Complexity is costly: a meta‐analysis of parametric and non‐parametric methods for short‐term population forecasting

Short‐term forecasts based on time series of counts or survey data are widely used in population biology to provide advice concerning the management, harvest and conservation of natural populations. A common approach to produce these forecasts uses time‐series models, of different types, fit to time series of counts. Similar time‐series models are used in many other disciplines, however relative to the data available in these other disciplines, population data are often unusually short and noisy and models that perform well for data from other disciplines may not be appropriate for population data. In order to study the performance of time‐series forecasting models for natural animal population data, we assembled 2379 time series of vertebrate population indices from actual surveys. Our data were comprised of three vastly different types: highly variable (marine fish productivity), strongly cyclic (adult salmon counts), and small variance but long‐memory (bird and mammal counts). We tested the predictive performance of 49 different forecasting models grouped into three broad classes: autoregressive time‐series models, non‐linear regression‐type models and non‐parametric time‐series models. Low‐dimensional parametric autoregressive models gave the most accurate forecasts across a wide range of taxa; the most accurate model was one that simply treated the most recent observation as the forecast. More complex parametric and non‐parametric models performed worse, except when applied to highly cyclic species. Across taxa, certain life history characteristics were correlated with lower forecast error; specifically, we found that better forecasts were correlated with attributes of slow growing species: large maximum age and size for fishes and high trophic level for birds. Synthesis Evaluating the data support for multiple plausible models has been an integral focus of many ecological analyses. However, the most commonly used tools to quantify support have weighted models’ hindcasting and forecasting abilities. For many applications, predicting the past may be of little interest. Concentrating only on the future predictive performance of time series models, we performed a forecasting competition among many different kinds of statistical models, applying each to many different kinds of vertebrate time series of population abundance. Low‐dimensional (simple) models performed well overall, but more complex models did slightly better when applied to time series of cyclic species (e.g. salmon).     

Combining contemporary ecology and palaeolimnology to understand shallow lake ecosystem change

1. Palaeolimnology and contemporary ecology are complementary disciplines but are rarely combined. By reviewing the literature and using a case study, we show how linking the timescales of these approaches affords a powerful means of understanding ecological change in shallow lakes. 2. Recently, palaeolimnology has largely been pre‐occupied with developing transfer functions which use surface sediment‐lake environment datasets to reconstruct a single environmental variable. Such models ignore complex controls over biological structure and can be prone to considerable error in prediction. Furthermore, by reducing species assemblage data to a series of numbers, transfer functions neglect valuable ecological information on species’ seasonality, habitat structure and food web interactions. These elements can be readily extracted from palaeolimnological data with the interpretive assistance of contemporary experiments and surveys. For example, for one shallow lake, we show how it is possible to infer long‐term seasonality change from plant macrofossil and fossil diatom data with the assistance of seasonal datasets on macrophyte and algal dynamics. 3. On the other hand, theories on shallow lake functioning have generally been developed from short‐term (<1–15 years) studies as opposed to palaeo‐data that cover the actual timescales (decades–centuries) of shallow lake response to stressors such as eutrophication and climate change. Palaeolimnological techniques can track long‐term dynamics in lakes whilst smoothing out short‐term variability and thus provide a unique and important means of not only developing ecological theories, but of testing them. 4. By combining contemporary ecology and palaeolimnology, it should be possible to gain a fuller understanding of changing ecological patterns and processes in shallow lakes on multiple timescales.     

A better way to estimate population trends

Estimation of a population trend from a time series of abundance data is an important task in ecology, yet such estimation remains logistically and conceptually challenging in practice. First, the extent to which unequal intervals in the time series, due to missing observations or irregular sampling, compromise trend estimation is not well‐known. Furthermore, the predominant trend estimation method (loglinear regression of abundance data against time) ignores the possibility of process noise, while an alternative method (the ‘diffusion approximation’) ignores observation error in the abundance data. State‐space models that account for both process noise and observation error exist but have been little used. We study an adaptation of the exponential growth state‐space (EGSS) model for use with missing data in the time series, and we compare its trend estimation to the status quo methods. The EGSS model provides superior estimates of trend across wide ranges of time series length and sources of variation. The performance of the EGSS model even with half of the counts in the time series missing implies that trend estimates may be improved by diverting effort away from annual monitoring and towards increasing time series length or improving precision of the abundance estimates for years that data are collected.     

