To assess upstream and downstream movements in the artificial creek, we caught 100 larvae (under license of the ‘Struktur- und Genehmigungsdirektion (SGD) Nord’ of the Federal State of Rhineland-Palatinate, Germany) between mid-May and mid-June 2015 in a first-order creek (Beresbach, Federal State of Rhineland Palatinate, Germany) depending on their availability. Each individual was marked with fluorescent Visible Implant Alphanumeric (VIA) tags (size 1.2 mm x 2.7 mm) of the Northwest Marine Technology Inc. (see Wagner, et al. 22). The tags were injected at the base of the tail on the right side of the body. This treatment was approved by the Ethics Commission of Trier University and carried out in accordance with the relevant animal welfare guidelines and regulations of the Federal State of Rhineland-Palatinate. All methods are reported in accordance with the ARRIVE guidelines. Two different colours (neon-yellow and neon-orange) were used to facilitate differentiation of two experimental cohorts (see below). After tagging, the larvae were kept in a climatic chamber at 15°C for one week before being put into the artificial creek in order to ensure tag retention.
The artificial creek originated in a small initial pool from where the water flowed down in three bends over a total length of 32.5 m into a terminal pond. From here the water was pumped back into the initial pool, forming a closed circuit. The creek bed was made of black pool liner and filled with gravel and chalk-sandstone rocks. In the second bend there was a permanent pool. Between the second and the third bend there was a small waterfall of about 30 cm height. Since European fire salamander larvae are mainly found in places with low water velocity 17, 16 additional potholes were formed with gravel and stones to create small areas of reduced flow velocity. Slates of different sizes were added to provide hiding places. The slates also facilitated the search for larvae in hiding places. The widest point of 2.1 m was in the second bend at the pothole. Apart from that, the widest place of 70 cm was between the first and the second bend and the smallest place of 30 cm was directly behind the initial pool. The difference in height from the initial pool to the pond at the end of the creek was about 5–6 m. The inclination measured at randomly selected points within the course of the creek ranged from 0 to 58 %. On average, the inclination was 18 %; the upper part of the creek has a comparatively lower inclination than the lower part.
Two different types of traps were installed at different places within the creek (pairwise if possible) to record the movements of the larvae. Each pair consisted of a downstream and an upstream trap, which together completely blocked the creek bed (for details see Veith, et al. 19). In narrow places only the downstream trap fitted into the creek bed. This was the case at the beginning of the creek directly behind the initial pool and at the end of the creek directly in front of the pond. Four pairs of traps were installed in between. Water temperature was measured every hour by three data loggers (Tinytag loggers of Gemini Data Loggers UK Ltd., Chichester, UK), which were put beneath stones within the stream to not being exposed to direct radiation.
After weighing (to the nearest 0.001 g using a Kern EMB 200-3 field balance) and photographing on graph paper (to later determine the snout-vent-length, SVL, as the distance between the tip of the snout and the hind legs as well as the head width to the nearest 0.5 mm using the software Datinf®23), the larvae were randomly introduced at one of the 18 introduction sites (initial pool, permanent pothole and additional potholes modelled with slate). Of the 100 initially tagged larvae, 82 were finally introduced into the experiment (twelve lost their tag before introduction, and six developed signs of a beginning metamorphosis after being tagged). Depending on availability from the source creek, the first 18 larvae were introduced into the experiment on May 11, followed by two larvae on May 26, nine larvae on June 2, one on June 5, nine on June 9 (up to this date all larvae had yellow tags), 17 on June 16, ten on June 18 and 16 on June 19 (all these larvae had orange tags). In order to ensure sufficient feeding of the larvae, the artificial creek was stocked ten times with macroinvertebrates before the first larvae were introduced. The stocking was repeated once a week throughout the trial period. During the period of flood simulation experiments, the invertebrate stocking was done one day after each flood simulation experiment.
