Joint temporal trends in river thermal and hydrological conditions can threaten the downstream migration of the critically endangered European eel

Climate change is modifying the hydrological and thermal regimes of rivers worldwide, threatening the triggering of organisms’ key life-cycle processes. European eel (Anguilla anguilla) is a critically endangered fish species that migrates over several thousand kilometres between its rearing habitats in continental waters of Europe and North Africa and its spawning area in the Sargasso Sea. Downstream migration of adult eels occurs during periods of decreasing river water temperature associated with high discharge but changes in these environmental cues may affected eel migratory conditions. An innovative multivariate method was developed to analyse long-term datasets of daily water temperature, discharge and eel passage in two European rivers. Over the past 50 years, water temperature and discharge increased in both rivers during the downstream migration period from August to November. Silver eels preferentially migrated at temperatures between 10 and 20 °C combined with high discharge. Environmental changes have resulted in the migration of silver eels under warmer water temperatures. This example illustrates how the changes in environmental cues have led to a growing mismatch between the migratory conditions preferentially selected and those actually used, which may threaten the completion of the eel’s life cycle and ultimately the persistence of this already critically endangered species.


Material and methods
Study sites and data used. The Imsa watershed is located in southwestern Norway (58° 54' N, 5° 57' E) and covers an area of 128 km 2 (Fig. 1). The watershed is sparsely urbanised (1.5% of the total surface of the watershed) and dominated by natural areas (54.8%) and agriculture (43.8%) 26 . The hydrology of the Imsa River is mostly influenced by rainfall, with a gradual decrease in discharge from January to June, followed by a period of low flow in summer and an increase in fall (see Supplementary material, Fig. S1). A Wolf-type fish trap is located about 100 m upstream of the river mouth and has been operating throughout the year since 1975 to catch fish during their downstream migration. The distance from the trap to the upper limit of eel's suitable habitat in the Imsa watershed is approximately 20 km 18  The Burrishoole watershed is located in western Ireland (53° 56' N, 9° 35' W) and covers an area of 90 km 2 (Fig. 1). The watershed is dominated by natural areas (i.e., forests, peat soils and wetlands and lakes), sparsely populated and the main land-use is coniferous forestry and subsistence agriculture 27 . The hydrology of the Burrishoole River is spatey and mainly influenced by rainfall. Two upstream and downstream Wolf-type fish traps are located on two outflow rivers from Lake Feeagh to the brackish Lake Furnace and have been operating year-round since 1971 28,29 . The distance from the traps to the upper limit of eel's suitable habitat in the Burrishoole watershed is approximately 13 km 18 . In this study, water level in Lake Feeagh was used as a proxy for water discharge 18 . Due to major modifications in the monitoring of the lake water level in 2013, subsequent data were discarded to avoid potential bias. Water temperature and water level have been recorded daily since 1970. The data considered in the present study covered the period 1971-2013. Mean annual water temperature in the Burrishoole River varied between 9.2 and 12.4 °C (interannual average = 10.5 °C) and mean annual water level varied between 0.33 and 0.49 m (interannual average = 0.41 m). Annual and seasonal means of water temperature and discharge/water level are provided in Table S1 for both rivers.
In both rivers, the traps are checked once or twice every day throughout the year. Catches were summed up to get the number of migrating silver eels per 24 h. Biological years were defined according to the migration period of silver eels 16 (Fig. 2B), from the 1st of June of year N to the 31st of May of year N + 1. For the Imsa River, between 706 (in 1993) and 5,355 (in 1977) silver eels have been trapped yearly (interannual average = 2687 eels; Fig. 2A). For the Burrishoole River, between 1559 (in 1983) and 6,549 (in 1975) silver eels have been trapped yearly (interannual average = 3379 eels). Mann-Kendall trend tests were performed on annual number of silver eels caught to test for temporal trends over the study period.