Band‐like Charge Photogeneration at a Crystalline Organic Donor/Acceptor Interface

Organic photovoltaic cells possess desirable practical characteristics, such as the potential for low‐cost fabrication on flexible substrates, but they lag behind their inorganic counterparts in performance due in part to fundamental energy loss mechanisms, such as overcoming the charge transfer (CT) state binding energy when photogenerated charge is transferred across the donor/acceptor interface. However, recent work has suggested that crystalline interfaces can reduce this binding energy due to enhanced CT state delocalization. Solar cells based on rubrene and C₆₀ are investigated as an archetypal system because it allows the degree of crystallinity to be moldulated from a highly disordered to highly ordered system. Using a postdeposition annealing method to transform as‐deposited amorphous rubrene thin films into ones that are highly crystalline, it is shown that the CT state of a highly crystalline rubrene/C₆₀ heterojunction undergoes extreme delocalization parallel to the interface leading to a band‐like state that exhibits a linear Stark effect. This state parallels the direct charge formation of inorganic solar cells and reduces energetic losses by 220 meV compared with 12 other archetypal heterojunctions reported in the literature.     

Empirical evaluation of observation scale effects in community time series

Natural communities are highly complex and dynamic over time, with populations structured by numerous abiotic and biotic forces acting through direct and indirect pathways. Multispecies Autoregressive (MAR) modeling can be used to partition effects of variables that are interrelated and temporally autocorrelated in time series from natural systems. Here we address two main questions in applying MAR models to community time series. First, what is the effect of observation scale on interpretation of community dynamics? We used a 10‐year weekly planktonic time series from Lake Washington to construct multiple “biweekly” and “monthly” data sets, and compared resulting community interaction models. Direct abiotic effects and intraspecific autocorrelation were apparent using all data sets. Biotic interactions were more apparent using biweekly and monthly data, indicating that time lags longer than one week were necessary to detect numerical response to interspecific interactions. Second, we examined effects of dropping the winter months from our analyses to simulate the common practice of sampling only during the “growing season” in long‐term ecological studies. We found that biotic interactions remained similarly characterized in models using only non‐winter months, but that the importance of seasonal physical factors nearly disappeared in non‐winter models. Exclusion of winter data in sampling designs may therefore allow us to characterize biotic interactions, although it may not help us understand populations’ relationships to seasonal abiotic variables. The models supported many previous findings from experimental and qualitative investigations of Lake Washington community interactions, implying that MARs provided plausible characterizations of community dynamics, but some previously unconsidered relationships did emerge, such as the importance of cryptomonads and picoplankton for zooplankton growth. We conclude that explicit consideration of time lags in biotic response is necessary to understand relative importance of abiotic and biotic factors, and that sampling regime can therefore strongly influence our interpretations of community dynamics.     

Historical introgression among the American live oaks and the comparative nature of tests for introgression

Introgressive hybridization challenges the concepts we use to define species and infer phylogenetic relationships. Methods for inferring historical introgression from the genomes of extant species, such as ABBA‐BABA tests, are widely used, however, their results can be easily misinterpreted. Because these tests are inherently comparative, they are sensitive to the effects of missing data (unsampled species) and nonindependence (hierarchical relationships among species). We demonstrate this using genomic RADseq data sampled from all extant species in the American live oaks (Quercus series Virentes), a group notorious for hybridization. By considering all species and their phylogenetic relationships, we were able to distinguish true hybridizing lineages from those that falsely appear admixed. Six of seven species show evidence of admixture, often with multiple other species, but which is explained by introgression among a few related lineages occurring in close proximity. We identify the Cuban oak as the most admixed lineage and test alternative scenarios for its origin. The live oaks form a continuous ring‐like distribution around the Gulf of Mexico, connected in Cuba, across which they could effectively exchange alleles. However, introgression appears highly localized, suggesting that oak species boundaries and their geographic ranges have remained relatively stable over evolutionary time.     

POTENTIAL PITFALLS OF RECONSTRUCTING DEEP TIME EVOLUTIONARY HISTORY WITH ONLY EXTANT DATA,A CASE STUDY USING THE CANIDAE (MAMMALIA,CARNIVORA)

Reconstructing evolutionary patterns and their underlying processes is a central goal in biology. Yet many analyses of deep evolutionary histories assume that data from the fossil record is too incomplete to include, and rely solely on databases of extant taxa. Excluding fossil taxa assumes that character state distributions across living taxa are faithful representations of a clade's entire evolutionary history. Many factors can make this assumption problematic. Fossil taxa do not simply lead‐up to extant taxa; they represent now‐extinct lineages that can substantially impact interpretations of character evolution for extant groups. Here, we analyze body mass data for extant and fossil canids (dogs, foxes, and relatives) for changes in mean and variance through time. AIC‐based model selection recovered distinct models for each of eight canid subgroups. We compared model fit of parameter estimates for (1) extant data alone and (2) extant and fossil data, demonstrating that the latter performs significantly better. Moreover, extant‐only analyses result in unrealistically low estimates of ancestral mass. Although fossil data are not always available, reconstructions of deep‐time organismal evolution in the absence of deep‐time data can be highly inaccurate, and we argue that every effort should be made to include fossil data in macroevolutionary studies.     

