Search

To support ongoing marine spatial planning in New Zealand, a numerical environmental classification using Gradient Forest models was developed using a broad suite of biotic and high-resolution environmental predictor variables. Gradient Forest modeling uses species distribution data to control the selection, weighting and transformation of environmental predictors to maximise their correlation with species compositional turnover. A total of 630,997 records (39,766 unique locations) of 1,716 taxa living on or near the seafloor were used to inform the transformation of 20 gridded environmental variables to represent spatial patterns of compositional turnover in four biotic groups and the overall seafloor community. Compositional turnover of the overall community was classified using a hierarchical procedure to define groups at different levels of classification detail. The 75-group level classification was assessed as representing the highest number of groups that captured the majority of the variation across the New Zealand marine environment. We refer to this classification as the New Zealand “Seafloor Community Classification” (SCC). Associated uncertainty estimates of compositional turnover for each of the biotic groups and overall community were also produced, and an added measure of uncertainty – coverage of the environmental space – was developed to further highlight geographic areas where predictions may be less certain owing to low sampling effort. Environmental differences among the deep-water New Zealand SCC groups were relatively muted, but greater environmental differences were evident among groups at intermediate depths in line with well-defined oceanographic patterns observed in New Zealand’s oceans. Environmental differences became even more pronounced at shallow depths, where variation in more localised environmental conditions such as productivity, seafloor topography, seabed disturbance and tidal currents were important differentiating factors. Environmental similarities in New Zealand SCC groups were mirrored by their biological compositions. The New Zealand SCC is a significant advance on previous numerical classifications and includes a substantially wider range of biological and environmental data than has been attempted previously. The classification is critically appraised and considerations for use in spatial management are discussed.

Robust design of Marine Protected Areas in ocean environments is often challenging due to inadequate knowledge of biodiversity patterns, reflecting difficulties in the prediction of species distributions from sampling data that are often sparse or inadequate. Models that combine species and environmental data, such as Gradient Forests (GF), provide one analytical approach to this problem, efficiently combining available information to produce spatial models of species turnover throughout an area of interest. Spatial estimates of species turnover can then be classified to estimate spatial patterns in species composition; however, the performance of GF-based classifications within a conservation planning context has not previously been evaluated. Here we assess the utility for conservation planning (using the software Zonation) of a GF-based hierarchical classification that summarises spatial patterns in demersal fish composition in the oceans around New Zealand. Progressively more complex Zonation analyses assessed the effects of (i) varying the number of classification groups, (ii) adding information describing species turnover, and (iii) adding information describing spatial variation in demersal fish species richness. The best-performing GF-based conservation ranking used layers describing the distributions of 30 classification groups, demersal fish species turnover between these groups, and species richness. Conservation outcomes from this ranking were only marginally less efficient than those from a more conventional ranking that used 217 individual species distribution layers (7% less efficient). This indicates that GF-based classifications may provide a practical alternative for marine conservation planning. Additional advantages arise from the greater ease with which a single classification layer summarising complex biodiversity patterns can facilitate decision-making in participatory stakeholder processes.

Several international agreements and conventions require nations to establish Marine Protected Area (MPA) networks as an approach to alleviating biodiversity declines; however, a common problem in planning MPA networks is how to balance conservation objectives against economic objectives. Here, using the distributions of 102 biodiversity features and 7 extractive uses we trial the systematic conservation planning software Zonation as a decision-support tool to facilitate progress towards New Zealand's commitment to establishing a representative network of MPAs while providing for economic development. Our results indicate that: (i) New Zealand's existing MPAs provide on average 70% less representation of the input biodiversity features than would be achieved by an MPA network of equivalent area designed from the outset using Zonation; (ii) small increases in the geographic extent of existing protection results in rapid increases in representation of the selected biodiversity features when systematic conservation planning software is used to inform expansion of existing protection; and (iii) the impacts on existing resource users of an expanded MPA system can be minimized by using Zonation to identify areas that increase biodiversity representation, while avoiding areas where existing uses may be incompatible with marine protection. These results demonstrate the utility of systematic conservation planning software as a decision-support tool within a broader social process for MPA network design and implementation. The iterative application of tools such as Zonation during participatory processes that balance alternative uses could potentially lead to more informed, efficient and socially enduring outcomes that enhance the ability to establish representative MPA networks.

