River ecosystem resilience to extreme flood events

Abstract Floods have a major influence in structuring river ecosystems. Considering projected increases in high‐magnitude rainfall events with climate change, major flooding events are expected to increase in many regions of the world. However, there is uncertainty about the effect of different flooding regimes and the importance of flood timing in structuring riverine habitats and their associated biotic communities. In addition, our understanding of community response is hindered by a lack of long‐term datasets to evaluate river ecosystem resilience to flooding. Here we show that in a river ecosystem studied for 30 years, a major winter flood reset the invertebrate community to a community similar to one that existed 15 years earlier. The community had not recovered to the preflood state when recurrent summer flooding 9 years later reset the ecosystem back to an even earlier community. Total macroinvertebrate density was reduced in the winter flood by an order of magnitude more than the summer flood. Meiofaunal invertebrates were more resilient to the flooding than macroinvertebrates, possibly due to their smaller body size facilitating greater access to in‐stream refugia. Pacific pink salmon escapement was markedly affected by the winter flood when eggs were developing in redds, compared to summer flooding, which occurred before the majority of eggs were laid. Our findings inform a proposed conceptual model of three possible responses to flooding by the invertebrate community in terms of switching to different states and effects on resilience to future flooding events. In a changing climate, understanding these responses is important for river managers to mitigate the biological impacts of extreme flooding effects.

flows and associated habitats in which biological communities exist (Ledger & Milner, 2015), our overall understanding remains in its infancy (Coumou & Rahmstorf, 2012). Another key aspect of floods in addition to peak flow magnitude is their timing throughout the year, causing potentially different impacts, particularly with respect to biotic communities (George, Baldigo, Smith, & Robinson, 2015).
Of particular significance is the need to understand the effects of contrasting flooding events on community resilience and assembly (George et al., 2015;Pearsons, Li, & Lamberti, 1992). We define resilience as incorporating two elements (a) resistance of the taxa to the initial disturbance and/or (b) ability of the taxa to recover rapidly (Holling, 1973). A key question is how communities reassemble following flooding events and whether this makes the community more resilient or less resilient to further change following a major event. In addition, a full understanding of the effects of extreme flooding events across a range of organismal groups has previously been hindered by the lack of long-term predisturbance data to permit detailed insights into the interaction of community dynamics, successional processes, and river channel geomorphology (Poff et al., 1997).
In southeast (SE) Alaska, the summer of 2014 saw recordbreaking prolonged high rainfall creating a series of large, recurrent, and atypical flood events during the summer/early autumn.
At Bartlett Cove (SE Glacier Bay) June (133 mm) and July (211 mm) were the second wettest on record with July 12 the wettest July day on record (51 mm). Heavy precipitation continued into August with 222 mm of precipitation falling (fifth wettest summer month on record) (Menne et al., 2012). These events created an extreme high-frequency series of recurrent discharge peaks (Figure 1 Lemon Creek proximal to the study area). Significantly, these events followed an extreme winter flood in the same systems in November 2005 (Milner et al., 2013), with record rainfall (>650 mm in <72 hr) and widespread flooding across SE Alaska (>1 in 100 year flood).
Contrasting the effects of these events provides a unique opportunity to understand how the timing and recurrence of extreme climate events will alter river ecosystems and their subsequent recovery.
The main aim of this study was to examine the effects of two contrasting extreme flood events one in the winter and one in the summer on riverine habitat and the associated biological communities in the context of a long-term dataset. Specific objectives were to (a) examine whether the timing of the extreme events resulted in different biological effects, (b) assess how far each event reset the respective invertebrate communities, (c) determine the effect of the floods on the resilience of the different components of the biological community, and (d) develop a conceptual model of community response to extreme flooding events.

| Study area
In 1986, a continuous study was initiated of the ecosystem of Wolf  and fixed in place. Topographic height change from each floodplain bank was determined using a Sokkia dumpy level (Topcon, Tokyo, Japan), tripod, and staff. Floodplain height on the left bank was used as a control marker to account for differences in dumpy level setup, which allowed the cross sections to be comparable between years.

| Salmon and invertebrates
Adult pink salmon spawners were estimated using the average of counts by two observers walking the length of the stream, and juvenile coho salmon densities were estimated with minnow traps baited with salmon eggs and fished for 2 hr. From 1986, macroinvertebrates (animals > 1 mm) were collected annually in August or early September randomly from a representative sampling station Samples were collected randomly from the same sampling station with a Surber sampler (five replicates; 63μm mesh net). All invertebrates were preserved in 70% ethanol and later separated in the laboratory from detritus and inorganic matter. Macroinvertebrates were identified using Merritt and Cummins (1988), and Chironomidae larvae were identified using methods outlined in Milner et al. (2000).

