate that community-level abundance and species richness are higher in wetlands, grasslands, and forests than in croplands and rice paddies, and higher in croplands than in rice paddies. Forest species were more abundant in forests than in the other land cover types, and the abundances of wetland and grassland species were higher in wetlands and grasslands than in croplands or rice paddies. The abundance of bare-ground/farmland species was higher in croplands than in rice paddies. Detailed results are available in electronic supplementary material, S4–S6.

Total bird abundance was predicted to decline by 72.0% (90% CI: 54.0%–77.3%, 1850) or 75.5% (90% CI: 60.0%–79.4%, 1880) since the pre-clearance period, i.e. from approximately 2 100 000 (1850) or 2 400 000 (1880) territories in the pre-clearance period to 590 000 in 2016 (figures 2b and 3b). Similarly, community-level species richness (per 2 ha) was predicted to decline by 76.3% (90% CI: 62.1%–81.1%), from 9.3 in 1850 to 2.2 in 2016 (electronic supplementary material S5–S7). Responses to broad-scale land cover changes varied markedly among the four functional groups. Forest species were estimated to be the second most abundant group during the pre-clearance period but exhibited declines of 86% by 1900 and 94% by 2016 (figures 2c and 3b,c), compared to 1850. Wetland species were estimated to be dominant during the pre-clearance period and comprised greater than 34% of the bird community, but have since declined by greater than 88% (figures 2d and 3b,c). The sharpest decline (about 60%) in wetland species was estimated to have occurred between 1900 and 1950. The abundance of grassland species was predicted to increase to greater than 147% by 1900 but has since declined. The abundance of grassland species was estimated to decline by 65% since the pre-clearance period (figures 2e and 3b). The abundance of bare-ground/farmland species was estimated to have increased by greater than 229% between the pre-clearance period and 1900 (figures 2f and 3b) and this group has become dominant, comprising 59% of the community (figure 3c). The abundance of bare-ground/farmland species was estimated to have decreased since 1900 but has exhibited an overall increase of greater than 58% since the pre-clearance period.

4. Discussion

Focusing on a large, unique landscape where land cover maps were available for the pre-clearance period, we estimated avian richness and abundance prior to land cover change and found that broad-scale land cover change over the Ishikari Lowland resulted in a greater than 70% reduction in total abundance (approx. 1 500 000 territories) and species richness over the past 166 years. We also estimated that the abundances of wetland and forest species declined by greater than 88% (630 000 and 650 000 territories, respectively), whereas the abundance of bare-ground/farmland species increased by greater than 50% (170 000 territories). These results suggest that land cover change throughout the Ishikari Lowland resulted in community-level species loss because the dominant functional group in the lowland has shifted from the species-rich forest and wetland groups (23 and 10 species, respectively) to the species-poor bare-ground/farmland group (six species) (electronic supplementary material, S7).

Given the long history of land cover change in the Northern Hemisphere [8,9], our results suggest that turnover in community composition and drastic reductions in forest and wetland species had already occurred in large parts of the Northern Hemisphere prior to agricultural intensification. This is supported by archeological studies, suggesting drastic changes in bird community composition in Europe during the Bronze Age, when broad-scale land cover change was predicted to have occurred [55]. North America shares a similar history with Hokkaido and a study suggested that broad-scale land cover change might have resulted in drastic reductions in woodland and prairie specialist species [5]. These findings suggest that broad-scale land cover change caused by lifestyle changes from hunter–gatherer to agriculture can be one of the major anthropogenic drivers of biodiversity loss. Historical dynamics of biodiversity suggested by our results would provide baseline information to implement current conservation actions.

Protecting remnant habitats ought to be given the highest priority to conserve forest and wetland species that have drastically declined. Valuing abandoned farmland would also be effective, as the habitat suitability of abandoned farmland can be comparable to that of undisturbed habitats [33,56], and these habitats have become increasingly prevalent worldwide since the 1950s [4,9,57]. Such measures may be valuable in the Ishikari Lowland (electronic supplementary material S8). However, conservation actions should also focus on immediate problems resulting from current pressures, such as agricultural intensification. Conserving grassland and bare-ground/farmland species would be equally, or more, important in regions where both groups are severely threatened [10,11].