Observing climate change trends in ocean biogeochemistry: when and where

Understanding the influence of anthropogenic forcing on the marine biosphere is a high priority. Climate change‐driven trends need to be accurately assessed and detected in a timely manner. As part of the effort towards detection of long‐term trends, a network of ocean observatories and time series stations provide high quality data for a number of key parameters, such as pH, oxygen concentration or primary production (PP). Here, we use an ensemble of global coupled climate models to assess the temporal and spatial scales over which observations of eight biogeochemically relevant variables must be made to robustly detect a long‐term trend. We find that, as a global average, continuous time series are required for between 14 (pH) and 32 (PP) years to distinguish a climate change trend from natural variability. Regional differences are extensive, with low latitudes and the Arctic generally needing shorter time series (<~30 years) to detect trends than other areas. In addition, we quantify the ‘footprint’ of existing and planned time series stations, that is the area over which a station is representative of a broader region. Footprints are generally largest for pH and sea surface temperature, but nevertheless the existing network of observatories only represents 9–15% of the global ocean surface. Our results present a quantitative framework for assessing the adequacy of current and future ocean observing networks for detection and monitoring of climate change‐driven responses in the marine ecosystem.     

Detecting Change in Biological Rhythms: A Multivariate Permutation Test Approach to Fourier‐Transformed Data

Treatment‐related changes in neurobiological rhythms are of increasing interest to psychologists, psychiatrists, and biological rhythms researchers. New methods for analyzing change in rhythms are needed, as most common methods disregard the rich complexity of biological processes. Large time series data sets reflect the intricacies of underlying neurobiological processes, but can be difficult to analyze. We propose the use of Fourier methods with multivariate permutation test (MPT) methods for analyzing change in rhythms from time series data. To validate the use of MPT for Fourier‐transformed data, we performed Monte Carlo simulations and compared statistical power and family‐wise error for MPT to Bonferroni‐corrected and uncorrected methods. Results show that MPT provides greater statistical power than Bonferroni‐corrected tests, while appropriately controlling family‐wise error. We applied this method to human, pre‐ and post‐treatment, serially‐sampled neurotransmitter data to confirm the utility of this method using real data. Together, Fourier with MPT methods provides a statistically powerful approach for detecting change in biological rhythms from time series data.     

The development and persistence of alternative ecosystem states in a large,shallow lake

1. Palaeolimnological data were used to investigate drivers of the community of primary producers in Lake Mattamuskeet, North Carolina, U.S.A. This is a large, shallow lake with two basins currently dominated by phytoplankton and macrophytes. The two basins were divided in 1940 by the building of a roadway across the lake, which also corresponded with the divergence in their ecosystem state. 2. Photosynthetic pigments, organic matter and nutrients (P, N, C, S) were analysed in sediment cores from each basin to reconstruct the primary producer community over the past c. 100 years. We sought to answer two questions. First, what changes to the ecosystem resulting from the building of the roadway caused the development of different primary producer communities in the two basins? Second, why have the alternative ecosystem states persisted despite a variety of human perturbations since 1940? 3. K‐means cluster analysis and principal component analysis were applied to identify three sediment types based on photosynthetic pigment data: sediments indicating low productivity (low pigment concentrations), sediments associated with macrophytes (chlorophyll a and b) and with phytoplankton (alloxanthin and aphanizophyll). In addition, other palaeolimnological proxies measured, such as loss on ignition, total phosphorus, total organic carbon/total nitrogen and other nutrients, were different in post‐1940 sediments within the two basins. 4. These differences suggest characteristics, such as nutrient cycling, water depth and other physical changes resulting from roadway construction, combined to establish and maintain the differing communities of primary producers in the two basins. Furthermore, Fe/S dynamics and waterfowl herbivory probably contributed to the development of the two ecosystem states.     