Spatial classifications of the environment have previously been used to characterise biodiversity and to facilitate management planning at large spatial scales. Such classifications are more likely to be adopted if they can demonstrate integration of real patterns in habitats or biotic assemblages, in addition to environment. A previous classification used Gradient Forest analysis to derive 30 classes based on demersal fish assemblage patterns and environmental gradients. Here we provide a detailed description of the similarities and differences in the environment and fish assemblages of classes resulting from an updated classification using the same methodology. Environmental differences were associated with varying levels of differences in the distributions of fish species. At broad spatial scales, assemblages are differentiated primarily according to oceanographic conditions such as temperature and depth; at finer scales, patterns in species assemblages are more closely associated with more localised environmental conditions such as productivity, sea-surface temperature gradients and tidal currents. The 30-group classification allows complex biodiversity information to be summarised in ways accessible to stakeholder and environmental managers. Given the hierarchical nature of the classification, there is considerable scope to use a larger number of groups for applications at regional to local scales.

Predictions of invasion risk for seven non-indigenous fish species, ecological impact scores for individual species, and lake conservation rankings were linked to develop Invasion Risk Impact (IRI) and Lake Vulnerability (LV) indices that help identify New Zealand lakes most at risk of loss of conservation value from potential multi-species invasions. Species-specific IRI scores (the product of predicted invasion risk and species impact) highlighted Eurasian perch (Perca fluviatilis) and the brown bullhead (Ameiurus nebulosus), as the species most likely to spread and cause ecological harm in lakes. For 3431 lakes >1 ha throughout New Zealand, total IRI tended to be highest for lowland riverine and dune lakes most of which are already colonized by multiple invasive fish species. The LV index indicated that lakes most at risk of loss of conservation value from invasive fish impacts were predominantly (i) in the northern half of the North Island where several uncommon lake types occur, and (ii) along the west coast of the South Island where conservation value is often greater, largely because of low catchment modification. The IRI and LV indices can be used to assist with setting priorities for surveillance monitoring, advocacy, and response planning targeted at preventing the establishment of invasive fish in moderate-to-high value lakes most susceptible to ecological impacts. Both indices can be adapted to accommodate alternative impact and conservation scoring systems, providing a flexible tool for local- and national-scale assessments of lake vulnerability to fish invasion impacts. Copyright © 2016 John Wiley & Sons, Ltd.

Responses of macroinvertebrate communities to human pressure are poorly known in large rivers compared with wadeable streams, in part because of variable substrate composition and the need to disentangle pressure responses from underlying natural environmental variation. To investigate the interaction between these factors, we sampled macroinvertebrates from the following: (i) submerged wood; (ii) littoral substrates 1.5 m) benthic habitats in eleven 6th- or 7th-order New Zealand rivers spanning a catchment vegetation land cover gradient. Cluster analysis identified primary site groupings reflecting regional environmental characteristics and secondary groupings for moderate gradient rivers reflecting the extent of catchment native vegetation cover. Low pressure sites with high levels of native vegetation had higher habitat quality and higher percentages of several Ephemeroptera and Trichoptera taxa than sites in developed catchments, whereas developed sites were more typically dominated by Diptera, Mollusca and other Trichoptera. Partial regression analysis indicated that the combination of underlying environment and human pressure accounted for 77–89% of the variation in Ephemeroptera, Trichoptera and Plecoptera taxa richness, %Diptera and %Mollusca, with human pressure explaining more variance than underlying environment for %Mollusca. Analysis of replicate deepwater and littoral samples from moderate gradient sites at the upper and lower ends of the pressure gradient indicated that total Trichoptera and Diptera richness and %Diptera responded to land use differences in these boatable river catchments. Responses to human pressure were substrate specific with the combination of littoral and deepwater substrates providing the most consistent response and yielding the highest number of taxa. These results indicate that multiple substrate sampling is required to document the biodiversity and condition of boatable river macroinvertebrate communities and that spatial variation in the underlying natural environment needs to be accounted for when interpreting pressure–response relationships. Copyright © 2012 John Wiley & Sons, Ltd.

