Multivariate Autoregressive State-space Models Research Articles

Addressing multidimensional highly correlated data for forecasting in precision beekeeping

In recent years, there have been relevant advances in precision beekeeping. These advances are mainly focused on proposing sensor systems that collect crucial information for bee welfare, creating integrated architectures that allow beekeepers to monitor the current state of their hive through real-time data. However, there is a lack of predictive models that would allow beekeepers to anticipate specific events that endanger bee welfare and lead to a decline in productivity. Specifically, predictive approaches accounting for the high correlation among internal variables of beehives have not been developed to date. To address this research gap, multivariate predictive models, including auto-regressive state-space and time series models, have been implemented and applied to four different hives from the we4bee project. These models aim to predict the internal variables of beehives (four different temperatures, humidity, and weight) by utilizing the meteorological conditions to which the hives are exposed. A cross-validation adapted to time series data was employed for model generalization assessment. Prediction models based on vector time series exhibited superior performance in forecasting internal hive variables compared to multivariate auto-regressive state-space models. Overall, the approach based on the vector error correction model yielded the best balance between fit, prediction, and computational cost. The VEC-based approach produces predictions with maximum mean absolute errors of 177 (312)g in weight, 3.366 (3.802)% in humidity, and 1.122 (1.685)°C in temperature at 1 (3)-days ahead when dealing with beehives exhibiting a high degree of correlation in their internal variables. Moreover, the VEC-based approach requires less than a second to perform the time series fitting process, which makes it particularly interesting for application in big data environments. The integration of such models into a decision support system would meet the need of beekeepers to anticipate potential threats to the welfare of their bee colonies, streamlining their monitoring processes while eliminating the need for continuous inspections.

Different patterns of coupled predator-prey dynamics when the same species interact in different locations.

Individual predator and prey species exhibit coupled population dynamics in simple laboratory systems and simple natural communities. It is unclear how often such pairwise coupling occurs in more complex communities, in which an individual predator species might feed on several prey species and an individual prey species might be attacked by several predators. To examine this problem, we applied multivariate autoregressive state-space (MARSS) models to 5-year time-series of monthly surveys of a predatory fish, the eastern mosquitofish (Gambusia holbrooki), and its littoral zone prey species, the least killifish (Heterandria formosa), in three locations in north Florida. The MARSS models were consistent with coupled predator-prey dynamics at two of the three locations. In one of these two locations, the estimated densities of the two species displayed classic predator-prey oscillations. In the third location, there was a positive effect of killifish density on mosquitofish density but no detectable effect of mosquitofish density on killifish density. In all three locations, increased submergent vegetation cover was associated with increased prey density but not increased predator density. Eigenvalues analyses for the joint predator-prey dynamics indicated that one of the cyclic locations had more stable dynamics than the other locations. The three different patterns demonstrate that the dynamics of a pairwise predator-prey interaction emerge not only from the characteristics of the prey and the predator, but also those of the habitat and trophic web in which the predator and prey are embedded.

Inverse relationships between coyote and wild turkey population time series: Implications for future studies of predator–prey interactions

AbstractChanges in predator communities such as the emergence of new predators alter predator–prey interactions profoundly. Coyotes (Canis latrans) have expanded their geographic distributions from the western United States (US) to the Eastern US, potentially suppressing some prey populations as their relative abundance has increased. To assess the relationship between coyotes and one of their prey, the eastern wild turkey (Meleagris gallopavo silvestris), we employed time series analysis and multivariate autoregressive state‐space models (MARSSMs) to a 23‐year capture per unit effort data set from both species in the state of Mississippi, US. The best MARSSM model indicated density dependence in each species. The competing MARSSM indicated that increases in coyote relative abundance reduced the population growth rate of wild turkeys. We recommend that future studies use long‐term time series of multiple coexisting predators and their alternative prey to assess the effects of the predator guild on the dynamics of prey populations.