Our results suggest that assessing changes in the area occupied by various land cover types over multiple periods is key to understanding when, how, and to what extent communities and functional groups have declined. We found that each functional group responded to changes in the area of particular land cover types. For example, conversion from forest to cropland prior to 1900 may have led to drastic declines in forest species, but increases in bare-ground/farmland species. The expansion of grassland prior to 1900 may have mitigated the negative effects of wetland loss on wetland and grassland species; however, the subsequent expansion of croplands and rice paddies may have triggered declines in those groups. Photographs taken in Hokkaido between 1900 and the 1910s [58] indicate that there were few trees or wetland patches in cropland and rice paddy areas, supporting our assumption that these land cover types do not support forest, wetland or grassland species. Each functional group experienced a different trajectory of abundance changes in the Ishikari Lowland, resulting in turnover in the dominant functional group, and ultimately to community-level biodiversity losses.

We did not include the detection process in our statistical model and assumed that the detection probability was constant across land-cover types. We might underestimate bird abundance in forest, though that of forest species was generally higher in forest than other land cover types and bird detectability within 50 m can be constant across land-cover types [59,60]. We also reiterate that our calculated values represent somewhat crude estimates. Notably, we did not quantify species richness or abundance in urban areas or open water due to data limitations. These should be considered in future studies. Nevertheless, our literature review (electronic supplementary material S3) revealed that, in most cases, estimates of past abundance (per 2 ha) from other field studies were within the 95% credible intervals of our analyses. Furthermore, it is worth noting that our estimates of reductions in bird abundance and species richness might be conservative. This is because we were unable to include some locally extinct or large-bodied species that are especially vulnerable to habitat loss [61]. For example, the yellow-breasted bunting, Emberiza aureola, was historically quite common throughout the Ishikari Lowland, but its population had collapsed by the 1990s [62]. In addition, Blakiston's fish-owl, Ketupa blakistoni, has not been confirmed in the Ishikari lowland since the 1950s (electronic supplementary material, S7).

5. Conclusion

Combining reconstruction of historical land cover maps with modelling the relationships between bird abundance and land cover types allowed us to infer the role of long-term and broad-scale land cover change in the transformation of terrestrial biodiversity. The negative effects of documented reductions in bird communities are not expected to be confined to the Ishikari Plain, as other plains in Hokkaido have experienced similar land cover change over the same timeframe [42]. Our results suggest that the transition from hunter–gatherer to the farming lifestyles triggered serious biodiversity losses, where these changes probably would have occurred in any landscape where farmland is now dominant.

Furthermore, 72% of the breeding bird species in the Ishikari Lowland are migratory and their main wintering grounds are located between southern Japan and southeastern Asia. Therefore, the broad-scale land cover change that occurred in Hokkaido 66–166 years ago might also have triggered massive bird declines and losses in ecosystem function throughout the migration flyways. Migratory species breeding in the Northern Hemisphere are suggested to be negatively affected by contemporary wintering habitat loss and degradation in tropical regions [63,64]; however, our results indicate that historical broad-scale land cover change in the Northern Hemisphere might have already triggered declines in migratory species until 1400–1700s, suggesting the threatening drivers have been shifting to tropical regions. The conservation of wetlands and forests remains a key challenge to the conservation of global biodiversity, as further agricultural and bioenergy expansion is predicted to occur in boreal regions, where the richness of migratory species, including many wetland and forest species, is at its highest [65,66].

Ethics

Data used in this study were compiled in accordance with the current laws of Japan and with relevant guidelines and regulations.

Data accessibility

The datasets and codes supporting this article have been deposited in the Dryad Digital Repository: https://doi.org/10.5061/dryad.9w0vt4bgz [67].

Electronic supplementary material is available online [68].

Authors' contributions

M.K.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, validation, visualization, writing—original draft; Y.Y.: conceptualization, data curation, methodology, project administration, supervision, validation, writing—review and editing; M.S.: conceptualization, project administration, supervision, validation, writing—review and editing; M.H.: data curation, investigation, methodology, writing—review and editing; H.O.: data curation, resources, validation, writing—review and editing; M.O.: data curation, resources, validation, writing—review and editing; T.M.: conceptualization, data curation, funding acquisition, resources, supervision, validation, writing—review and editing; F.N.: funding acquisition, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the Environmental Research and Technology Development Fund of the Environmental Restoration and Conservation Agency of Japan (JPMEERF20154004, JPMEERF20184005, JPMEERF20202002); JSPS KAKENHI (JP19J20957, JP20H04380).