Summary 1. Modification of natural landscapes and land-use intensification are global phenomena that can result in a range of differing pressures on lotic ecosystems. We analysed national-scale databases to quantify the relationship between three land uses (indigenous vegetation, urbanisation and agriculture) and indicators of stream ecological integrity. Boosted regression tree modelling was used to test the response of 14 indicators belonging to four groups – water quality (at 578 sites), benthic invertebrates (at 2666 sites), fish (at 6858 sites) and ecosystem processes (at 156 sites). Our aims were to characterise the ecological response curves of selected functional and structural metrics in relation to three land uses, examine the environmental moderators of these relationships and quantify the relative utility of metrics as indicators of stream ecological integrity. 2. The strongest indicators of land-use effects were nitrate + nitrite, delta-15 nitrogen value (δ15N) of primary consumers and the Macroinvertebrate Community Index (a biotic index of organic pollution), while the weakest overall indicators were gross primary productivity, benthic invertebrate richness and fish richness. All indicators declined in response to removal of indigenous vegetation and urbanisation, while variable responses to agricultural intensity were observed for some indicators. 3. The response curves for several indicators suggested distinct thresholds in response to urbanisation and agriculture, specifically at 10% impervious cover and at 0.1 g m−3 nitrogen concentration, respectively. 4. Water quality and ecosystem process indicators were influenced by a combination of temperature, slope and flow variables, whereas for macroinvertebrate indicators, catchment rainfall, segment slope and temperature were significant environmental predictor variables. Downstream variables (e.g. distance to the coast) were significant in explaining residual variation in fish indicators, not surprisingly given the preponderance of diadromous fish species in New Zealand waterways. The inclusion of continuous environmental variables used to develop a stream typology improved model performance more than the inclusion of stream type alone. 5. Our results reaffirm the importance of accounting for underlying spatial variation in the environment when quantifying relationships between land use and the ecological integrity of streams. Of distinctive interest, however, were the contrasting and complementary responses of different indicators of stream integrity to land use, suggesting that multiple indicators are required to identify land-use impact thresholds, develop environmental standards and assign ecological scores for reporting purposes.

We develop a high-resolution conservation prioritization analysis for New Zealand's rivers and streams that simultaneously consider both the present state (representation) of ecosystems, and the prioritization of management actions designed to mitigate ongoing human impacts on their expected future state (retention). As input we used information about the geographic distributions of river ecosystem groups and their compositional similarity, species richness, present condition as compared to their estimated pristine state, and upstream and downstream connectivity. Candidate management actions included riparian planting, establishment of wetlands on tile-drain outflows, and use of riparian buffer strips in plantation forests. The analysis, carried out at a 1-ha resolution for a study area of 22,000 km2 in Southland, New Zealand, demonstrates a credible range of options for management intervention, particularly in lowland streams under serious threat from agricultural intensification. The proposed analysis can be replicated elsewhere for terrestrial, freshwater, or marine systems using publicly available software.

SUMMARY 1. Previous attempts to identify nationally important wetlands for biodiversity in New Zealand were based on expert panel opinions because quantitative approaches were hampered by a lack of data. We apply principles of systematic conservation planning to remote sensing data within a geographical information system (GIS) to identify nationally important palustrine and inland saline wetlands. 2. A catchment-based classification was used to divide New Zealand into 29 biogeographic units. To meet representation goals, all wetland classes need to be protected within each unit. 3. We mapped current and historic wetlands down to a minimum size of 0.5 ha. Over 7000 current wetlands were mapped using standardised satellite imagery and a collection of point or polygon data. Historical extent was estimated from soil information refined using a digital elevation model. The current extent of wetlands is 10% of the historic extent, which is consistent with previous estimates. 4. A classification was produced using fuzzy expert rules within a GIS to identify seven wetland classes: bog, fen, swamp, marsh, pakihi ⁄ gumland, seepage and inland saline. Swamps and pakihi ⁄ gumland are the most common, but the former has sustained the greatest reduction in area with only 6% of its historical extent remaining. A preliminary field assessment of classification accuracy in the Otago region found only 60% agreement, mainly because of the misclassification of marshes into swamps. 5. Wetland condition was estimated using six measures of human disturbance (natural cover, human-made impervious cover, introduced fish, woody weeds, artificial drainage and nitrate leaching risk) applied at three spatial scales: the wetland’s catchment, a 30-m buffer around the wetland and the wetland itself. Measures were transformed and combined into a single condition index. More than 60% of remaining wetlands had condition indices