| Statistical analyses
All statistical tests were completed using Minitab v15 or R v3.3.2 except Nonmetric Multidimensional Scaling (NMDS) which was undertaken using PRIMER v6 with each year included in the ordinations. Analyses were run with macroinvertebrate and meiofauna log 10 (abundance + 1) data. Both analyses were conducted using Bray-Curtis dissimilarity matrices and 2000 restarts. Persistence was determined using the index of Jaccard (1912) and year pairs for both macroinvertebrates (23 pairs) and meiofauna (17 pairs).
Nonparametric multivariate analysis of variance (PERMANOVA) tested the null hypothesis that differences in stream macroinvertebrate community composition between year groups before and after the flood (i.e., 1996-2005 vs. 2006-2008 vs. 2010-2013) were not different to those within year groups. Analyses were run using Bray-Curtis (BC) dissimilarity scores, with 10,000 permutations. Generalized least squares (GLS) regression of the two key chironomid species was applied to the time series of log 10 transformed Diamesa davisii and Pagastia partica abundance after initial analysis revealed significant autocorrelation. Models took the form P. partica ~ D. davisii + e, where e = an error term modeled as a first-order autoregressive process from the lag1 autocorrelation coefficient.   (Table 1).

| Invertebrate community structure
Nonmetric  F I G U R E 3 (a) NMDS plots for macroinvertebrates from 1986 to 2015 using mean abundance data from ten replicates collected in August/early September and (b) meiofauna from 1994 to 2015 using mean abundance data from five replicates collected during the same time period. The immediate postflood years are outlined in red and black and the arrows highlight significant shifts in the macroinvertebrate community structure between the groups of years. Numbers represent years  (Paterson, 1993;Robertson, Lancaster, & Hildrew, 1995) and the greater densities postflooding may reflect increased connectivity between the upstream lake and the stream channel.

| Invertebrate persistence
The largest decrease in macroinvertebrate persistence followed the recurrent 2014 flooding with only 24% similarity compared to 54% similarity following the 2005 flood reset (see Appendix Table 1 Paratrichocladius was the only taxon that has been found over the entire 30-year record and has persisted through all the major flood events. Persistence of the meiofaunal community decreased to a similar extent following both floods (46% similarity compared to 75% and 66% in preceding year pairs) (Appendix Table 1). However, the meiofaunal community was resilient following both floods in that total abundance of the meiofaunal community showed negligible change. TA B L E 2 Taxa eliminated and colonizing following the two flooding events and the number of years newly colonizing taxa persisted in the community following these events is given in brackets   (3) Ostracoda (2) Planorbidae

| D ISCUSS I ON
The magnitude of extreme flooding events is predicted to increase in the future with climate change, while the predictability of peak flow timings is likely to decrease (Ledger & Milner, 2015). Our study dem-  Year 1986 1987 1988 1989 1990 1991 1992 1993 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999  stages is low to nonexistent as few adults emerge at this time of year due to the stream being potentially ice covered in this part of Alaska.
Thus, the potential for rapid recolonization (and thus overall community resilience) may be reduced when extreme events occur in winter. can drive catastrophic shifts in ecosystems (Scheffer, Carpenter, Foley, Folke, & Walker, 2001), and findings from this study suggest alternative state theory can be adopted to propose a conceptual framework ( Figure 6) of how two differing extreme flooding events can influence a riverine invertebrate community during successional change and may cause a shift to an alternative state where the community is more or less resilient to future extreme flooding events.

Persistent shifts in overall macroinvertebrate
Our proposed framework should then be tested more widely with data from multiple areas and flood events.
Over time following glacial recession communities show a central successional tendency to shift toward state A. Consider then two extreme flooding events, potentially occurring at different times of the year, either in the same year or in different years. The first extreme flooding event (T 1 ) , can have two outcomes (1) a shift from state A to state B (evidenced by solid lines) or (2) the community is resilient to change either through initial resistance or rapid recovery (as evidenced by dotted lines) and remains at state A. A shift to state B in outcome 1 can then have three potential effects on the resilience of the community to a further extreme flooding event at T 2 -in scenario (a), the shift makes the community less resilient and will more easily reset further to state C which is less complex and similar to an early successional stage (i.e., increased likelihood of following solid line trajectory; Figure 6a), or in scenario (b) the shift makes the community more resilient to further change and thus more difficult for an extreme event to shift the community from B to C (i.e., increased likelihood of following dotted line trajectory; Figure 6b), or in scenario (c) the shift to state B does not affect the resilience of the community to future flooding events and is similar in resilience to state A (Figure 6c).
The traits possessed by taxa comprising the community will inevitably influence community response to extreme flooding. Certain traits may confer initial resilience to extreme flooding (i.e., the community remains at state A; outcome 2) or following flooding and outcome 1 such traits may be more common in the alternative state community so that after further flooding, the community is more resilient to disturbance and less likely to shift to state C (i.e., outcome 1b). For example, the trait of small body size may confer or improve resilience in hydraulically turbulent conditions (Segura, Siqueira, & Fonseca-Gessner, 2014;Snook & Milner, 2002) because smaller body size facilitates access to refugia during high flows following storm events (Mesa, 2012). The persistence of a number of the smaller chironomid species following the extreme events documented here supports this idea, and small F I G U R E 6 Conceptual changes in river invertebrate communities following two extreme flooding events occurring at different times T 1 and T 2 . Event T 1 may shift a community from state A to state B which is less complex and at an earlier successional stage, and therefore (a) less resilient, (b) more resilient, or (c) unchanged in resilience to a further extreme event at T 2 (see text) body size could also account for the observed resilience in the meiofaunal community following the two flooding events (Robinson, 2012).

