We conducted our study in southwestern Hyogo Prefecture, Japan (Figure 1). We selected this area because of its abundant ponds (>5000 ponds within ∼780 km²) and the increasing interest in conserving regional freshwater biodiversity (Akasaka et al., 2010). The climate is warm and temperate, with a mean annual temperature of 14.4°C (mean minimum, 3.5°C in January; mean maximum, 26.4°C in August) and a mean annual precipitation of 1183 mm based on data at the Miki Climatological Observatory, which is located within the study area at 145 m a.s.l. The elevation range of the study area is <160 m and it generally decreases from the north to the south.

Irrigation ponds, which account for virtually all the existing ponds in this region, were created extensively before the 1910s to provide water for rice cultivation (Uchida, 2003), and more than 8% of the ponds have been lost in the last half century because of urbanization, residential development, or abandonment of agriculture (Uchida, 2003). The macrophyte biodiversity of the remaining ponds has decreased drastically, especially after the 1980s (Ishii and Kadono, 2003). In our previous studies of ponds in this region, surface area ranged from 107 to 29,204 m² and potential maximum depth from 1.1 to 8.1 m (Akasaka et al., 2010; Akasaka and Takamura, 2011, 2012), although the majority of the ponds in those studies do not overlap with the ponds surveyed here.

We haphazardly chose 369 irrigation ponds for our study (Figure 1). The selection of ponds was not strictly randomized; we selected ponds based on ease of access and with of a variety of sizes and characteristics. We surveyed each pond for macrophyte occurrence using an inflatable boat and a rake at the approximate time of peak macrophyte biomass (late August to September) in either 2006 or 2008. We recorded all species of floating-leaved, submerged, and free-floating submerged macrophytes using the nomenclature of Kadono (2014).

To describe the water quality for each macrophyte community type (for details, see section “Classification of the Macrophyte Communities in Ponds”), we measured six physicochemical parameters of the pond water (Secchi-disk transparency, pH, suspended solids, and concentrations of total nitrogen, total phosphorus, and chlorophyll a) once late in the same growing season when the macrophytes were surveyed. Water samples for chemical analysis were collected at a depth of 0.5 m at the approximate center of each pond, and field measurements were conducted at the same location (see Supplemental Material A for detailed methods). Unfortunately, water-quality data for 17 ponds were lost due to an accident, and thus water quality for each community type was summarized based on the remaining 352 ponds.

To evaluate the relative importance of landscape and local variables in determining the macrophyte community composition, we conducted a two-step analysis. Unlike a simple ordination analysis, this approach allows us to visualize a possible non-linear relationship. In addition, the results are easier to interpret for the non-specialist unfamiliar with interpreting biplot output, which is an important consideration when the goal is to support practical conservation planning.

In the first step, we clustered ponds according to their macrophyte community. To conduct the cluster analysis, we used the Jaccard distance measure and Ward’s linkage method. We identified the appropriate number of groups (we refer to each group as a “community type”) by examining the dendrogram while considering the need for parsimony and ease of interpretation. We tested for significant differences among the communities identified by the cluster analysis by using analysis of similarities (ANOSIM; Clarke, 1993), and used the Jaccard distance measure and 10,000 permutations in this analysis. Ponds with three or more macrophyte species (148 ponds) were subjected to the cluster analysis, whereas ponds with one or two macrophyte species (129) and those with no macrophytes (92) were treated as independent community types. These latter two community types were integrated with the community types identified by the cluster analysis in our subsequent analyses.