Using multivariate autoregressive state‐space models to examine stock structure of Greenland halibut in the North Atlantic

AbstractAccurate information on population structure is essential for effective fisheries management. Greenland halibut (Reinhardtius hippoglossoides) in the North Atlantic is managed as four separate offshore stocks. We use Multivariate Autoregressive State‐Space (MARSS) models to assess population structure by means of abundance and biomass trends in four regions (Norwegian Sea, Iceland, Southeast Greenland, and Northwest Atlantic) where three offshore stocks are recognized: (1) Baffin Bay–Davis Strait (Northwest Atlantic stock), (2) Southeast Greenland and Iceland (West Nordic stock (WNS)), and (3) the Barents and Norwegian Seas (Northeast Arctic stock). We formulated model alternatives, using bottom trawl survey data from each region for 1996–2019, to evaluate support for different population structures. Abundance and biomass observations from each region were linked to growth rate parameters in MARSS models and the impact of climate (North Atlantic Oscillation Index) and fishing (commercial catches) on stock dynamics was investigated. Top models identified the Northwest Atlantic as an independent population. Best‐fit models treated Greenland halibut in the WNS as two independent populations (east and west), with potential connections between eastern Iceland and the western Barents Sea. These results suggest a mismatch between current stock perception and management boundaries in the Northeast Atlantic.

Four decades of climatic fluctuations and fish recruitment stability across a marine-freshwater gradient.

Investigating the effects of climatic variability on biological diversity, productivity, and stability is key to understanding possible futures for ecosystems under accelerating climate change. A critical question for estuarine ecosystems is, how does climatic variability influence juvenile recruitment of different fish species and life histories that use estuaries as nurseries? Here we examined spatiotemporal abundance trends and environmental responses of 18 fish species that frequently spend the juvenile stage rearing in the San Francisco Estuary, CA, USA. First, we constructed multivariate autoregressive state‐space models using age‐0 fish abundance, freshwater flow (flow), and sea surface temperature data (SST) collected over four decades. Next, we calculated coefficients of variation (CV) to assess portfolio effects (1) within and among species, life histories (anadromous, marine opportunist, or estuarine dependent), and the whole community; and (2) within and among regions of the estuary. We found that species abundances varied over space and time (increasing, decreasing, or dynamically stable); and in 83% of cases, in response to environmental conditions (wet/dry, cool/warm periods). Anadromous species responded strongly to flow in the upper estuary, marine opportunist species responded to flow and/or SST in the lower estuary, and estuarine dependent species had diverse responses across the estuary. Overall, the whole community when considered across the entire estuary had the lowest CV, and life histories and species provided strong biological insurance to the portfolio (2.4‐ to 3.5‐fold increases in stability, respectively). Spatial insurance also increased stability, although to a lesser extent (up to 1.6‐fold increases). Our study advances the notion that fish recruitment stability in estuaries is controlled by biocomplexity—life history diversity and spatiotemporal variation in the environment. However, intensified drought and marine heatwaves may increase the risk of multiple consecutive recruitment failures by synchronizing species dynamics and trajectories via Moran effects, potentially diminishing estuarine nursery function.

Spatial patterns of Plasmodium vivax transmission explored by multivariate auto-regressive state-space modelling - A case study in Baoshan Prefecture in southern China.

The transition from the control phase to elimination of malaria in China through the national malaria elimination programme has focussed attention on the need for improvement of the surveillance- response systems. It is now understood that routine passive surveillance is inadequate in the parasite elimination phase that requires supplementation by active surveillance in foci where cluster cases have occurred. This study aims to explore the spatial clusters and temporal trends of malaria cases by the multivariate auto-regressive state-space model (MARSS) along the border to Myanmar in southern China. Data for indigenous cases spanning the period from 2007 to 2010 were extracted from the China's Infectious Diseases Information Reporting Management System (IDIRMS). The best MARSS model indicated that malaria transmission in the study area during 36 months could be grouped into three clusters. The estimation of malaria transmission patterns showed a downward trend across all clusters. The proposed methodology used in this study offers a simple and rapid, yet effective way to categorize patterns of foci which provide assistance for active monitoring of malaria in the elimination phase.

Long-term plankton community dynamics and influencing factors in a man-made shallow lake, Lake Dishui, China

Understanding how long-term environmental variation and biotic interaction determine biological assemblages is critical for evaluating the response of biological communities to anthropogenic and climate changes. We sampled plankton and water quality from a man-made lake monthly from 2006 to 2015 for the period from after creation to 13-year old of the lake. We used correlation, variation partitioning, multivariate autoregressive state-space models, and dynamic factor analyses to disentangle interactions of biotic and abiotic factors in regulating the dynamics of plankton size groups. We found that annual average total abundances decreased 86% for phytoplankton and 78% for zooplankton, while concentrations increased 1.8 times for total nitrogen and 2.7 times for total phosphorus during the study period. We concluded that the decline of plankton abundance was largely attributed to the predation of zooplankton on phytoplankton, fish predation on both phytoplankton and zooplankton, and linkages with the decline of water salinity, pH, and the increase of clarity. The decline of plankton abundance was apparently not resulted from long-term change in nutrient concentrations. These conclusions were further supported by that abundances of nano-phytoplankton and micro-phytoplankton oscillated in opposite direction with extremely low density of Cladocera (< 1 individual/L) during the study period. Our conclusions were also supported by the strong interactions between Rotifera and macro-phytoplankton and between nano-phytoplankton and Copepoda, as well as by the describability of the six plankton size groups in a shared trend that was opposite to the trends of water clarity and planktivorous fish stock density.

