To investigate the distribution of suitable habitats for the Japanese giant salamander, we defined the study area, based on the actual range of the species provided by the IUCN (IUCN SSC Amphibian Specialist Group 2022), covering the western part of Japan (western Honshu Island, Kyushu and Shikoku) (Fig. 1). The eastern limit of the actual extent of the species corresponds to the ‘Japanese Alps’, a series of mountain ranges bisecting Honshu Island and supposed to represent a natural barrier to the distribution of the species. To investigate the potential suitable areas in Japan and to verify if the natural barrier could prevent the species from reaching favourable areas to the north of Japan, we included the entirety of the Honshu, Shikoku and Kyushu islands in the study, which extend eastwards from 129.34°E to 142.06°E and northwards from 30.99°N to 41.54°N, covering a total area of 282,916 km².

The occurrence data of Japanese giant salamanders were obtained from our field surveys as well as from previously published studies in Hyogo, Hiroshima, Gifu, Mie, Nara, Oita, Osaka, and Tottori Prefectures (Kuwabara et al. 2005; Tochimoto et al. 2007; Okada et al. 2008; Taguchi and Natuhara 2009; Bjordahl et al. 2020). In total, we gathered 1,228 presence records of Japanese giant salamanders. These data were collected between 1970 and 2024 (primarily after 1995) and align with the data range of the environmental predictors used in the study. Only high-resolution occurrence data (mainly GPS points) were used in this study, which fall within the grid cells used for environmental data. To avoid spatial autocorrelation issues (i.e. overfitting bias) in establishing the SDMs (Dormann et al. 2007, 2013), the data were filtered using the “spThin” package in R (Aiello-Lammens et al. 2015). Spatial filtering of the data is considered to be a robust correction method for SDMs (Kramer-Schadt et al. 2013; Boria et al. 2014; Fourcade et al. 2014). We randomly selected occurrence data localities that were at least 1 km apart from each other. We selected this distance because this study was conducted at a 1 km² grid scale and because the mean potential migratory distance of Japanese giant salamanders lies typically within this spatial scale (Taguchi 2009). After the thinning procedure, we obtained and used 477 presence data to run the models (Fig. 1).

To assess the distribution of the Japanese giant salamander, we used the presence data and their corresponding environmental characteristics, considering climatic, topographic and land-use predictors at a 1 km² resolution. As it is recommended to avoid too many predictors in the same models and to focus on variables likely to influence the distribution of the species (Ficetola et al. 2014; Fourcade et al. 2018), we used a limited set of environmental variables to perform the models, selected according to the Japanese giant salamander’s ecology (Okada et al. 2008; Harris et al. 2013; Bjordahl et al. 2020; Soley-Guardia et al. 2024). Four climatic variables were selected among the 19 bioclimatic variables of WorldClim version 2 climate dataset (Fick and Hijmans 2017) at a high (1 km²) spatial resolution: mean temperature of the wettest quarter (Bio8), mean temperature of the warmest quarter (Bio10), precipitation of the warmest quarter (Bio18) and precipitation of the coldest quarter (Bio19). Bio8 and Bio10 were selected because high temperature may be a critical condition for the survival of the Japanese giant salamander. An increase in air temperature is indeed expected to raise water temperature as well as reduce dissolved oxygen concentration, which is a major parameter for salamander respiration. Bio18 and Bio19 were included because the Japanese giant salamander is entirely aquatic and lives in stream ecosystems; precipitation is, therefore, a critical parameter for survival. The mean slope of each 1 km² grid cell was selected over mean altitude as a topographical predictor because mean altitude was highly correlated with temperature predictors and to focus on the specific aspects of slopes (e.g. on water flow). Agricultural (including all types of agricultural activities) and urban areas (i.e. considering all kinds of man-made artificial areas) were included in the model processing as land-use variables. Japanese giant salamander populations can be found in various types of environments, such as agricultural areas and they are also occasionally found in densely built urban areas (Okada et al. 2008). We also made sure to use uncorrelated predictors. The Pearson correlation coefficient was used to assess multicollinearity between selected variables (all selected variables with R < 0.558; Suppl. material 1: table S1; Suppl. material 2: table S2 for details on each data sources) (Elith et al. 2006; Fourcade et al. 2018).

