Study site
In this study, a total of 20 floodplain ponds were surveyed in the Tokachi region, including Ikeda and Toyokoro towns, in northern Japan (42° 78' 60"–90' 27" N, 143° 42' 82"–59' 82" E). The average annual temperature in this region is 5.8°C and the average annual precipitation is 869.7 mm (1981–2020 average: Japan Meteorological Agency 2019). The floodplain ponds surveyed in this study are located in the lower part of the Tokachi River, and the surrounding landscape is dominated by farmlands, such as croplands and pastures (Fig. 1; Fig. 2). Many floodplain ponds are interconnected through agricultural ditches. Historically, several floodplain ponds were distributed along the meandering main watercourses of the lower Tokachi River. In the 1880s, however, the Tokachi River was straightened as a flood control measure, and most protected inland areas along the river were converted to cities and farmlands (Okuyama & Fujomaki 2001). Consequently, overbank flooding rarely occurred in the study region. We used Aerial photographs to determine whether the floodplain ponds surveyed in this study are remnant floodplain ponds or those created as a result of river channelization (Geographical Survey Institute 2020). For the surveys, floodplain ponds were selected by considering a wide range of variations in the area of floodplain ponds and the degree of hydrological connectivity.
Environmental factors
Local- and landscape-scale environmental variables were selected to explain the structure and composition of the aquatic communities (that is, species richness and species coverage). Local-scale environmental variables were measured at the same time as the vegetation survey, and landscape-scale environmental variables were obtained from the National Land Numerical Information (Geographical Information Authority of Japan 2005) and 1:25,000 scale vegetation map GIS data (Ministry of Environment 2017).
The local-scale environmental variables were the physical environment (water depth, water depth variation, area of water surface, and turbidity) and nutrient indices (dissolved total nitrogen (DTN) and dissolved total phosphorus (DTP) concentrations). For each quadrat in each floodplain pond, water depth and turbidity were measured using an aluminum staff and a multi-item water quality meter (WQC-24, DKK-TOA Co., Ltd., Tokyo, Japan), respectively. The average values of these measurements for each floodplain pond were used as local-scale variables. The water depth variation in each floodplain pond was calculated as the standard deviation of the measured water depth. The water surface area of each floodplain pond was calculated using the data available from the National Land Numerical Information (Geographical Information Authority of Japan 2005) using QGIS. To analyze DTN and DTP, surface water was collected at five locations per floodplain pond and the samples were immediately filtered using glass fiber filter paper (0.7 µm, GF/F, GE Healthcare, Chicago, the United States). The sample filtrate was transferred to the laboratory and stored at -18°C until further analysis. Subsequently, DTN and DTP in the filtrate were analyzed using a flow-injection analyzer (AACS-4, BL-TEC Inc., Osaka, Japan).
Landscape scale environmental variables included the connectivity of watercourses and land use ratio around the floodplain ponds. Connectivity was represented by the decrease in the integral index of connectivity (dIIC). The dIIC considers one habitat as a connecting element between other habitats. It can also be calculated without knowing the coefficient of dispersion, which is specific to the target species (Baranyia et al. 2011). The dIIC was calculated as follows:
where i,j represent any floodplain pond combination, ai represents the area of the floodplain pond i, AL represents the total area of all floodplain ponds, and nlij represents the number of links in the shortest paths between floodplain ponds i and j.
The IIC represents the connectivity of floodplain ponds as an entire landscape (all floodplain ponds along the lower Tokachi River). The value of dIICk represents the percentage reduction in the IIC that occurs when wetland k is lost (that is, the importance of wetland k in the entire floodplain pond network); floodplain ponds with larger dIICk values contribute more towards maintaining the network. The dIIC can be calculated based on the length of the watercourses, assuming that the floodplain ponds are functionally connected. This length is called the threshold distance and can be set on the basis of the territory and dispersal distance of living organisms (Baranyia et al. 2011). However, the dispersal distance of aquatic plants through the watercourses in the study area is unknown. Therefore, based on the studies conducted by Ishiyama et al. (2014, 2015, 2020), the threshold distances ere determined as 0.5, 1, 3, 5, 7.5, 10, 12, and 14 km. To evaluate the importance of connectivity exclusively through the watercourses, the distances were set as less than or equal to 14 km, which did not include the main river channel. The dIIC was calculated using Conefor 2.6 software (Saura & Torné, April 2012).
The percentage of farmland, urban, and farmland + urban area around the floodplain pond were calculated for the analyses, as we assumed that the land use around the studied floodplain ponds would affect the nutrient conditions in the water. The outer buffers by stage were determined as 10, 50, 100, 500, and 1000 m from the pond edge to detect the most influential spatial scale for each nutrient index. Prior to that, we used the data obtained by extracting and reclassifying the corresponding land use from 1:25000 scale vegetation map GIS data (Ministry of Environment 2017). We used QGIS (version 2.18.24) for all GIS analyses. The environmental variables were subjected to natural logarithmic transformation to improve normality and standardized to make different units comparable.
Data analysis
Structural equation modeling (SEM) was used to investigate the factors affecting aquatic plant communities in the surveyed areas. The species richness and coverage ratio for each aquatic plant species were used as the objective variables (Fig. 3). The explanatory variables were selected from the following 29 environmental factors. The landscape-scale variables were dIIC (0.5, 1, 3, 5, 7.5, 10, 12, and 14 km) and land use (farmland, urban, farmland + urban ratios for 10, 50, 100, 500, and 1000 m buffers). The local-scale variables were the nutrient indices (DTN and DTP) and physical environment information (water depth, water depth variation, turbidity, and area). However, with respect to the physical environment, only water depth variation could be selected in the SEM; the objective variable for this was the species richness of aquatic plants. Owing to the small sample size (n = 20), all variables included in this study could not be analyzed. The fully developed SEM model consisted of one variable from the dIIC, 1–2 variables from land use, and 0–1 variable from the physical environment categories based on our sample size. Subsequently, then ran the SEM was run by swapping one variable for each category (1–2 for land use, 0–1 for physical environments) to detect the model with the highest fitting degree (that is, lowest Akaike's information criteria). Model fitting was also checked with the RMSEA (root mean square error of approximation) < 0.06. The software R ver. 3.5.1 (R Core Team 2018) and the lavaan package ver. 0.6–5 (Rosseel 2019) were used for the analysis.