Trends in annual water temperature and discharge associations. Water temperature and discharge data were analysed using the Choc method 8 to determine which associations of water temperature and discharge have become significantly more or less frequent over the years. First, yearly densities of probability were estimated with kernels at each pair of water temperature and discharge conditions. The bandwidth parameter (H), which defines the degree of smoothing of the kernel curves, was selected according to the plug-in method 30 . Second, Mann-Kendall trend tests were used on the densities of probability over the entire study period for each pair of water temperature and discharge conditions to detect monotonic temporal trend. Statistical significance was assessed by comparing results with those of artificial datasets generated under the null hypothesis, which implied that the frequency of a given water temperature and discharge association did not change over time. This approach was applied separately to Imsa and Burrishoole rivers for the full range of biological years available to reveal annual environmental trends. Discharge and water level were log-transformed prior to analyses for normalization purposes and to limit data over-dispersion.
Trends in water temperature and discharge associations available during the downstream migration period: the available niche. The Choc method was applied to daily water temperature and discharge/water level data covering the migration season (from August to November; Fig. 2B), which represented the key time period when silver eels experience possible environmental changes, to detect trends in the available niche, i.e., the occurrence of water temperature × discharge associations during the period of silver eel migration. Kernels estimated during the migration period were denoted Kde env .
Trends in water temperature and discharge associations effectively used by silver eels during downstream migration: the effective niche. The daily proportion of migrating silver eels was calculated per biological year as the daily count divided by the total counts of the biological year. To detect whether changes in the available niche have led to changes in the environmental conditions effectively used for www.nature.com/scientificreports/ downstream migration, i.e., the 'effective' niche, we applied another Choc analysis on water temperature and discharge data restricted to the migration season (similarly to the available niche) with the exception that days were weighted by their daily proportion of migrating silver eels. Kernel estimates weighted by the daily proportions were denoted Kde eel .
Preferential niche of silver eel migration and its change over time: the preferential niche. To characterize the preferential niche for silver eel migration (i.e., water temperature and discharge associations preferentially selected by eels to migrate), we built an electivity index based on the Ivlev feeding electivity index 31,32 , which was commonly used in gut content analyses to reveal prey preferences of predators. This index identified prey "selected" or "rejected" by a predator, following the standard terminology generally associated with diet analyses. Consequently, we adapted it to refer to the environmental conditions "selected" or "rejected" by silver eels for their downstream migration by taking into account, for each association, the frequency of a water temperature × discharge association in the environment, the occurrence of a migration event and the number of eels migrating under this association. The electivity of a water temperature × discharge association E'(T,Q) was calculated as follows: With Kde eel (T,Q) and Kde env (T,Q) the estimated kernel densities of probability of occurrence of water temperature T and discharge Q association estimated on the basis of the effective and available niches, respectively. Electivity index values close to -1 corresponded to associations "rejected" by silver eels whereas values close to 1 indicated preferentially "selected" associations. As index values were highly fluctuating and not reliable for low kernel estimations, the electivity index was not calculated for the 5% rarest environmental associations. These conditions corresponded mostly to periods of low discharge that are not suitable for silver eel migration.
To get an overall picture of the preferential niche over the whole study period, electivity indices were first computed using kernel density estimations fitted on all years pooled together (i.e., average preferential niche). Then, to investigate temporal trends in this niche, yearly electivity indices were computed for each water temperature × discharge association using yearly Kde eel and Kde env . Then, similar to a Choc analysis, Mann-Kendall trend tests were performed on the electivity indices of each pair of associations. Statistical significance was

Results
The annual number of silver eels migrating each year decreased significantly on the Imsa River between 1975 and 2019 and on the Burrishoole River between 1971 and 2013 ( Fig. 2A).