SDMs for the Japanese giant salamander were developed using MaxEnt 3.4.1 (Phillips et al. 2006). This is a presence – background method comparing the relationship between environmental conditions of species-specific locations to the conditions along the entire background (Phillips and Dudík 2008). MaxEnt has the advantage of balancing goodness-of-fit with model complexity and being relatively easy to calibrate (Muscarella et al. 2014). To carry out distribution modelling, we followed recommendations to avoid violation of key assumptions in SDMs (Sillero and Barbosa 2021; Sillero et al. 2021; Soley-Guardia et al. 2024). To ensure that the model captures the environmental conditions specific to the current distribution of the Japanese giant salamander, we used only the known range of the species and restricted the background to this area during the model training phase. Once the model was developed, we projected it on to a broader area, including the northern part of Honshu Island, to explore potential habitat suitability beyond the current range. The selection of background (or pseudo-absence) data localities for SDMs is an important consideration. SDMs such as MaxEnt require data of species presence and also of the available environmental conditions (background) in the area (Barbet-Massin et al. 2012). The choice of the spatial extent from which background data are taken can seriously affect model performances and variable importance (VanDerWal et al. 2009; Elith et al. 2010; Fourcade et al. 2014). As there was a sampling bias concerning our occurrence data, we used a bias file implemented in the MaxEnt models to also apply a bias correction on the selection of the background. We created a bias grid by delineating a Minimum Convex Polygon (MCP) around occurrence localities buffered by 50 km to also more broadly consider the watersheds in which the points of occurrence bordering the MCP are located (Peterson et al. 2011). In total, 10,000 background points were then randomly sampled from the selected background extent.

We considered ‘masked geographically structured rules’ (Radosavljevic and Anderson 2014) to partition occurrence data and the environmental predictors into training and validation data subsets. We chose to use the spatial partitioning ‘block’ method that increases model transferability across space and time. This is a method that fits our study objectives on projecting our SDM into a novel spatial extent (north of Japan) and estimating the impact of future climate change on species distribution and habitat suitability (Muscarella et al. 2014). Species occurrence data were partitioned in four parts (i.e. through a 4-fold spatial partition) by the latitude and longitude and both occurrence and background data were then attributed to a spatial group depending on their location. This method reduces spatial autocorrelation between training and validation data (Roberts et al. 2017).

Model tuning and selection were performed using the R package ENMeval 2.0 (Kass et al. 2021). Two main settings were used to adjust model fitting: features classes (FC), which regulate the format of the marginal response curves and the relationship between environmental and background data and regularisation multipliers (RM), representing a penalty of including variables in the model. For example, models with complex FC associations and lower RM values result in more complex model predictions (Elith et al. 2011; Phillips et al. 2017).

We tested different combinations of simple and complex feature classes – L, LQ, H and LHQ (L = linear, Q = quadratic, H = hinge). We excluded the “product” and “threshold” feature classes as their omission results in easier to interpret models (Kass et al. 2020). For each feature class combination, we tested nine regularisation multipliers (1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5). This parameter control model complexity, with small values leading to overly complex models and high values to more simple models (Radosavljevic and Anderson 2014). The different RM values combined with the feature classes resulted in a total of 36 parameter combinations. Model selection was performed using the corrected Akaike Information Criteria (AICc) (Akaike 1974). Top models with ΔAICc < 2 were selected (Burnham and Anderson 2002). If several models were selected, we chose the models with the highest AUC ratio for the partial ROC performance metric (Peterson et al. 2011). To highlight the relationship between habitat suitability and environmental predictors for the Japanese giant salamander, we constructed species response curves for each environmental variable.

As SDMs selected by minimising AICc may not perform well on withheld data (Galante et al. 2018), we determined the confidence and effect sizes of the performance metrics of the tuned model by comparing them to those calculated with null models. If the performance metrics of the empirical model did not differ from metrics calculated with the null models, the confidence in the evaluation metrics was considered as low (Raes and ter Steege 2007). We built null models using the same occurrence data, with the same spatial partitioning parameters (4-fold partition) as the empirical model and using 100 iterations to create different null models. For each iteration, n localities across the training folds were selected, where n is equal to the total number of real occurrence localities in these folds and then evaluated the model on the real Japanese giant salamander occurrences in the withheld fold (also using the same background values). We used a L (linear) feature class and a regularisation multiplier of 5 to create the null models, which correspond to simple models. Null models were generated using the R package ENMeval 2.0 (Kass et al. 2021).