In the second step, we assessed the relationship between the occurrence of community types, including those identified in the cluster analysis and the two additional ones, and the five environmental factors. To do so, we modeled the occurrence of each community type using multinomial regression (Venables and Ripley, 2002). Landscape factors (TWI and proportions of urban area and paddy field) and local factors (pond surface area, proportion of artificial embankment) were used as the explanatory variables. We also added TWI² as an explanatory variable, because preliminary analysis indicated that the relationship between the occurrence of community types and TWI can be unimodal. We constructed 64 (= 2⁶) candidate models that included all possible combinations of the explanatory variables. The model with the lowest value of Akaike’s information criterion (AIC; Burnham and Anderson, 2002) was selected as the best-fit model. We further ranked the relative contribution of each variable by calculating the corresponding Akaike parameter weight (Burnham and Anderson, 2002), which indicates the relative strength of the relationship between the variable and the response variable (i.e., community type) and represents the probability that the variable is included in the actual best model (Burnham and Anderson, 2002). Significance of each variable was also tested with a log-likelihood ratio test. We did not explicitly analyze the geographic location of each pond, because we found that Moran’s spatial autocorrelation coefficient (I) in the residuals on the probability of the best-fit model was low (I < 0.18) and most were non-significant (e.g., Van Langevelde and Wynhoff, 2009; Supplemental Material B). The magnitude of the correlations (|r|) between the explanatory variables was <0.54, except for the relationship between TWI and TWI². Although we measured water-quality variables in 352 of the 369 ponds, these were not included as explanatory variables because of the well-reported effect of the five explanatory variables on the water-quality variables (e.g., Verhoeven et al., 2006); the inclusion of non-independent variables in a linear model as explanatory variables violates the fundamental assumption of dependence, which would not only disrupt parameter estimation but also complicate the interpretation of the responsible process and causal relationships.

We compared the six physicochemical parameters of pond water quality for each community type by using a general linear model. We set the link function of the general linear model as identity for pH and as log₁₀ for the other parameters. For each parameter, we created models that included 52 possible combinations of the community types by merging the five community types identified by the cluster analysis (for details of the community types, see section “Classification of the Macrophyte Communities in Ponds”). The most plausible combination was determined by choosing the one with the lowest AIC (Burnham and Anderson, 2002).

All statistical analyses were conducted by using R version 3.4.2 (R Development Core Team, 2017). We used function hclust in package stats (R Core Team and Contributors Worldwide, 2017) for cluster analysis, adonis in package vegan (Oksanen et al., 2011) for ANOSIM, multinom in package nnet (Ripley, 2016) for multinomial regression, dredge in package MuMIn (Barton, 2018) to create and estimate all possible combinations of explanatory variables in the multinomial regression, Anova in package car (Fox and Weisberg, 2017) for likelihood ratio tests, and lm in package stats (R Core Team and Contributors Worldwide, 2017) for the general linear model.

The 369 surveyed ponds captured a wide range of landscape and local features of ponds (Table 1). In total, we found 48 macrophyte species in the ponds, with the number per pond ranging from 0 to 14 (mean ± SD, 2.55 ± 2.67). However, 92 ponds (28%) had no macrophytes and 129 (40%) had only one or two species (Table 2). Trapa japonica was the most dominant macrophyte species in the surveyed ponds (observed in 174), followed by Potamogeton octandrus Poir. (58), Nymphoides indica (L.) Kuntze (58), and Utricularia australis R. Br. (55).

The cluster analysis divided the macrophyte communities with three or more species into three community types (Figure 2). Type I frequently harbored the species Nuphar saikokuensis Shiga et Kadono (observed in 67% of the Type I ponds), B. schreberi (67%), U. australis (64%), and N. tetragona (62%; Table 2). Type II most often included N. indica (97%), Myriophyllum ussuriense (Regel) Maxim. (56%), and T. japonica (50%). Type III most often contained T. japonica (79%), Utricularia aurea Lour. (41%), and H. verticillata (39%). These three community types differed significantly in their species compositions (pairwise ANOSIM, P < 0.001). We also defined Type IV (one or two species) and Type V (no macrophytes). Type IV frequently harbored T. japonica (64%; Table 2).

The six physicochemical water parameters had a wide range of variation, suggesting that the surveyed ponds cover a broad portion of the eutrophication gradient. All the parameters showed significant differences among the five community types (Table 3). Overall, Type V without macrophytes occurred in more eutrophic waters, represented by low water transparency and high contents of total nitrogen, total phosphorus, chlorophyll a, and suspended solids, followed by Type IV, and then by the remaining three Types (I to III). Only Type V showed an alkaline pH, and Type I had the most acidic water (Table 3).

Among the 64 candidate models that delineated the relationship between the community type in a pond and the six explanatory variables, the model that included all of the variables (including TWI²) was selected as the best-fit model, and all six variables were significant (Table 4). The ΔAIC of the second best-fit model (i.e., the difference in AIC from the that of the best-fit model), which employed all variables except proportion of artificial embankment, was 3.36, and those of the other 62 models were >7.7, indicating that all the variables that we selected play some role in determining the community type.

