Increased Variance as a Leading Indicator of Reorganization in Alaska Marine Ecosystems: An Empirical Test
School of Fisheries and Ocean Sciences, University of Alaska Fairbanks
Marine ecosystems may respond to external environmental or human forces—for example, climate change or fishing—with abrupt ecological reorganizations that can be economically and socially harmful to Alaska fishing communities. Currently, there is no way to predict these reorganization events. In this study, researchers will conduct a retrospective analysis of 12 different crustacean commercial fisheries with the goal of developing a technique called a "variance tracking system" to monitor and track ecosystem variables prior to and during reorganization events. With such a system, scientists hope to be able to predict sudden ecological change, and to help Alaska fisheries managers and fishing communities develop proactive responses and minimize resulting economic and social disruption.
An outstanding problem of ecosystem-based fisheries management is the propensity of ecosystems to undergo abrupt transitions between alternate stable states, whereby ecosystems respond to certain forces such as fishing or environmental change, with sudden reorganization once a critical threshold is reached. Most of these natural variables are rare in marine ecosystems, occurring on decadal time scales, but they are disruptive to fishing communities when they occur.
Although Alaska marine ecosystems are intensively monitored, currently there is no ability to detect an increased chance of sudden ecological transition. This lack of an early detection ability is a particular problem given uncertainty over the ecological implications of climate change in Alaska.
Simply speaking, "variance" is a term given to the many factors or parameters that describe an ecosystem state. By tracking and monitoring these parameters, scientists hope to develop a way to predict sudden and dramatic changes in the state of an ecosystem. The idea that the variance of key parameters can be used to track the resilience, or ability of an ecosystem to resist sudden change, is based on the idea that as the ecosystem loses resilience, it becomes easier for sudden changes to alter current stable state; and it then takes longer for the system to return to its former state. In general, an ecosystem exhibits more variance as resilience declines.
Alaska crustacean fisheries offer an excellent opportunity to examine the idea that tracking system variance offers a plausible way to predict system changes, since Alaska crustacean stocks experienced widespread collapse in the 1970s and 1980s due to the combined effects of overfishing and climate-forced ecological reorganization. In this study, researchers will conduct a retrospective analysis of 12 different crustacean populations described by 22 different time series with a cumulative length of 735 years.
Why is this an Alaska Sea Grant project?
One of Alaska Sea Grant's six key goals outlined in the 2009–2013 Strategic Plan is sustained, well-managed, and healthy marine, coastal, and watershed ecosystems in Alaska. The program pursues this goal through support of research that provides decision-makers with science-based information that can be used to craft well-informed policies governing the use and conservation of Alaska's marine and coastal resources.
How will researchers conduct their study?
This project centers around five specific goals. Three of these goals address variance dynamics within individual populations: (1) test for consistent increases in the variance of selected parameters prior to collapses in crustacean fisheries that occurred during ecological reorganization following the 1976–77 Pacific decadal oscillation (PDO) regime shift, (2) test for “false positives” by testing for increasing trends in variance in situations when fisheries did not collapse, and (3) compare the ability of fisheries-generated and fisheries-independent data sets to detect declining resilience. Two further study goals will employ aggregate results from different crustacean populations to: (4) develop management reference points for variance indication of impending fisheries collapse, and (5) contribute to the ability to measure resilience across ecosystems by aggregating variance parameters across multiple populations.
Alaska Department of Fish and Game
NOAA Alaska Fisheries Science Center
What researchers learned
Researchers report six major findings:
1. Spatial variability in catches increased prior to historical collapses in Alaska crustacean fisheries.
2. Increasing variability could be detected up to five years prior to a collapse.
3. Collapsing and non-collapsing fisheries showed statistically significant differences in variability trends, indicating that "false positive" signals were not present in the non-collapsing fisheries in our study.
4. Variability among vessels in catch and CPUE did not increase prior to collapse, indicating that vessel-scale parameters do not act as early warning indicators for collapse.
5. Skew in catch distributions did not increase prior to collapse, contradicting model predictions.
6. Increased variability was a persistent characteristic of post-collapse fisheries, contradicting model predictions of temporary increases in variability around collapse points.
To our knowledge, this study is the only application to date of variance tracking in a fisheries management context, and it is also one of a few studies to date to test the approach at very large spatial scales (i.e., over a shelf area on the order of 105 km2). The use of rising variance and other generic statistical indicators to provide early warning of impending shifts in ecosystems and populations remains a field that is dominated by modeling studies, with relatively little empirical research (Scheffer et al. 2009). Accordingly, we believe that our empirical work will make a valuable contribution to the understanding of the utility of these indicators in real-world settings.
In addition to the tests of spatial variance as an indicator of fisheries collapse that we identified in our proposal, we were able to test the hypothesis that increases in the skew of parameter distributions should signal declining resilience. Addition of this parameter to our analysis reflects the growing number of critical transition indicators proposed in the modeling literature, and the results of this additional analysis are reported in the paper from our study. This addition to our analysis compensated to some degree for difficulties that we experienced with fisheries-independent data sources that we identified in our proposal. During the course of our research we came to realize that most fishery-independent data sources that were available to us included too few years of data prior to population collapse to offer meaningful tests for trends in variability. When planning our research, we had assumed that a period of five years pre-collapse would be adequate for testing our hypotheses. However, inspection of our data revealed very high levels of background noise in parameters of interest, and a review of the literature on fisheries indicators suggested that ten years of data was the minimum for providing adequate statistical power for meaningful tests for trends. Although we expended considerable effort attempting to reconstruct usable fisheries-independent survey data, changes in survey methodology and the short duration of available time series ultimately led us to conclude that available survey data were inadequate for testing our hypotheses (see details below). The same was true of size class data from dockside sampling of commercial catches—these data were almost exclusively available from post-collapse fisheries, so that we could not test hypotheses concerning pre-collapse patterns of variability among size classes within exploited populations.