Incorporating spatial synchrony in the status assessment of a threatened species with multivariate analysis

Spatial synchrony—correlated abundance fluctuations among distinct populations—is associated with increased extinction risk but is not a component of widely-used extinction risk assessments (e.g., IUCN Red List, U.S. Fish and Wildlife Service's Species Status Assessment). Alongside traditional viability metrics (i.e., the number of populations, their spatial extent, the status of each population), consideration of spatial synchrony in these assessments may provide additional insight into extinction risk as well as the relative importance of intrinsic and extrinsic factors on population dynamics. We demonstrate a method for estimating abundance trends in populations of the endangered freshwater fish, the amber darter (Percina antesella), while simultaneously assessing empirical support for existence of spatial synchrony among its two populations in the Conasauga and Etowah rivers in Georgia, U.S.A. Our analysis was performed using multivariate autoregressive state-space (MARSS) models with annual sampling data from 1996 to 2018 at 16 sites distributed between the two rivers. Our results indicate that amber darter populations have declined substantially, with 9% annual losses in both the Conasauga and Etowah rivers, suggesting rangewide imperilment. Furthermore, model selection indicated little support for models with fully independent dynamics between rivers, which may compound overall extinction risk. This analysis demonstrates the utility of tools such as MARSS models for assessing spatial synchrony and long-term population trajectories of imperiled species, resulting in improved vulnerability assessments that do not assume independence among separate populations.

Drivers shaping episodic and gradual changes in phytoplankton community succession: Taxonomic versus functional groups

Describing temporal changes in phytoplankton communities is complicated owing to (i) multivariate environmental drivers, (ii) inter-specific relationships, and (iii) various species. With long-term research data from the lower Nakdong River from 1993 to 2016, we examined the temporal changes at two scales—episodic (from weekly to monthly) and long-term (yearly)—and screened the potential environmental drivers. Phytoplankton community component patterns were modeled with the drivers as covariates, using multivariate autoregressive state-space (MARSS) models, to assess their response to environmental drivers and biotic interactions. We assumed that compared to taxonomic classification, functional classification would obtain a better identification of community response to temporal variability. Over 24 years, the succession patterns of the dominant taxonomic and functional groups decreased in diversity, with the greatest decreases in biomass of Bacillariophyceae and group D (mainly the diatom Stephanodiscus hantzschii), and coincided with the introduction of group H1 (dinitrogen-fixing nostocaleans). The potential drivers for these changes were precipitation, water level, and total nitrogen (TN) for taxonomic groups and TN, total phosphorus, and euphotic zone depth for functional groups. The results of the MARSS model and temporal trends for each driver indicated that the increases in the water level and light availability were mostly related with the taxonomic and functional groups, respectively. The model for functional groups proposed a total of 24 significant inter-group relationships, where five relationships supported the succession patterns of dominant groups in the Nakdong River. Combined with the effects of increased light availability, a positive relationship between groups H1 and M (mainly Cyanobacteria and Microcystis aeruginosa) appears to induce cyanobacterial bloom development over a long period. These results can be fundamental information for river system management concerning the resulting cascading effects of changes in environmental drivers and inter-group relationships on the phytoplankton community composition.

Using multivariate state-space models to examine commercial stocks of redfish (Sebastes spp.) on the Flemish Cap

There are three different species of redfish (Sebastes spp.) in the waters of the Flemish Cap (Division 3M, NAFO Regulatory Area): S. fasciatus, S. mentella, and S. norvegicus. Historically, S. fasciatus and S. mentella have been managed together as a single stock because of similar biology and difficulty in species identification. Here we use multivariate autoregressive state-space models to examine the abundance trajectories of the three species and to determine whether they can be treated as a single stock for management purposes or whether they should be treaty separately. We also included covariates to evaluate relationships with climate, commercial catch, and the abundance of predators and (or) competitors and prey. We did two separate analyses: (i) a single-period analysis over the full time series and (ii) a blocked, two-period analysis over different regulatory periods. In both analyses, the best-fit model included separate trajectories for each species at each depth but one overall stock growth rate; both also included commercial catches as a covariate. These analyses suggest that a single assessment for the Sebastes complex is acceptable.