Exploring the temporal variability of a food web using long‐term biomonitoring data

Ecological communities are constantly being reshaped in the face of environmental change and anthropogenic pressures. Yet, how food webs change over time remains poorly understood. Food web science is characterized by a trade‐off between complexity (in terms of the number of species and feeding links) and dynamics. Topological analysis can use complex, highly resolved empirical food web models to explore the architecture of feeding interactions but is limited to a static view, whereas ecosystem models can be dynamic but use highly aggregated food webs. Here, we explore the temporal dynamics of a highly resolved empirical food web over a time period of 18 years, using the German Bight fish and benthic epifauna community as our case study. We relied on long‐term monitoring ecosystem surveys (from 1998 to 2015) to build a metaweb, i.e. the meta food web containing all species recorded over the time span of our study. We then combined time series of species abundances with topological network analysis to construct annual food web snapshots. We developed a new approach, ‘node‐weighted’ food web metrics by including species abundances to represent the temporal dynamics of food web structure, focusing on generality and vulnerability. Our results suggest that structural food web properties change through time; however, binary food web structural properties may not be as temporally variable as the underlying changes in species composition. Further, the node‐weighted metrics enabled us to detect that food web structure was influenced by changes in species composition during the first half of the time series and more strongly by changes in species dominance during the second half. Our results demonstrate how ecosystem surveys can be used to monitor temporal changes in food web structure, which are important ecosystem indicators for building marine management and conservation plans.     

The use of long‐term monitoring data for studies of planktonic diversity: a cautionary tale from two Swiss lakes

相似文献   

Foraging strategies of generalist and specialist Old World nectar bats in response to temporally variable floral resources

Foraging theory predicts that generalist foragers should switch resources more readily, while specialist foragers should remain constant to preferred food resources. Plant‐pollinator interactions provide a convenient system to test such predictions because floral resources are often temporally patchy, thus requiring long‐lived pollinators to switch resources seasonally. Furthermore, flowering phenologies range from ‘steady‐state’ (low‐rewarding but highly reliable) to ‘big‐bang’ (high‐rewarding but ephemeral) plant species. We assessed how nectarivorous Old World bats respond to this temporally variable floral environment by examining their diets throughout the year. Over 15 months of fieldwork in southern Thailand, we simultaneously: (1) recorded the flowering phenologies of six bat‐pollinated plant taxa; and (2) assessed the diets of seven common flower‐visiting bat species. As predicted, the generalist nectarivore (Eonycteris spelaea) frequently switched diets and utilized both big‐bang and steady‐state resources, while the specialist nectarivores (Macroglossus minimus and M. sobrinus) foraged on one or two steady‐state plant species year‐round. Our results suggest that larger and faster bat species are able to fly longer distances in search of big‐bang resources, while smaller bat species rely on highly predictable food resources. This study supports the theory that generalist foragers have flexible diets, while specialist species restrict foraging to preferred floral resources even when other floral resources are more abundant. Moreover, these findings demonstrate how plant flowering phenology and pollinator diet breadth can shape the frequency and constancy of pollinator visits; we further discuss how such interactions can influence the potential extent of gene flow within a patchy floral environment.     

Age,state, environment,and season dependence of senescence in body mass

Senescence is a highly variable process that comprises both age‐dependent and state‐dependent components and can be greatly affected by environmental conditions. However, few studies have quantified the magnitude of age‐dependent and state‐dependent senescence in key life‐history traits across individuals inhabiting different spatially structured and seasonal environments. We used longitudinal data from wild female yellow‐bellied marmots (Marmota flaviventer), living in two adjacent environments that differ in elevation and associated phenology, to quantify how age and individual state, measured as “time to death,” affect body mass senescence in different environments. Further, we quantified how patterns of senescence differed between two biologically distinct seasons, spring, and late summer. Body mass senescence had an age‐dependent component, expressed as a decrease in mass in old age. Overall, estimated age‐dependent senescence was greater in females living in the more favorable lower elevation environment, than in the harsher higher elevation environment, and greater in late summer than in spring. Body mass senescence also had a state‐dependent component, captured by effects of time to death, but only in the more favorable lower elevation environment. In spring, body mass gradually decreased from 2 years before death, whereas in late summer, state‐dependent effects were expressed as a terminal decrease in body mass in the last year of life. Contrary to expectations, we found that senescence was more likely to be observed under more favorable environmental conditions, rather than under harsher conditions. By further demonstrating that senescence patterns differ among seasons, our results imply that within‐year temporal environmental variation must be considered alongside spatial environmental variation in order to characterize and understand the pattern and magnitude of senescence in wild populations.     

