We conducted the field survey in the Kyushu of southern Japan (Fig. 2a,b). The rice-transplanting schedule in Kyushu, is generally late (in late June24) because it is necessary to delay rice-transplanting timing after the harvest of wheat as a back crop22,23. However, since the technique of early transplanting was established around 1960, early transplanting (in early April24) has been practiced instead of a back crop of wheat in some areas in Kyushu25,26, so the rice-transplanting schedules differ at the regional scale. We established two study regions with early-transplanting schedules (Karatsu in northern Kyushu; N33.4, E129.9 and Amakusa in central Kyushu; N32.5, E130.1), and two study regions with late-transplanting schedules (Itoshima in northern Kyushu; N33.6, E130.2 and Uki in central Kyushu; N32.6, E130.8) (Fig. 2c–f). Each study region is 10–20 km2, and the distances between the regions are 20–140 km. Climatic conditions between the regions are similar (Supplementary Table S3). The size of each region is much larger than the size of the typical territory of breeding buzzards (approximately 500 m radius from the nest18).
We examined the distribution of buzzards in 2019. Buzzards are migratory birds that breed in Japan, northeastern China, and the Russian Far East in summer, and overwinter in the Ryukyu Islands, Southeast Asia, and southern China27. In our study regions, buzzards start breeding in April, soon after returning from their wintering area. Buzzards incubate their eggs from late April until late May when they hatch. Once the eggs hatch, buzzards feed nestlings. Then nestlings start to fledge in late June, but the adults continue feeding their fledglings for several weeks. Buzzards migrate to their wintering grounds around October. The breeding season of buzzards thus overlaps largely with the rice production season, but there is a slight but significant differences in seasonality, i.e., paddies are already planted and flooded before hatching in early rice-transplanting schedules, while not yet flooded in late rice-transplanting schedules.
Because breeding buzzards were thought to prefer mosaic landscapes of farmland and forest17,19, we established 62 study blocks that included edges between farmland and forest in each study region (Fig. 2c–f; northern-early: 17 central-early: 11 northern-late: 17 central-late: 17). The study blocks were 400 m square and located at least 700 m apart from each other, a distance determined from the knowledge that buzzards intensively use an area of 200 m from their nests18. To examine the presence/absence of breeding buzzards in each block, we conducted 2 days of 30-min observations during the breeding season (April to July). Using a pilot study, we determined that this observation time was enough to minimize the possibility of missing buzzards. We identified breeding individuals based on displays, feeding behaviors, and territorial behaviors.
Land use survey
During the brood-rearing period (late June to early July) in 2019, we recorded the land use (forest, grassland, flooded paddies, non-flooded paddies) in each block. We then used these data to create a land use map in each block using QGIS3.16.228 and overlaying it on Google Earth aerial photographs in 2017.
Prey species survey
We surveyed the distribution of prey species in paddies and grasslands in the study regions in 2019 and 2020. Based on the previous studies on the feeding habits of buzzards17,18,19,20,21, we surveyed the distribution of frogs and orthopterans larger than 3 cm as prey of this size is considered their main prey. We established survey transects in paddies and grasslands in our study blocks. We conducted surveys twice each year during the brood-rearing period (late June to early July), when breeding buzzards need a large amount of prey. We walked along the transects and counted the number of prey species observed within 0.5-m of both sides. This survey method is suitable to assess prey availability29 because buzzards visually search for prey (e.g.20,30). A total of 148 20 m-transects were placed in paddies in 34 blocks and 157 15-m transects were placed in grasslands in 37 blocks, and each transect was surveyed in both or either 2019 and 2020. In the paddy transects, we recorded the height and coverage of vegetation, the ditch characteristics (none, concrete ditch, earthen ditch), the surrounding land use (10 m width from the transect: flooded paddy, non-flooded paddy, grassland, forest, stone wall, and road), and flooding or non-flooding in the paddy field adjacent to transects. In grassland transects, we recorded the height and coverage of vegetation and the grassland types on which the transect was located (abandoned land, orchard, farmland, bank, forest edge).
To investigate the habitat selection of buzzards, we used a generalized linear model with a binomial error distribution. We used the edge length between the landscape elements and forest as independent variables, because the edge length, rather than the area of the landscape elements, is known to be an important determinant for buzzard distribution19. We prepared a land use map and calculated the edge length between the landscape elements and forest by using the field calculator of QGIS, and values were standardized (mean = 0 and SD = 1).
To explore the variation in habitat selection across regions with different transplanting schedules, we first used a model that included the interaction term of transplanting schedules and landscape elements as independent variables. The length of paddy-forest edges, grassland-forest edges, the transplanting schedules, the interaction term of the edge length and the transplanting schedules, and the study regions were included as independent variables (details of independent variables: Supplementary Table S4). The presence/absence of breeding buzzards was included as a dependent variable. We analyzed the full model and all sub-models containing different combinations of all independent variables, including the null model. We regarded models that had ΔAIC values (the difference between the AIC value of the focal model and that of the best-fit model) of < 2 as competing models and conducted model averaging31 for them. Model averaging was performed using the natural average method to avoid shrinkage of coefficient values32. Independent variables were considered influential when the 95% confidential intervals did not cross zero32.
As the interaction term turned out to be significant, we then analyzed the models separately for the early and the late transplanting regions. We included the length of paddy-forest edge, length of grassland-forest edge, and study region as independent variables (Supplementary Table S4), and conducted model averaging as in the first model. We confirmed that correlation coefficients of all pairs of independent variables were < 0.6, indicating no serious multicollinearity.
Prey species model
To clarify the relationship between frogs abundance in paddies and surrounding landscape, we used a generalized linear mixed model with a zero-inflated Poisson error distribution. We included the height and coverage of vegetation, shape of ditch adjacent to transects, surrounding land use, area of the fields with transects, flooding conditions in the fields, survey year and study region as independent variables, and study blocks as a random variables (Supplementary Table S4). We standardized the independent variables (mean = 0, SD = 1) to allow direct comparisons between estimated model effects. Since rice frogs (Fejervarya kawamurai) were abundant in the prey species surveys, we used the abundance of rice frogs as the dependent variable. We conducted model selection and model averaging using the same method as in the buzzard model above. We excluded models that included independent variables that have high correlation coefficients (> 0.6) to avoid serious multicollinearity.
For grasshoppers, we also used a generalized linear mixed model with zero-inflated Poisson distribution. We included the height and coverage of vegetation, land use of the grassland with transects, the survey year and the study region as independent variables, and study blocks as a random variable (Supplementary Table S4). We standardized the independent variables (mean = 0, SD = 1). Since Gampsocleis buergeri was by far the most abundant in the prey species surveys in grasslands, we used their abundance as the dependent variable. We conducted model selection and model averaging using the same method as the buzzards’ models. As above, we excluded the models that included independent variables that were highly correlated (> 0.6) to avoid serious multicollinearity.
We performed all analyses in R 4.0.333, using the glmmTMB packages34 for model fitting, the MuMIn package35 for model selection and averaging, and the ggplot236 for graphic illustration or results.
Source: Ecology - nature.com