Making the most of incomplete long-term datasets: the MARSS solution

ABSTRACT Long-term field-based monitoring is essential to develop a deep understanding of how ecosystems function and to identify species at risk of decline. However, conducting long-term field-based research poses some unique challenges due to the harsh environmental conditions or extreme weather events that may be encountered. Such conditions are especially likely to occur in arid environments. Fieldwork issues can arise from vehicle breakdowns, wildfires and heavy rainfall events, all of which can delay or even cancel data collection. In addition, long-term monitoring typically requires multiple observers, which may add observation bias to estimates of measured parameters. Thus there is an increasing need to develop new statistical techniques that take advantage of the power of long time-series datasets that also are incomplete. Here we discuss multivariate autoregressive state-space (MARSS) modelling; a relatively new statistical technique for modelling long-term time-series data. MARSS models allow u...

Using state-space models to predict the abundance of juvenile and adult sea lice on Atlantic salmon

Sea lice are marine parasites affecting salmon farms, and are considered one of the most costly pests of the salmon aquaculture industry. Infestations of sea lice on farms significantly increase opportunities for the parasite to spread in the surrounding ecosystem, making control of this pest a challenging issue for salmon producers. The complexity of controlling sea lice on salmon farms requires frequent monitoring of the abundance of different sea lice stages over time. Industry-based data sets of counts of lice are amenable to multivariate time-series data analyses.In this study, two sets of multivariate autoregressive state-space models were applied to Chilean sea lice data from six Atlantic salmon production cycles on five isolated farms (at least 20 km seaway distance away from other known active farms), to evaluate the utility of these models for predicting sea lice abundance over time on farms. The models were constructed with different parameter configurations, and the analysis demonstrated large heterogeneity between production cycles for the autoregressive parameter, the effects of chemotherapeutant bath treatments, and the process-error variance. A model allowing for different parameters across production cycles had the best fit and the smallest overall prediction errors. However, pooling information across cycles for the drift and observation error parameters did not substantially affect model performance, thus reducing the number of necessary parameters in the model. Bath treatments had strong but variable effects for reducing sea lice burdens, and these effects were stronger for adult lice than juvenile lice. Our multivariate state-space models were able to handle different sea lice stages and provide predictions for sea lice abundance with reasonable accuracy up to five weeks out.

Declining patterns of Pacific Northwest steelhead trout (Oncorhynchus mykiss) adult abundance and smolt survival in the ocean

Examination of population abundance and survival trends over space and time can guide management and conservation actions with information about the spatial and temporal scale of factors affecting them. Here, we analyzed steelhead trout (anadromous Oncorhynchus mykiss) adult abundance time series from 35 coastal British Columbia and Washington populations along with smolt-to-adult return (smolt survival) time series from 48 populations from Washington, Oregon, and the Keogh River in British Columbia. Over 80% of the populations have declined in abundance since 1980. A multivariate autoregressive state-space model revealed smolt survival four groupings: Washington and Oregon coast, lower Columbia River, Strait of Juan de Fuca, and Puget Sound – Keogh River populations. Declines in smolt survival rates were seen for three of the four groupings. Puget Sound and Keogh River populations have experienced low rates since the early 1990s. Correlations between population pairs’ time series and distance apart illustrated that smolt survival rates were more positively correlated for proximate populations, suggesting that important processes, including those related to ocean survival, occur early in the marine life of steelhead.

Using a multivariate auto-regressive state-space (MARSS) model to evaluate fishery resources abundance in the East China Sea, based on spatial distributional information

The abundance index (AI) is a representative indicator used to assess the state of fishery resources. Conventional AI is generally calculated by summing the catch per unit of effort (CPUE) weighted by the size of each fishing area. However, CPUE data has many missing values owing to annual changes in operational fishing areas, and this can lead to a considerable bias in the estimated AI. To obtain an unbiased AI, a multivariate auto-regressive state-space (MARSS) model was used to estimate and interpolate missing values in a spatially arranged, long-term bottom-trawl CPUE dataset for yellow seabream Dentex hypselosomus and largehead hairtail Trichiurus japonicus in the East China Sea. As expected, increasing the number of analyzed fishing grids improved interpolation accuracy, but remarkably increased the time required for the analysis. Reducing the maximum number of expectation–maximization (EM) iterations in the maximum likelihood procedure was an effective way to practically reduce analysis time, while keeping the accuracy of the estimation. Thus, this EM-reduction MARSS model was applied to the entire CPUE datasets of yellow seabream and largehead hairtail to address the annual shifts in their AIs and their seasonal migration.