Transition among the three species-rich community types (Types I to III) was apparent among the two natural variables (i.e., TWI and surface area; Figure 3). The relationship between TWI and the probability of occurrence of Type I was unimodal, and the occurrence of Type I maximized at TWI = 7.3 (Figure 3A). After a slight initial decrease, an increase in TWI increased the probability of occurrence of Types II and III until peaking at TWI = 10.6 and 9.8, respectively, and both then gradually decreased. Increase in surface area decreased the probability of occurrence of Type I, whereas it increased the probabilities of occurrence of Types II and III (Figure 3D). In contrast, the three anthropogenic variables (i.e., proportions of urban area, paddy field, and artificial embankment) collectively decreased the probability of occurrence of Types I to III, except for slight increases in the occurrence of Types II and III with increasing proportion of artificial embankment (Figures 3B,C,E). The declining trend was sharpest for Type I for all three anthropogenic variables.

Overall, the probabilities of occurrence of Types IV and V (few or no macrophytes) increased with increases in the three anthropogenic variables (Figures 3B,C,E). In particular, a clear increase in the occurrence of Type V was observed as the proportion of urban area increased (Figure 3B), whereas that of Type IV was evident along with an increase in the proportion of paddy field (Figure 3C). Consequently, over 85% of the ponds were predicted to harbor either of these two community types when the proportion of urban area or paddy field equals 1. In contrast, the occurrence probability of these two types gently increased with increasing proportion of artificial embankment (Figure 3D).

To conserve regional pond biodiversity, previous studies noted that conservation actions must be taken not only on a focal pond but also in areas surrounding the pond (Kirchner et al., 2003; Williams et al., 2004; Declerck et al., 2006; Akasaka and Takamura, 2011, 2012). Although these studies indicated the need for conservation plans and actions focused on a larger spatial extent than an individual pond, they provided no basis to determine the actual spatial extent needed for conservation plans. In that sense, the transition of macrophyte community types along the gradient of topographic configuration that we detected provides a potential guideline, indicating that the spatial extent of a conservation plan should cover an area that includes the full gradient of topographic configurations (i.e., covering ponds from those in the upper hilly landscape to those in the lower flat landscape) in the region. In practice, we propose first zoning the area based on the topographic configuration (i.e., TWI value), and then selecting conservation priority ponds within each zone. Multiple ponds need to be conserved in each zone because β diversity of aquatic macrophytes (variation in species composition among ponds) is substantially larger than their α diversity (mean number of species in a pond) even if the topographic configuration is similar (Akasaka and Takamura, 2012). Based on our findings, when determining the zones, a TWI value around 7.3 is a plausible partitioning value to delineate a zone that targets macrophyte communities occurring in acidic, relatively oligotrophic water (i.e., Type I). Further research in different regions is required to determine widely applicable partitioning value(s).

Because urbanization and expansion of agriculture area strongly regulated the emergence of species-rich macrophyte communities, restriction of land development around the targeted ponds is a key issue for successful conservation of regional pond biodiversity. Such restrictions might be implemented by local and national governments, for example, by setting aside the areas for conservation or recommending eco-friendly rice farming while providing subsidies.

At a local scale, conservation action will be necessary to prevent further establishment of artificial embankments, which destroy macrophyte habitat (Nishihiro et al., 2006a). In some cases, however, to maintain the strength of a pond dike and to prevent disaster caused by collapse of the dike, artificial embankments are unavoidable. The use of the seed bank in the sediments (Nishihiro et al., 2006b) might mitigate the impacts caused by such construction and might restore the target macrophyte communities. Pond managers may want to integrate two or more small adjacent ponds to create a larger pond in order to increase the amount of water storage and reduce the labor cost for pond management (Takamura, 2012). To maintain existing macrophyte communities, however, the decision to enlarge a pond should be made carefully, especially when the pond is small and located in a less urbanized hilly area, because pond surface area significantly influences the macrophyte community. Although our results highlight the greater impact of anthropogenic land use as compared to that of artificial embankment, we believe that local action should continue to limit increases of artificial embankment because habitat provision is the basis of conservation as well as restoration. Joint efforts among entities that mainly work on different spatial scales (i.e., national and local governments, conservation groups, and pond managers) will be key to ensuring regional pond biodiversity.