In the Imsa River, high discharge conditions have increased in recent years in comparison to the 1970s as well as days with low discharge (log-scale values of − 2.5-0; natural scale values of 0.1-1 m 3 s -1 ) and cold temperature (< 10 °C) (Fig. 4). On the contrary, low and intermediate water temperatures (< 15 °C) associated with . Schematic representation of the methodological framework and links between the performed analyses. Daily water temperature T and discharge Q are used to identify trends in the available niche over the years y. T and Q, weighted by the daily proportions of eel passage N, are then used to identify trends in the effective niche. The available and effective niches are finally used to calculate an average preferential niche over the study period and trends in the preferential niche over the years y.  (Fig. 4). These results apply to trend analyses at both annual and seasonal scales (but with respect to water temperature and discharge seasonal ranges; Fig. S2). Percentages of significant changes in the occurrence of environmental associations are provided in Table S2.
Concerning the migration season, occurrences of situations where water temperature was above 10 °C associated with high discharge have increased compared to the past (available and effective niches) on the Imsa River (Fig. 5). The average preferential niche corresponded to silver eels preferentially migrating at (1) water temperature between 10 and 15 °C, regardless of the discharge, and (2) water temperature between 15 and 20 °C with high discharge (Fig. 5). The preferential niche has remained essentially the same over time (see Fig. S3) but the selection of warm temperatures has increased compared to low temperatures (see trends in the preferential niche, Fig. 5).
In the Burrishoole River, water temperatures between 10 and 17 °C associated with high water level have increased during the migration season over the study period (i.e., the available niche; Fig. 5). Trends in the effective niche were different from trends in the available niche: eels migrated more frequently than in the past at warm water temperatures (> 17 °C) or at intermediate water temperatures associated with low discharges. Similar to the Imsa River, silver eels preferentially migrated at water temperatures between 10 and 20 °C at high water level. Trends in the preferential niche demonstrated increased migration at high water temperature (> 17 °C) over the study period.
The results on these two rivers were similar: the main association within the available niche that became more frequent was warm temperatures associated with high discharges. These trends in the available niche drove totally (for the Imsa River) or partially (for the Burrishoole River) the trends within the effective niche. Preferential niches were also similar among rivers, with a selection of water temperature between 10 and 20 °C associated with high discharges, but this preference gradually tended towards warmer water temperatures.

Discussion
In this article, we proposed an innovative method to link the occurrence of a species, or of a key life-cycle event, to various environmental cues simultaneously through the ecological niche concept 35 . Unlike traditional statistical methods, our method explicitly considers the correlation structure between the variables considered. This joint treatment of the variables is essential since the interaction between them may entail a differential impact on the ecological process of interest. For instance, a high discharge can trigger eel migration but the ecological consequences of this migration (such as energy costs) may not be the same depending on water temperature. We showed that in two watersheds located in the northern part of the geographic range of the European eel, high discharges were increasingly frequent, especially associated with warm water temperatures, both annually and during the core migration period (from August to November). We found that these trends in the frequency of . Two-dimensional heatmaps of water temperature (X-axis; °C) and log-transformed discharge/water level (Y-axis; log m 3 s −1 or log m) associations for the Imsa and Burrishoole rivers at the annual scale. Water temperature × discharge/water level associations that have become more or less frequent over the study period are shown in blue or red, respectively. Light and dark colours correspond to non-significant and significant trends, respectively (e.g., dark red corresponds to significant decreasing trends, light blue to non-significant increasing trends). The grey line delineates the convex hull (i.e., the smallest space encompassing all points of the annual dataset). See Fig. S2 for the results at the seasonal scale. www.nature.com/scientificreports/ environmental associations within the available niche led to a modification towards warmer water temperatures of the conditions used by silver eels to migrate. Concerning the proposed approach, the effective and preferential niches can diverge: while eels preferentially "select" specific conditions, migration can occur in other conditions (even conditions identified as usually "rejected") if the most suitable ones are not available. For example, we observed an increasing trend in the use of warm water temperatures in the effective niche on the Burrishoole River. However, this does not mean that these conditions are now the most frequently used by eels to migrate: these conditions were rare at the beginning of the study period and we detected an increase in the occurrence of such conditions compared to a near-absence at the beginning of the period. Additionally, we focus on the migration period (from August to November) because similar environmental conditions may occur at other times of the year but are not used by the eels because the migration peak has already occurred. Therefore, the eels' preference for a condition would be lowered if we considered the whole biological year. In our application of the Choc analysis, we assumed that eels display immediate responses to environmental conditions as we compared daily migrations to daily environmental conditions (though kernel smoothing reduces the immediacy). This is supported by the fact that the two watersheds studied are small and that changes in local conditions (e.g., precipitation, wind) have a very rapid effect on the thermal and hydrological conditions of the rivers. We therefore assumed that the initiation of their migration and capture by the trap occurred within one day.