We generated future projections of the Japanese giant salamander distribution for three periods: 2050 (average for 2041–2060), 2070 (average for 2061–2080) and 2090 (average for 2081–2100). We considered two shared socioeconomic pathways (SSPs) scenarios (i.e. climatic scenarios), based on the representative concentration pathways (RCPs): SSP126, a remake of the RCP2.6, equal to the optimistic scenario (i.e. with a reduction of carbon dioxide emissions in the future) and SSP370, a scenario between RCP6.0 and RCP8.5, which represents the pessimistic scenario (i.e. with a continuous increase of carbon dioxide emissions in the future) in our study. We averaged the climate data among four global circulation models (GCMs) ­– CMCC-ESM2, MIROC6, IPSL-CM6A-LR and MRI-ESM2-0 – to reduce the uncertainties among the models in order to produce the maps of future habitat suitability of the Japanese giant salamander. Future climate projections for all models were retrieved from WorldClim version 2 (Fick and Hijmans 2017). In addition, we ran multivariate environmental similarity surface analysis (MESS) to evaluate and identify potential high extrapolation risk when projecting the model to new areas and new periods (Elith et al. 2010). MESS measures the similarity between the climatic conditions at a point in time and the reference layer. Thus, we ran a MESS analysis to compare the climatic variables of the reference layer (corresponding to the extent of the bias grid created to sample the background) that we used to develop the model and the climatic variables of the projection layer (all Honshu, Shikoku and Kyushu Islands) and we also ran six MESS to compare the present climatic conditions of all Japan and the future conditions (six different future conditions). All the MESS analysis highlighted small areas of extrapolation, meaning a high environmental similarity for each condition tested and, therefore, a low extrapolation risk when projecting the model. One example of MESS results is presented in the supplementary materials (Suppl. material 3: fig. S1).

The selected model for current habitat suitability and the projections were mapped on to the study area using a log-log (‘cloglog’) approximation of the probability of presence of the species (Phillips et al. 2017). In order to calculate the surface areas of areas favourable to the species, probability values (ranging from 0 to 1) were classified into four categories using the Jenks natural breaks classification method (Jenks 1963): unsuitable (0–0.11), low suitability (0.11–0.37), moderate suitability (0.37–0.72) and high suitability (> 0.72). Classification and map rendering were performed using ArcMap 10.8.1 (ESRI) and QGIS 3.34.6.

The selected MaxEnt model was mainly influenced by climatic predictors with a total contribution of 76.21%, which was split into 36.73% for the precipitation of the warmest quarter (Bio18), 13.77% for the precipitation of the coldest quarter (Bio19), 13.54% for the mean temperature of the warmest quarter (Bio10) and 12.17% for the mean temperature of the wettest quarter (Bio8). Precipitation and temperature variables contributed to 50.5% and 25.71% of the model, respectively. The contribution of the mean slope was of 11.62%. Land-use predictors (urban and agricultural) contributed to 9.36% and 2.81%, respectively. All contribution values are based on the permutation importance of each variable.

According to the response curves of the five most contributing environmental predictors, the probability of presence of Japanese giant salamanders was the highest in areas with an average air temperature around 20–24°C, whereas all areas between freezing average temperature and 20° air temperature, as well as those warmer than 24°C, were less favourable. The probability of presence of the Japanese giant salamanders is high at relatively low average precipitation, with increase of precipitations favourable in the coldest quarter (around 100–300 mm) and unfavourable in the wettest quarter, with a sharp decrease of suitability above 500 mm precipitation. Beside climatic variables, the probability of presence increased with the mean slopes of the surrounding lands up to 17% but decreasing for steeper areas (Fig. 2). The two land-use variables (agricultural and urban areas) were the two least contributing variables (Suppl. material 5: fig. S3).

The current habitat suitability map for the Japanese giant salamander (Fig. 3) highlights the presence of highly suitable areas for the species both within and beyond the current range of the species. Focusing on the current distribution area of the species (59,193 km² in total), 30% were identified as unsuitable (17,712 km²), 37% with low suitability (21,853 km²), 22% with moderate suitability (13,166 km²) and 11% with high suitability (6,462 km²). In the area beyond the range of the species (223,723 km²), unsuitable areas represent 36% (81,276 km²), low suitability 30% (66,073 km²), moderate suitability 15% (33,866 km²) and high suitability 19% (42,508 km²).

Figure 2. 