Details of the analysis conducted and summary results follow.
Indicators of marine ecosystem status typically require ten years of data for trends to be detected (Nicholson and Jennings 2004). Catch data for twelve collapsing fisheries met this criterion (mean pre-collapse n = 17 years), as did data for two non-collapsing fisheries (mean n = 26 years). While we worked with biologists responsible for maintaining different survey data sets in attempts to reconstruct usable survey time series, these efforts were often frustrated by changes to survey methodology and/or the general development of standardized fisheries-independent surveys just as crustacean populations were beginning to collapse in the late 1970s and early 1980s. In the end, surveys for only two fisheries (Bering Sea snow crab and St. Matthew Island blue king crab) met the criterion of at least 10 year of pre-collapse data. Given the high variability that we observed among catch time series in pre-collapse trends (i.e., only six of 12 fisheries showed statistically significant increases in variability pre-collapse, although a random-effects model showed strong evidence of increasing variability across all 12 fisheries, P < 0.0001), we concluded that a sample of two time series was inadequate for testing our hypotheses with survey data.
Catch data in our study were highly (positively) skewed, and the coefficient of variation, one of the more common measures of variability, produces systematic errors when used with skewed data (McArdle et al. 1990). We therefore used the standard deviation of log-transformed data (SDL), which is less sensitive to skew and maintains the desirable characteristic of being independent of the mean for commonly observed mean-variance scaling coefficients (McArdle et al. 1990). Our unit of analysis was the Alaska Department of Fish and Game statistical area. We also tested time series for increases in the skew of catch distribution among statistical areas, as spatial skew is also predicted to increase prior to a critical transition (Guttal and Jayaprakash 2009).
We analyzed trends in spatial variability and skew (slope on year) for each collapsing fishery with linear regression, using estimates of residual standard errors pooled across all 12 collapsing fisheries, with package nlme in the computer language R (Pinheiro and Bates 2000). These tests were one-tailed, since our hypotheses predict rising values (positive slopes) prior to collapse. We then made global estimates of slope across all collapsing fisheries, again using package nlme, using either full random-effects models (i.e., random-effect intercepts and slopes) or mixed-effects models (random-effect intercepts and fixed-effect slopes; Pinheiro and Bates 2000). Both random- and mixed-effects models estimated first-order autocorrelation as a separate parameter, and Aikaike's Information Criterion (AIC) was used to select the best model for tests of each hypothesis. We also used the best model, as determined by AIC, to test for "false positives" by comparing slopes between collapsing and non-collapsing fisheries. Trends for collapsing fisheries were calculated up to the year prior to collapse, while trends for non-collapsing fisheries were calculated across entire time series. Comparisons between collapsing and non-collapsing fisheries were also one-tailed, as our hypotheses predict greater slopes for collapsing than non-collapsing fisheries.
In cases where our hypotheses were supported, we assessed the length of warning that might be generated by the indicator in question by curtailing the data included in analysis 1–5 years in advance of the collapse, and again using either fixed- or random-effects models (depending on AIC results) to estimate trends in spatial variability across fisheries. These estimates were again subjected to one-tailed tests, which produced analysis of the ability to detect trends, given at least ten years of data in hand, from 1 to 5 years prior to a collapse. We also repeated our analysis using vessels as the sample unit, both for data on catch (n = 14 fisheries) and catch per unit effort (CPUE, kg per pot lift, n = 10 fisheries).
Fraterrigo, J.M., and J.A. Rusak. 2008. Disturbance-driven changes in the variability of ecological patterns and processes. Ecol. Lett. 11:756–770.
Guttal, V., and C. Jayaprakash. 2009. Spatial variance and spatial skewness: Leading indicators of regime shifts in spatial ecological systems. Theor. Ecol. 2:3–12.
McArdle, B.H., K.J. Gaston, and J.H. Lawton. 1990. Variation in the size of animal populations: Patterns, problems and artifacts. J. Anim. Ecol. 59:439–454.
Nicholson, M.D., and S. Jennings. 2004. Testing candidate indicators to support ecosystem-based management: The power of monitoring surveys to detect temporal trends in fish community metrics. ICES J. Mar. Sci. 61:35–42.
Pinheiro, J.C., and D.M. Bates. 2000. Mixed-effects models in S and S-plus. Springer, New York.
Scheffer, M., J. Bascompte, W.A. Brock, V. Brovkin, S.R. Carpenter, V. Dakos, H. Held, E.H. van Nes, M. Rietkerk, and G. Sugihara. 2009. Early-warning signals for critical transitions. Nature 461:53–59.
This new management tool has the potential to aid fisheries managers in anticipating ecological reorganization, and give them time to make management decisions that might avert or limit social, economic, and environmental consequences. Also the tool has the potential to offer coastal communities time to plan for and respond to ecological changes.
This study validated variance tracking as a management tool for predicting fisheries collapse. In the example of the Alaska king crab fishery collapse, researchers were able to detect increasing variability five years prior to the historical crustacean fishery collapses.
Venues for communicating our results include a completed manuscript for publication in a peer-reviewed journal, a series of presentations at statewide, national and international scientific conferences, a contribution to the annual Stock Assessment and Fisheries Evaluation report for the North Pacific Fishery Management Council, and public outreach through statewide media coverage, a public presentation in Kodiak, and a permanent display in the Kodiak NOAA lab.
Litzow, M.A., F.J. Mueter, and J.D. Urban. 2013. Rising catch variability preceded historical fisheries collapses in Alaska. Ecological Applications 23(6):1475–1487. http://doi.org/10.1890/12-0670.1