The water flow of the artificial creek was kept constant at a rate of 48 l/min. Towards the end of the experiment, four flood simulations were initiated (June 30, July 4, July 8 and July 12). To simulate flood peaks, the pumping rate could be increased to 192 l/min; additional water could be pumped into the system from a 1000 liter water tank at the same maximum flow rate of 192 l/min. Two different types of flooding were simulated to test different stress situations caused by increased water flow. We first simulated a continuous but gentle increase in flow rate within the first 10 minutes, followed by a constant flood peak of 384 l/min for 70 minutes. After that, the circulating water flow was kept constant at 192 l/min until the next morning and then dropped to the regular 48 l/min. This was to simulate an increased water flow, e.g. after a thunderstorm. The second type of flood simulated an irregular and potentially more stressful flow of water; it started directly with the maximum flow rate of the first pump (192 l/min) and simulated another sudden increase in discharge (in less than a minute) towards 384 l/min, with two short interruptions of lower discharges (192 l/min) for five minutes each. Again, the circuit water flow was then kept constant at 192 l/min until the next morning, when it dropped to the regular 48 l/min.
The entire experiment took place from May 11 to July 15. The traps were inspected for larvae twice a week. During the flood simulations traps were checked every other day. The larvae were counted as either moving downstream (Nd) or moving upstream (Nu). VIA tags were read with the help of UV light, weight and SVL were measured at each inspection event as described above. Larvae with yellow tags were returned randomly to one of the 18 introduction sites (‘random treatment’); this ensured that data could be collected across the entire artificial creek. Larvae with orange tags were released at the closest introduction site in the direction in which they had moved (‘directional treatment’): below a downstream trap or above an upstream trap. This directional treatment was designed to mimic movement histories that were not interrupted by our treatment. To gather information from a non-captured control group, we searched for larvae in the introducing sites between the traps after every third trap control event (Ns). They were identified and measured as described above to obtain representative samples of body size and nutritional status and subsequently returned to their pothole of capture. To ensure that our search for larvae at the introduction sites had not unintentionally induced additional up and down movements, the traps were then checked again for larvae. Larvae that entered metamorphosis or had lost their tag were removed from the experiment. The nutritional status of all larvae was calculated as Scaled Mass Index (SMI) 24.
Prior to estimating population size, we tested if the two reintroduction treatments of trapped larvae might have caused a heterogeneity in their capture probability. We used the subroutine ‘closed captures’ in MARK 25 to compare the probabilities of the models of 26 including group-specifity: g*M0 (equal capture probabilities), g*Mt (time-dependent capture probabilities), g*Mb (behavioural effects on capture probabilities), g*Mh (group-specific heterogeneity in capture probabilities) and combinations thereof. We used only capture histories of trapped larvae, since non-trapped larvae were not exposed to different reintroduction treatments. We expected a model including group specific capture probabilities (g*Mh, g*Mbh, g*Mth or g*Mbth) to be the best model if the two reintroduction groups would differ in their capture probabilities. The Otis, et al. 26 models were compared with or without group specific capture probability. Model selection was based on AICc values.
We used the software MARK also to estimate the population size for each control event. We calculated the survival probability (Φ), the capture probability (p) and the entering probability (pent) using the subroutine POPAN 27, which provides a Cormack-Joly-Seber (CJS) estimate for open populations 28–30. We selected the best fitting population model using the AICc criterion. To see how well the estimated population sizes reflect reality, we determined the minimum and maximum possible population sizes for each control event. The minimum value for a given control day was obtained by adding the number of individuals actually caught on that day and the number of individuals not caught at this day but on later days. The maximum value was calculated by subtracting dead or removed individuals from the number of larvae introduced into the system prior to this capture day. Based on the estimated population size, we calculated drift rates for each control day.
The data were tested for normal distributions using the Shapiro-Wilk test 31. Depending on the result of the Shapiro-Wilk test, drift rates were compared using a paired t-test for hypothesis I, the unpaired t-test for Hypothesis II, III and IV and the Wilcoxon-Mann-Whitney test for non-normally distributed data for hypotheses V. To test hypotheses V, drift rates were correlated with water temperature (average per interval between two trap control events across all three loggers) using the Spearman’s rank correlation; for this test, we had to confine to the data prior to the start of water flow manipulations in order to avoid that the then drastically increased water flow would mask potential temperature effects.
To account for co-variation of parameters, we performed a generalized linear mixed model (GLMM). We included SMI, head width, water temperature and treatment group as independent variables and analyzed them by stepwise AIC to identify the main factors that influence larval drift as binomial response variable. In addition, we used ID of larvae and date of measurement as random effects to account for variation in individual behavior and for multiple drift events with the same temperature due to same sampling interval. GLMM was performed using the glmer function of the lme4 package in R 32,33. Statistical tests were also performed using the software packages R 33.