1. Classifications that group rivers and streams with similar ecological characteristics are used increasingly to underpin conservation and resource management planning. Uses include identifying systems that may respond similarly to human activities or management actions, setting guidelines and standards to manage human impacts, interpreting data from inventory (survey) and monitoring, and identifying priority sites for conservation management.2. Traditional approaches to river classification have been based mostly on delineating landscape units (ecoregions), often by grouping adjacent catchments having similar ecological character. However, use of this approach can be complicated by marked local heterogeneity of river systems. Instead, classifications may be more ecologically relevant if individual river or stream segments having similar environmental conditions are grouped together, independent of their geographic locations. The latter approach also allows the use of more rigorous classification procedures, including newly emerging techniques that optimise the ability of a classification to discriminate patterns in parallel sets of biological data.3. Here, we explore the use of one of these newer techniques, generalised dissimilarity modelling (GDM), an extension of generalised regression techniques, that defines an optimal set of transformations of candidate environmental predictors to maximise explanation of species turnover in site-based biological data.4. Using two biological data sets describing the distributions of freshwater fish and macroinvertebrates and a candidate set of functionally relevant environmental variables, we used GDM to identify the variables, weightings and transformations that best explain biological dissimilarities across sites. We then used these as input to a multivariate classification of 567000 river and stream segments throughout New Zealand. Weightings and transformations of these variables were also specified from the GDM analysis. The matrix of transformed environmental predictors was classified in a two-stage process, using non-hierarchical mediod clustering to define an initial set of 400 groups, with relationships between these groups then defined using hierarchical clustering.5. The resulting classification better discriminates sites with similar biological character than previous classifications, particularly at higher levels of classification detail. Key factors contributing to this success include the use of detailed, segment-specific environmental variables, coherently accounting for the longitudinal connectivity inherent in rivers including its implications for the construction of biologically relevant predictors, and the use of a modelling technique (GDM) designed to specifically analyse biological turnover and its relationships with environment. © 2010 Blackwell Publishing Ltd.

SUMMARY 1. Broad-scale assessment of stream health is often based on correlative relationships between catchment land-use categories and measurements of stream biota or water chemistry. Few studies have attempted to characterise the response curves that describe how measures of ecosystem function change along gradients of catchment land use, or explored how these responses vary at broad spatial scales. 2. In autumn 2008, we conducted a survey of 84 streams in three bioregions of New Zealand to assess the sensitivity of functional indicators to three land-use gradients: percentage of native vegetation cover, percentage of impervious cover (IC) and predicted nitrogen (N) concentration. We examined these relationships using general linear models and boosted regression trees to explore monotonic, non-monotonic and potential threshold components of the response curves. 3. When viewing the responses to individual land-use gradients, four of five functional indicators were positively correlated with the removal of native vegetation cover and N. In general, weaker and less responsive models were observed for the IC gradient. An analysis of the response to multiple stressors showed d 15 N of primary consumers and gross primary productivity (GPP) to be the most responsive functional indicators to land-use gradients. The multivariate models identified thresholds for change in the relationship between the functional indicators and all three land-use gradients. Apparent thresholds were

We applied a multivariate statistical modelling technique called boosted regression trees to derive relationships between environmental conditions and the distribution of the adult stage of the cyclopoid copepod Oithona similis in the Southern Ocean. Nearly 20 000 samples from the Southern Ocean Continuous Plankton Recorder survey (87% from East Antarctica) were used to model the probability of detection (presence) and relative abundance of adults of this zooplankton species in surface waters. We demonstrate that it is possible to obtain reasonable models for both the presence (area under the Receiver Operating Characteristic curve of 0.77) and relative abundance (28–35% variance explained) of adult O. similis between November and March in much of the Southern Ocean. No investigation was possible where the environmental characteristics were not well represented by the SO-CPR dataset, namely, the Argentine shelf, Weddell Sea, and the frontal region north of the Amundsen Sea, or under sea-ice. Our analyses support the hypothesis that adult O. similis abundance is related to environmental conditions in a broadly similar way throughout the Southern Ocean. Compared to a compilation of net-haul data from the literature, the abundance model explained 34% of the variance in surface concentrations of adult stages of this species, and 23–59% of the variance in depth-integrated abundance of copepodite and adult stages combined. The models show higher occurrence and elevated abundances in a broad circumpolar band between the Antarctic Polar Front and the southern boundary of the Antarctic Circumpolar Current (approximately 54–64°S). Evidence of diel vertical migration by adults of this species north of 65°S was found, with surface abundances 20% higher at night than during the day. There was no evidence of diel migration south of 65°S. Five potential “hotspots” of adult O. similis were identified: in the southern Scotia Sea, two areas off east Antarctica, in the frontal zone north of the Amundsen Sea, and a small area in the outer Bellingshausen Sea. We recommend that a database of all available net-haul data on Oithona similis in the Southern Ocean be created to facilitate further investigations on the circumpolar distribution of this species.