Response curves from the MaxEnt model of the five predictor variables contributing the most to the habitat suitability of the Japanese giant salamander. For each response curve, the grey area represents the range of values observed in Japan and the dotted brown vertical lines delineate the minimum and maximum values of the occurrence data. Photo of Japanese giant salamander by M. Denoël.

Figure 3. 

Habitat suitability prediction map of the Japanese giant salamander according to the optimal selected MaxEnt model. Brown thick contour lines represent the current distribution of the Japanese giant salamander according to the IUCN.

The results of the projections of future changes in habitat suitability (Fig. 4) indicate that climate change is expected to lead to a restriction of the current distribution ranges. Decreases in habitat suitability and range expansion are predicted both within and beyond the currently occupied areas (Fig. 4). Considering the extent of the current distribution of the Japanese giant salamander, there will be a loss of habitat suitability in more than 50% of the geographic area at any time in the future (2050–2090) and this, for both future climate scenarios (Fig. 5). The loss of suitable habitats is more important when considering the pessimistic scenario for all time periods, with the most important loss (in 64.5% of the area) predicted in 2090. In 2050, the proportion of areas gaining in suitability is higher in the pessimistic scenario (13.6%) than in the optimistic scenario (11.1%). The largest increase in suitability is observed in 2070 and 2090 with the optimistic scenario, with an increase in 14.5% and 14.3% of the area, respectively. It is slightly lower for the pessimistic scenario for the same periods (10.7% and 9.1%, respectively).

Figure 4. 

Projection maps of changes in habitat suitability of the Japanese giant salamander for three time periods considering an optimistic (SSP126) and a pessimistic scenario (SSP370) for future climate change. Brown lines represent the current distribution of the Japanese giant salamander according to the IUCN. Change in habitat suitability was determined by subtracting the current habitat suitability (Fig. 1) from the predicted future habitat suitability raster at 1 × 1 km resolution (Suppl. material 6: fig. S4).

Figure 5. 

Sankey diagrams of the predicted evolution of habitat suitability within the current distribution range of the Japanese giant salamander for each time period (2050, 2070 and 2090) and climate change scenario (SSP126: optimistic scenario; SSP370: pessimistic scenario). A gain of habitat suitability was considered when the change in habitat suitability was > 0.1, a loss was identified when the change was < -0.1 and stability when the change was between -0.1 and 0.1.

Of the seven tested environmental predictors, six had a contribution to the distribution of the species greater than 10% (Fig. 2) and the four most contributing variables are climatic. They show that precipitation and temperature are important drivers of the distribution of the species. More specifically, precipitation of the warmest and the coldest seasons are the most contributing variables. Precipitation is closely related to the living environment of giant salamanders. The amount and periodicity of precipitation affect water quantity and stream velocity (Allan et al. 2021), a critical factor for salamanders as cutaneous respiration is essential to their survival. An increase in water velocity and mixing facilitate the dissolution of oxygen in water (Ultsch 2012). However, our results also indicate that excessive precipitation could be a detrimental factor for salamander presence, as strong currents or prolonged flooding are likely to negatively affect their populations. Such large excesses of water flows specifically occur during the warmest months. Studies investigating the effects of flooding on salamander populations, including the hellbender, highlight that flood events particularly affect larval stages, leading to a decline in adult recruitment (Nickerson et al. 2007; Walls et al. 2013). This may also interfere with nesting – a key element in the life of giant salamanders where the male guards the eggs in a submerged cavity, which can be particularly affected in the case of flooding (Okada et al. 2015; Takahashi et al. 2017).