Region‐wide changes in marine ecosystem dynamics: state‐space models to distinguish trends from step changes

Regime shifts are sudden changes in ecosystem structure that can be detected across several ecosystem components. The concept that regime shifts are common in marine ecosystems has gained popularity in recent years. Many studies have searched for the step‐like changes in ecosystem state expected under a simple interpretation of this idea. However, other kinds of change, such as pervasive trends, have often been ignored. We assembled over 300 ecological time series from seven UK marine regions, covering two to three decades. We developed state‐space models for the first principal component of the time series in each region, a common measure of ecosystem state. Our models allowed both trends and step changes, possibly in combination. We found trends in three of seven regions and step changes in two of seven regions. Gradual and sudden changes are therefore important trajectories to consider in marine ecosystems.     

Direct negative density‐dependence in a pond‐breeding frog population

Understanding population dynamics is critical for the management of animal populations. Comparatively little is known about the relative importance of endogenous (i.e. density‐dependent) and exogenous (i.e. density‐independent) factors on the population dynamics of amphibians with complex life cycles. We examined the potential effects of density‐dependent and ‐independent (i.e. climatic) factors on population dynamics by analyzing a 15‐yr time series data of the agile frog Rana dalmatina population from Târnava Mare Valley, Romania. We used two statistical models: 1) the partial rate correlation function to identify the feedback structure and the potential time lags in the time series data and 2) a Gompertz state‐space model to simultaneously investigate direct and delayed density dependence as well as climatic effects on population growth rate. We found evidence for direct negative density dependence, whereas delayed density dependence and climate did not show a strong influence on population growth rate. Here we demonstrated that direct density dependence rather than delayed density dependence or climate determined the dynamics of our study population. Our results confirm the findings of many experimental studies and suggest that density dependence may buffer amphibian populations against environmental stress. Consequently, it may not be easy to scale up from individual‐level effects to population‐level effects.     

Nonlinear impacts of climatic variability on the density‐dependent regulation of an insect vector of disease

Aedes aegypti is one of the most common urban tropical mosquito species and an important vector of dengue, chikungunya, and yellow fever viruses. It is also an organism with a complex life history where larval stages are aquatic and adults are terrestrial. This ontogenetic niche shift could shape the density‐dependent regulation of this and other mosquito species, because events that occur during the larval stages impact adult densities. Herein, we present results from simple density‐dependent mathematical models fitted using maximum likelihood methods to weekly time series data from Puerto Rico and Thailand. Density‐dependent regulation was strong in both populations. Analysis of climatic forcing indicated that populations were more sensitive to climatic variables with low kurtosis, i.e., climatic factors highly variable around the median, rainfall in Puerto Rico, and temperature in Thailand. Changes in environmental variability appear to drive sharp changes in the abundance of mosquitoes. The identification of density‐independent (i.e., exogenous) variables forcing sharp changes in disease vector populations using the exogenous factors statistical properties, such as kurtosis, could be useful to assess the impacts of changing climate patterns on the transmission of vector‐borne diseases.     

Stepwise changes in viable but nonculturable Vibrio cholerae cells

Many bacterial species are known to become viable but nonculturable (VBNC) under conditions that are unsuitable for growth. In this study, the requirements for resuscitation of VBNC‐state Vibrio cholerae cells were found to change over time. Although VBNC cells could initially be converted to culturable by treatment with catalase or HT‐29 cell extract, they subsequently entered a state that was not convertible to culturable by these factors. However, fluorescence microscopy revealed the presence of live cells in this state, from which VBNC cells were resuscitated by co‐cultivation with HT‐29 human colon adenocarcinoma cells. Ultimately, all cells entered a state from which they could not be resuscitated, even by co‐cultivation with HT‐29. These characteristic changes in VBNC‐state cells were a common feature of strains in both V. cholerae O1 and O139 serogroups. Thus, the VBNC state of V. cholerae is not a single property but continues to change over time.     

An Australian Multi‐Regional Waste Supply‐Use Framework

The production of waste creates both direct and indirect environmental impacts. A range of strategies are available to reduce the generation of waste by industry and households, and to select waste treatment approaches that minimize environmental harm. However, evaluating these strategies requires reliable and detailed data on waste production and treatment. Unfortunately, published Australian waste data are typically highly aggregated, published by a variety of entities in different formats, and do not form a complete time‐series. We demonstrate a technique for constructing a multi‐regional waste supply‐use (MRWSU) framework for Australia using information from numerous waste data sources. This is the first MRWSU framework to be constructed (to the authors' knowledge) and the first sub‐national waste input‐output framework to be constructed for Australia. We construct the framework using the Industrial Ecology Virtual Laboratory (IELab), a cloud‐hosted computational platform for building Australian multi‐regional input‐output tables. The structure of the framework complies with the System of Environmental‐Economic Accounting (SEEA). We demonstrate the use of the MRWSU framework by calculating waste footprints that enumerate the full supply chain waste production for Australian consumers.