The two 50-year datasets mobilized allowed us to robustly characterize the current effective and preferential niches for eel downstream migration. Using observations and measurements, we assessed past effects of global change on silver eel migration on the Imsa and Burrishoole rivers. A perspective of this work would be to further develop this method to allow the projection of possible effects of future hydroclimatic changes on the ecological niches for eel downstream migration, by combining it with climate and population dynamics models. Furthermore, to capture a more comprehensive picture of the effect of hydroclimatic trends on eel migration, it would be necessary to apply our approach to watersheds distributed throughout the species' geographic range. However, despite multiple requests, we did not find any other long-term datasets that combined daily measurements of water temperature, river discharge and eel passage. Another limitation of our work was that the number of silver eels never became limiting by weighting the days with the daily proportions of annual migrants. We considered this assumption to be reasonable since we restricted the analysis to the core migration season, and the migrants appeared to be evenly distributed over from August to November. Future developments of the method could make it possible to weight observations by the daily number of migrating eels divided by the sum of remaining eels, with the risk of giving too much importance of late migratory episodes. The last migrants may be rather "in a hurry" to migrate, whatever the environmental conditions, which could have a significant impact on the preferential niche.
Global climate change has affected the rainfall-driven hydrological regimes of the Imsa and Burrishoole watersheds over the past 50 years. We detected an increase in the frequency of high discharge events on both rivers. This finding is consistent with studies that predict a severe alteration of natural flow regime by 2050 36 with Burrishoole rivers . Two-dimensional heatmaps of water temperature (X-axis; °C) and logtransformed discharge/water level (Y-axis; log m 3 s −1 or log m) are plotted for the period August to November (i.e., the migration season) representing trends in the available niche, effective niche, average preferential niche and trends in the preferential niche. Discharge/water level and temperature associations that have become more or less frequent over the study period are shown in blue or red, respectively. Light or dark colours correspond to non-significant and significant trends, respectively. The grey line delineates the convex hull. Within the preferential niche, the 5% of the less frequent associations correspond to the white colour within the convex hull while the selected or rejected associations are shown in green or purple, respectively. www.nature.com/scientificreports/ a significant increase in annual precipitation 37 as well as in the annual number of rainy days 38 . The increase in precipitation level is expected to be considerable in winter (between + 15 and + 30%) while remaining moderate in summer (+ 5%) 36 . Extreme rainfall events are also expected to be more common and intense in small coastal watersheds 39 similar to the Imsa and Burrishoole watersheds. Concurrently, water temperature increased on both rivers, as previously reported on the Imsa River over the last 20 years 29 and on the Burrishoole River since 1960 40,41 . While global change is expected to have spatially heterogeneous effects on the hydrological regime of rivers (i.e., increases in summer and fall discharges in northern Europe; decreases in southern and southwestern Europe 36 ), the patterns observed over the last few decades were similar for the two rivers considered in this study. The almost 50-year time-series from the Imsa and Burrishoole rivers provide a unique environmental baseline for assessing how global change has affected the environmental cues that regulate many ecological processes. Despite a decrease in the number of migrating silver eels 29 , there were enough individuals caught every year to study the environmental conditions associated with eel downstream migration. Given that the European eel is a panmictic species (i.e., individuals originating from a single spawning area 42 ) and environmental conditions are comparable within both rivers, the associations selected were unsurprisingly similar between the Imsa and Burrishoole rivers, with silver eels preferentially migrating at water temperature above 10 °C associated with high discharge. These preferentially selected conditions are consistent with existing observations 18,24 but the possibility to migrate within a wide range, between 4 and 23 °C, has been observed 43 . Given that water temperature and discharge appear to be the predominant environmental cues triggering eel migration 15 , some other abiotic (e.g., photoperiod, barometric pressure, lunar phase or turbidity) or biotic factors (e.g., shoal effect 44 ) interact to determine the intensity of migration 18 and may explain individuals migrating under "rejected" conditions 45 . This wide range of water temperature can also be explained because the migration phenomenon is not only explained by the immediate environmental conditions, but also by the conditions experienced by eels in the months prior to migration 18 .