Another important parameter for water velocity and dissolved oxygen is the slope. This parameter also appears to be very important for the distribution of salamanders. The higher the mean slope of the surrounding environment, the higher the probability of the presence of salamanders becomes. However, there is an upper slope limit beyond 17%, where suitability decreases. Steeper slopes result in higher water velocities and, in mountainous areas, streams tend to be smaller as the slope increases, which may not be ideal for giant salamanders. However, the mean slope of the terrain does not necessarily reflect the slope of the rivers. Our results are comparable to those on habitat preferences of the Chinese giant salamander (Andrias davidianus), where slope, temperature and precipitations strongly influence habitat suitability for this species (Zhang et al. 2020). The land-use variables tested here (i.e. agricultural areas and urban areas) had only a small contribution to the distribution of the species. These results corroborate the known local data on the habitat preferences of Japanese giant salamanders (Okada et al. 2008). This contrasts with the situation of the Chinese giant salamanders, which are harvested for food and, therefore, less likely to persist near urban areas (Turvey et al. 2018). Nests and larvae of Japanese giant salamanders are often found in small headwaters (Bjordahl et al. 2020). These streams are often surrounded by broadleaf forests, tree plantations or agricultural areas. Our models suggest that an increase of urbanisation may increase the probability of finding the Japanese giant salamanders. This is, in fact, caused by the urbanisation and associated agriculture which takes place more along the main river network than outside, which means that it often follows the habitat of the salamanders. The long-term persistence of these populations in such areas may, therefore, be uncertain as also shown by the low abundance of the species or the absence of breeding sites (pers. obs.). This is particularly reinforced by the fact that giant salamanders have a long life expectancy (Klein et al. 2024) and may then subsist, but not breed in some habitats for many years following disturbance. Therefore, the long-term viability of salamander populations in urban areas is precarious, due to the artificialisation of streams and rivers (e.g. concrete banks and dams), decreasing catchments and breeding sites availability, as well as impeding connectivity (Tochimoto et al. 2007; Okada et al. 2008; Taguchi and Natuhara 2009; Bjordahl et al. 2020; pers. obs.).

The projection of the current potential distribution (Fig. 3) highlights that, within the current distribution range of the species, many areas present a moderate (red colour) and high suitability (green colour). However, several regions of the distribution of the species provided by the IUCN (IUCN SSC Amphibian Specialist Group 2022) include unsuitable habitats projected by our model, especially throughout Shikoku Island. Few detailed data about the distribution in Shikoku Island are available in literature and, therefore, our study provides a foundation for future mapping efforts, focusing on areas of medium to high suitability and under-sampled regions, as it has been done with the hellbender (Freake and DePerno 2017). More specifically, there is a particular need for more surveys on the large islands of Shikoku and Kyushu. Although palaeontological evidence of the species presence from the Ehime Prefecture in Shikoku was reported by Shikama and Hasegawa (1962), Ikoma (1963) mentioned the discovery of only two individuals on the island without providing a map. Kobara (1985) reported scattered occurrences across the island (included in unfavourable areas according to our models) and Matsui (2008) pinpointed a single locality in Shikoku Island, from Tokushima Prefecture (these occurrence references were not included in the occurrence data of the present study as the precise location were not available). Similarly, the presence of the species in Kyushu Island was only clearly mapped in the Oita Prefecture. In Honshu Island, the distribution is split into two main areas, with most locality data coming from the western part. Although some mountain barriers may be responsible for this separation, which also explains the genetic difference between these two groups of populations (Matsui et al. 2008), some of the areas appeared as suitable and would deserve future surveys. Moreover, within their distribution range, giant salamanders were found in some areas classed as moderately suitable according to our models suggesting either conservation issues (i.e. possible future declines) or the need for complementary samplings to fine tune the models. In this perspective, the use of eDNA surveys could give promising data to be followed up by visual surveys of positive sites (Bjordahl et al. 2020; Neto et al. 2020; Hidaka et al. 2024). In the northern part of Japan, numerous habitats have been identified as potentially highly suitable for Japanese giant salamanders. This finding aligns with our initial hypotheses, suggesting that the climatic differences between the northern and southern regions of Japan do not account for the current absence of the species in the north. Instead, the biogeography of the Japanese giant salamander and the natural barrier formed by the Japanese Alps in the centre of Honshu Island appears to be the primary factor preventing the species from establishing a wider distribution.