Population assessment using multivariate time-series analysis: A case study of rockfishes in Puget Sound.

Estimating a population's growth rate and year‐to‐year variance is a key component of population viability analysis (PVA). However, standard PVA methods require time series of counts obtained using consistent survey methods over many years. In addition, it can be difficult to separate observation and process variance, which is critical for PVA. Time‐series analysis performed with multivariate autoregressive state‐space (MARSS) models is a flexible statistical framework that allows one to address many of these limitations. MARSS models allow one to combine surveys with different gears and across different sites for estimation of PVA parameters, and to implement replication, which reduces the variance‐separation problem and maximizes informational input for mean trend estimation. Even data that are fragmented with unknown error levels can be accommodated. We present a practical case study that illustrates MARSS analysis steps: data choice, model set‐up, model selection, and parameter estimation. Our case study is an analysis of the long‐term trends of rockfish in Puget Sound, Washington, based on citizen science scuba surveys, a fishery‐independent trawl survey, and recreational fishery surveys affected by bag‐limit reductions. The best‐supported models indicated that the recreational and trawl surveys tracked different, temporally independent assemblages that declined at similar rates (an average of −3.8% to −3.9% per year). The scuba survey tracked a separate increasing and temporally independent assemblage (an average of 4.1% per year). Three rockfishes (bocaccio, canary, and yelloweye) are listed in Puget Sound under the US Endangered Species Act (ESA). These species are associated with deep water, which the recreational and trawl surveys sample better than the scuba survey. All three ESA‐listed rockfishes declined as a proportion of recreational catch between the 1970s and 2010s, suggesting that they experienced similar or more severe reductions in abundance than the 3.8–3.9% per year declines that were estimated for rockfish populations sampled by the recreational and trawl surveys.

The long-term population dynamics of common wasps in their native and invaded range.

Populations of introduced species are often thought to perform differently, or experience different population dynamics, in their introduced range compared to their native habitat. Differences between habitats in climate, competition or natural enemies may result in populations with varying density dependence and population dynamics. We examined the long-term population dynamics of the invasive common wasp, Vespula vulgaris, in its native range in England and its invaded range in New Zealand. We used 39years of wasp density data from four sites in England, and 23years of data from six sites in New Zealand. Wasp population time series was examined using partial rate correlation functions. Gompertz population models and multivariate autoregressive state-space (MARSS) models were fitted, incorporating climatic variation. Gompertz models successfully explained 59-66% of the variation in wasp abundance between years. Density dependence in wasp populations appeared to act similarly in both the native and invaded range, with wasp abundance in the previous year as the most important variable in predicting intrinsic rate of increase (r). No evidence of cyclic population dynamics was observed. Both the Gompertz and MARSS models highlighted the role of weather conditions in each country as significant predictors of annual wasp abundance. The temporal evolution of wasp populations at all sites was best modelled jointly using a single latent dynamic factor for local trends, with the inclusion of a latent spring weather covariate. That same parsimonious multivariate model structure was optimal in both the native and invaded range. Density dependence is overwhelmingly important in predicting wasp densities and 'wasp years' in both the native and invaded range. Spring weather conditions in both countries have a major influence, probably through their impact on wasp colony initiation and early development. The population dynamics in the native range and invaded range show no evidence of cyclic boom-and-bust dynamics. Invasive species may not exhibit different population dynamics despite considerable variation in abundances throughout their distribution.