Water temperature and discharge conditions observed during the downstream migration period of eels (i.e., August-November) have changed significantly on both rivers since the 1970s. Significant trends (either increases or decreases in frequency of occurrence) were detected in the effective niche for about 25% and 35% of all associations of water temperature and discharge conditions for the Imsa and Burrishoole rivers, respectively (see Table S2 for percentages of significant trends in each niche). We have shown an increased preference towards water temperatures above 15 °C and niche margins (i.e., potentially suboptimal environmental conditions) but such high temperatures remain scarce and most eels migrated at lower temperatures. A shift towards warmer temperature used is not surprising: both rivers are in the northern part of the species geographic range, where river temperatures are below the temperature optimum for eel growth 41,46 . When comparing silver European eels caught at different latitudes, growth rate tends to increase in the southern regions because of warmer temperatures 41,47 , although it has since been shown that this relationship appears to be more complex 48 . Increasing water temperature due to climate change is expected to increase eel growth in northern regions of Europe 46 , which includes the Imsa and Burrishoole rivers, and expand the geographic range of the European eel northward 49 . The hydrological regime may become the main restricting factor to eel migration, with summer discharges too low to trigger migration or requiring more energy resources for migrants. Along with this, the number of eels migrating in August in the Burrishoole River is positively related to the water level 29 . High summer discharge events have recently been more frequent and discharge variations may have stimulated a progressive early trigger of the migration on the Burrishoole River over the study period as observed by Sandlund et al. 18 , who have also proposed that this shift might be induced by a change in spring conditions leading to earlier silvering. In contrast, in the Mediterranean region, warm water temperatures combined with low discharges prevented silver eels from migrating during summer/fall and delayed migration until a water temperature drop in December/ January 50 , several months later than in northern latitudes 16 . The contrasting trends in environmental conditions induced by global climate change is a real challenge for a panmictic species whose evolutionary capacities for local adaptation are limited 51 .

Conclusion
The Imsa and Burrishoole rivers are located at different latitudes but have similar characteristics in terms of European eel subpopulations size and migration dynamics. We have developed a method to quantify the joint temporal trends in water temperature and discharge associations and their consequences on the available, effective and preferential niches of the downstream migration of European eel. Since the 1970s, silver eels have increasingly migrated during conditions of warm water temperatures (above 15 °C) associated with high discharge, a scenario which has become more frequent in the Imsa and Burrishoole rivers. So far, these conditions do not seem to have been limiting for the resilient European eel, but if the warming trend persists as expected in the coming decades, it will be very difficult to anticipate the practical consequences for the population and a mismatch between silvering "readiness" and the appropriate environmental window for migration could be feared. Besides, one could expect the environmental trends within the species geographic range to be even more heterogeneous than between the two watersheds studied, raising the issue of asynchronous migration between the northern and southern regions. Although we cannot definitively conclude on the impact of such environmental changes on the downstream migration of eels, we have broadly scanned the possible consequences, which may hold drastic implications for the persistence of an already critically endangered species. Importantly, we have developed a method that is particularly well suited to the analysis of long-term monitoring data and could be applied to non-panmictic species with a high homing rate (e.g., the Atlantic salmon Salmo salar), which may exhibit local adaptation and greater adaptability.