Complexity
In addition, some species have developed traits related to life history evolution to avoid predictable large magnitude flooding which include fast development to aerial adult stage and egg-laying synchronized with seasonal lower flows (Lytle & Poff, 2004;Southwood, 1977).
Where the community becomes more resilient to future change following a shift to a different state (Figure 6b), a further shift from state B to state C may require an extreme event (T 2 ) that is higher intensity (size or duration) and/or has differing timing and predictability than the first event. The adaptations outlined earlier that confer resilience may only be invoked when the event happens "predictably" at specific times of the year relative to an organism's life cycle. Timing of the flooding event may thus be critical; for example, during the summer, many aquatic insects are in their aerial stage and can lay their eggs rapidly following a flood, whereas in winter, the potential for aerial colonization is limited as the majority of insects are in the egg or larval stage and are therefore potentially lost during the flood. The order of magnitude difference in reduction of total macroinvertebrate abundance following the November flooding event compared to the summer flooding events could thus relate to life history. The effect of a summer extreme flooding event may also be reduced for taxa that undergo continuous reproduction, as is the case for many meiofauna, because higher temperature results in rapid population increases and therefore greater resilience.
In addition, long-term changes in the habitat template (Gothe et al., 2017) as well as dispersal limitations (Brown et al., 2018) can restrict recovery of ecosystems following disturbance. In the case of WPC, the physical habitat template has not regained the slower flowing habitats evident before the 2005 flood and thus recolonization by Dysticidae, Gammaridae, Planorbidae, and Corixidae had not occurred by 2015. While there may be dispersal limitation effects impeding the recolonization of these groups, populations are present in nearby ponds and therefore the lack of geomorphological recovery would appear to be a major reason for their continued absence in the river. Conversely, increased sedimentation after the 2014 flood in some areas of the river has enhanced the abundance of Oligochaetae worms and may have impacted some of the chironomid species following the summer flooding causing overall lower community persistence. The lack of resilience of some groups and the shift in community states is undoubtedly linked to the (potentially long term) changes in the habitat template caused by the flood events. The findings clearly demonstrate fugitive taxa, like D. davisii grp, depend upon major disturbances to maintain populations in rivers illustrating the role of extreme flood events in enhancing river biodiversity. This taxon can also provide an indication of potential past extreme flood events when they appear in the record. Other groups were surprisingly resilient; juvenile coho salmon recovered rapidly following the 2005 flood despite the continued lack of geomorphological complexity in the stream with respect to their preferred pool habitat. This finding bolsters our previous suggestion that markedly different responses according to the organismal group mean that caution is required when applying ecosystem theories and concepts to predict responses to flood events at the whole river ecosystem scale (Milner et al., 2013).
Our findings have significant implications for the management of rivers to conserve biodiversity in light of increased incidences of extreme flood events. The implementation of mitigation strategies such as the preservation of remnant population refuges and proximal colonizing courses may be more critical according to the timing of the event, especially if floods occur during the winter or early spring rather than the summer. Continued disturbances may necessitate restoration of geomorphic complexity (i.e., the physical habitat template), because the natural recovery of complexity is very slow. This approach could facilitate the recolonization of taxa that have been eliminated and enable the reversal of the documented shifts in community states following these extreme events. More realistically, managers must accept dynamic change as a natural component of river ecosystems (Mainstone, 2017) and rivers should be allowed to flood and rework their morphology and biodiversity, and allow the persistence of fugitive species. However, where extreme events occur during system recovery (e.g., river restoration) from previous disturbances that have caused reduced biodiversity, then restoration practices that increase the resilience of the system to these events may need to be implemented to prevent prolonged extensions to recolonization periods (Reich & Lake, 2015).

ACK N OWLED G M ENTS
We thank those who have carried out research at Wolf Point Creek for long field seasons over the 30 years to develop the long-term record, including Sarah Conn, Elizabeth Flory, Ed Malone, Kieran Monaghan, Ian Phillips, Amanda Veal, and Mike McDermott.
We are grateful to Captain Jim Luthy, of the mv Nunatak, who for many years supplied our remote field camp in Goose Cove and latterly to Captains Justin Smith and Todd Bruno of the mv Capelin who facilitated access to the stream. We are grateful to numerous others in the National Park Service, not mentioned above, for other logistic and field support. Research has been supported by funding from various bodies including NERC (GR9/2913, NE/E003729/1, NE/E004539/1, and NE/E004148/1 NE/M0174781/1), the Royal Society, and the Universities of Birmingham, Leeds and Roehampton. LE is funded by a University of Leeds Anniversary Research Scholarship.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R ' S CO NTR I B UTI O N S
A.M.M. initiated the study of WPC and collected many of the preflood samples. A.M.M., A.L.R., M.J.K., and L.E.B were responsible for the funding applications to study the postfloods ecosystem, research design and planning, data collection and analysis, and writing the manuscript. J.P. and LC assisted with fieldwork and analyzed