A Multivariate Control Chart for Autocorrelated Tool Wear Processes

Full automation of metal cutting processes has been a long held goal of the manufacturing industry. One key obstacle to achieving this ambition has been the inability to monitor completely the condition of the cutting tool in real time, as premature tool breakage and heavy tool wear can result in substantial costs through damage to the machinery and increasing the risk of non‐conforming items that have to be scrapped or reworked. Instead, the condition of the tool has to be indirectly monitored using modern sensor technology that measures the acoustic emission, sound, spindle power and vibration of the tool during a cut. An online monitoring procedure for such data is proposed. Firstly, the standard deviation is extracted from each sensor signal to summarise the state of the tool after each cut. Secondly, a multivariate autoregressive state space model is specified for estimating the joint effects and cross‐correlation of the sensor variables in Phase I. Then we apply a distribution‐free monitoring scheme to the model residuals in Phase II, based on binomial type statistics. The proposed methodology is illustrated using a case study of titanium alloy milling (a machining process used in the manufacture of aircraft landing gears) from the Advanced Manufacturing Research Centre in Sheffield, UK, and is demonstrated to outperform alternative residual control charts in this application. © 2016 The Authors Quality and Reliability Engineering International Published by John Wiley &amp; Sons Ltd.

Spatial and temporal synchrony in reptile population dynamics in variable environments.

Resources are seldom distributed equally across space, but many species exhibit spatially synchronous population dynamics. Such synchrony suggests the operation of large-scale external drivers, such as rainfall or wildfire, or the influence of oasis sites that provide water, shelter, or other resources. However, testing the generality of these factors is not easy, especially in variable environments. Using a long-term dataset (13-22years) from a large (8000km(2)) study region in arid Central Australia, we tested firstly for regional synchrony in annual rainfall and the dynamics of six reptile species across nine widely separated sites. For species that showed synchronous spatial dynamics, we then used multivariate follow a multivariate auto-regressive state-space (MARSS) models to predict that regional rainfall would be positively associated with their populations. For asynchronous species, we used MARSS models to explore four other possible population structures: (1) populations were asynchronous, (2) differed between oasis and non-oasis sites, (3) differed between burnt and unburnt sites, or (4) differed between three sub-regions with different rainfall gradients. Only one species showed evidence of spatial population synchrony and our results provide little evidence that rainfall synchronizes reptile populations. The oasis or the wildfire hypotheses were the best-fitting models for the other five species. Thus, our six study species appear generally to be structured in space into one or two populations across the study region. Our findings suggest that for arid-dwelling reptile populations, spatial and temporal dynamics are structured by abiotic events, but individual responses to covariates at smaller spatial scales are complex and poorly understood.

Quasi-extinction risk and population targets for the Eastern, migratory population of monarch butterflies (Danaus plexippus).

The Eastern, migratory population of monarch butterflies (Danaus plexippus), an iconic North American insect, has declined by ~80% over the last decade. The monarch’s multi-generational migration between overwintering grounds in central Mexico and the summer breeding grounds in the northern U.S. and southern Canada is celebrated in all three countries and creates shared management responsibilities across North America. Here we present a novel Bayesian multivariate auto-regressive state-space model to assess quasi-extinction risk and aid in the establishment of a target population size for monarch conservation planning. We find that, given a range of plausible quasi-extinction thresholds, the population has a substantial probability of quasi-extinction, from 11–57% over 20 years, although uncertainty in these estimates is large. Exceptionally high population stochasticity, declining numbers, and a small current population size act in concert to drive this risk. An approximately 5-fold increase of the monarch population size (relative to the winter of 2014–15) is necessary to halve the current risk of quasi-extinction across all thresholds considered. Conserving the monarch migration thus requires active management to reverse population declines, and the establishment of an ambitious target population size goal to buffer against future environmentally driven variability.

Population diversity in Pacific herring of the Puget Sound, USA.

Demographic, functional, or habitat diversity can confer stability on populations via portfolio effects (PEs) that integrate across multiple ecological responses and buffer against environmental impacts. The prevalence of these PEs in aquatic organisms is as yet unknown, and can be difficult to quantify; however, understanding mechanisms that stabilize populations in the face of environmental change is a key concern in ecology. Here, we examine PEs in Pacific herring (Clupea pallasii) in Puget Sound (USA) using a 40-year time series of biomass data for 19 distinct spawning population units collected using two survey types. Multivariate auto-regressive state-space models show independent dynamics among spawning subpopulations, suggesting that variation in herring production is partially driven by local effects at spawning grounds or during the earliest life history stages. This independence at the subpopulation level confers a stabilizing effect on the overall Puget Sound spawning stock, with herring being as much as three times more stable in the face of environmental perturbation than a single population unit of the same size. Herring populations within Puget Sound are highly asynchronous but share a common negative growth rate and may be influenced by the Pacific Decadal Oscillation. The biocomplexity in the herring stock shown here demonstrates that preserving spatial and demographic diversity can increase the stability of this herring population and its availability as a resource for consumers.