Our projections of the distribution of the Japanese giant salamander in the upcoming decades using SDMs suggest that suitable areas for the species will severely decrease both within and beyond the current distribution range of the species. The decrease of highly suitable areas will be more moderate beyond the current distribution range of the species (Fig. 4), especially if we consider the optimistic scenario. These results suggest that climate change may have direct negative effects on Japanese giant salamander populations. It is in accordance with previous studies of the Chinese giant salamander (Zhang et al. 2020) and the North American giant salamander (Sutton et al. 2023) as for these other species, the predictions made revealed that future distributions may be severely reduced under climate change. The maps of habitat suitability changes (Fig. 4) and future habitat suitability (Suppl. material 6: fig. S4) show that, within the current distribution range, the most suitable areas will be mainly restricted to inland regions of Honshu and Kyushu Islands in the future, while on the island of Shikoku, highly suitable areas seem to be absent by 2050. Future climatic projections suggest that the intensity of precipitations and temperature will increase (but with local variations) in all regions of Japan (Takabatake and Inatsu 2022). One main issue of the increase of heavy and sustained rainfalls is that it increases the frequency of flooding events (Shakti et al. 2020; Hatsuzuka et al. 2021), a noticeable disturbance for the species, suggesting a potential negative influence on Japanese giant salamander survival. On the other hand, the life habits of giant salamanders, especially their fully aquatic life and benthic mode of locomotion, may turn out to be a protection from climate extremes (Brooks and Kindsvater 2022). This lifestyle and habitat may have supported the survival of giant salamanders as a fully aquatic clade for millions of years. However, data are currently lacking to investigate how water temperature would be locally affected by increasing global temperatures and, therefore, long-term monitoring on rivers is needed to assess such changes more precisely. A study conducted on the North American giant salamander (Cryptobranchus alleganiensis) showed that a 2°C increase in temperature decreased the immune function and growth rate and increased physiological stress of the individuals (Terrell et al. 2013).

Our findings suggest that the northern part of the Honshu Island may be able to hold large areas of suitable habitats for the Japanese giant salamander in the present and the future (Figs 3, 4, Suppl. material 6: fig. S4). However, access to these potential areas is limited as the main barrier to salamanders’ expansion to the north is represented by the ‘Japanese Alps’, a mountain chain bisecting Honshu Island (Fig. 1) and, therefore, fully preventing the direct cross and dispersal of this species, which could not climb terrestrial environments outside of riverine areas. Ultimately, natural dispersal to the north is, therefore, unrealistic. Considering that habitat suitability within the current distribution range of giant salamanders will severely decrease in the future (Fig. 4), increased habitat fragmentation could exacerbate dispersal limitations in this area. This may increase the vulnerability of the species to climate change by preventing access to potentially suitable habitats (Caplat et al. 2016). This very specific condition allows us to keep in mind that distinction must be made between the potential distribution (all the suitable habitats) and the actual distribution (also called realised distribution) of a species (De Kort et al. 2020). In our study, we predicted the potential distribution of the Japanese giant salamander without accounting for dispersal constraints, specifically to illustrate how the ‘Japanese Alps’ restrict its distribution primarily to the south-western regions of Japan. Another major obstacle to dispersal is the low habitat connectivity caused by dams, which can reduce or block movements of this species along the inhabited watercourses (Tochimoto 1995; Taguchi and Natuhara 2009; Takahashi et al. 2016).

In addition to highlighting that the Japanese giant salamander populations would be threatened by future climate change, it is necessary to keep in mind the current endangered status of this species. The conservation status rank of the species was changed from Near Threatened to Vulnerable in the 2022 IUCN Red List (IUCN SSC Amphibian Specialist Group 2022), but is likely to be reassessed as many populations are, in fact, declining. Moreover, due to the long life expectancy of this species, the real status of the populations is difficult to assess. This species is facing many other threats and one of the most disturbing is habitat degradation and fragmentation (Takahashi et al. 2016). Habitat fragmentation is particularly due to dam construction for agriculture and flood control to protect human settlements (Taguchi and Natuhara 2009; Borzée et al. 2024). These facilities deeply affect the habitat connectivity within and between waterways, as well as the suitability of breeding microhabitats. Thus, we have identified potential future suitable habitats for Andrias japonicus within the current distribution range, but without connectivity between habitats, they would be inaccessible to the giant salamanders. The study of habitat connectivity is fundamental to biodiversity conservation (Correa Ayram et al. 2016), but such data are not yet available for modelling at the country scale. As a fully aquatic species, Japanese giant salamanders have limited migration abilities when facing barriers in the rivers and long-distance migration is unrealistic. These findings suggest that the highest conservation priority for Japanese giant salamanders is to characterise their ecology at a local scale, such as habitat preferences and implement the most effective conservation actions possible to preserve populations and improve habitat connectivity (Bjordahl et al. 2020). Moreover, the decreasing trend of the human population in Japan, especially in rural areas, could benefit Japanese giant salamander populations by reducing human pressure on giant salamander habitats. For example, an increasing number of small agricultural dams are no longer maintained and eventually burst (pers. obs.), restoring habitat connectivity. However, the construction of facilities to control floods and damages due to erosion is increasing in Japan (Itsukushima 2023), threatening many habitats of the Japanese giant salamander and it is directly linked to climate change.