We demonstrate a novel method for spatial conservation prioritization at a community-level that takes account of: (i) an environmentally-based classification of the landscape into community classes; (ii) similarities between community classes to allow for community complementarity-based selection; (iii) variation in species richness; (iv) variation in human impacts on ecological integrity; and (v) requirements for maintenance of upstream–downstream connectivity in riverine systems. While this technique has generic application, we demonstrate its application using a biologically-trained environmental classification of New Zealand’s river network. Our analysis produces a priority ranking of planning units (here 4th order catchments or sub-catchments) and performance estimates in terms of expected biodiversity returns given varying degrees of geographic protection. Accounting for community similarity ensures high protection for distinct habitat classes with low similarity to other classes; our results indicate a 28% loss in conservation efficiency of the highest-ranked 10% of the landscape if it is ignored. Accounting for human pressures and connectivity also had clear influences on spatial priority rankings, indicating the need to consider these factors in the conservation planning process.

Aim The highly adaptable estuarine crab (Carcinus maenas) has successfully invaded five temperate geographic regions outside of its native Europe. Here, we determine which environmental factors predict the current distribution of C. maenas and what the potential geographic range of this species might be. We also investigated whether the invasion potential of C. maenas differs with respect to the origin of a native subpopulation. Location Models were developed using global observation records of C. maenas. Methods Boosted regression trees were used to model observations from the (1) native, (2) invasive, (3) southern European, (4) northern European and (5) the combined native and invasive geographic ranges of C. maenas. Results Most established invasions were predicted mainly based on temperature. Interestingly, the environment encountered by established invasions failed to predict the majority of northern European populations; suggesting that invasion potential may differ between distinct native populations. Supporting this suggestion, a model of northern European populations, distinguished from southern European populations based on genetic structure, only predicted established invasions south of Nova Scotia. By contrast, a model of southern European populations predicted most established invasions. Main conclusions These results suggest that invasion potential depends on the European origin of an invasive population and that most invasions have arisen from southern Europe. Finally, a model based on combined native and invasive ranges of C. maenas identified potential geographic range extension along many currently invaded coastlines and the potential invasion of countries like Chile, China, Russia, Namibia and New Zealand.

Species distribution models (SDMs) are numerical tools that combine observations of species occurrence or abundance with environmental estimates. They are used to gain ecological and evolutionary insights and to predict distributions across landscapes, sometimes requiring extrapolation in space and time. SDMs are now widely used across terrestrial, freshwater, and marine realms. Differences in methods between disciplines reflect both differences in species mobility and in “established use.” Model realism and robustness is influenced by selection of relevant predictors and modeling method, consideration of scale, how the interplay between environmental and geographic factors is handled, and the extent of extrapolation. Current linkages between SDM practice and ecological theory are often weak, hindering progress. Remaining challenges include: improvement of methods for modeling presence-only data and for model selection and evaluation; accounting for biotic interactions; and assessing model uncertainty.

We used statistical models to predict the distributions of 15 native diadromous fish species across New Zealand's river and stream network, and demonstrate their potential use for guiding the restoration of freshwater ecosystems. Models were fitted to an extensive collection of field samples describing the distributions of individual fish species, coupled with a set of environmental predictors chosen primarily for their functional relevance to diadromous fish species. Models were fitted to observations of species occurrence using boosted regression trees (BRT), an advanced regression technique that combines excellent predictive performance with good description of relationships between species occurrence and individual environmental predictors. Environment‐based predictions of species distributions were then made for all river and stream segments. We explored the use of these predictions to guide the selection of sites where diadromous species are likely to occur, and to set restoration targets, ...

International audience; Numerical clustering has frequently been used to define hierarchically organized ecological regionalizations, but there has been little robust evaluation of their performance (i.e., the degree to which regions discriminate areas with similar ecological character). In this study we investigated the effect of the weighting and treatment of input variables on the performance of regionalizations defined by agglomerative clustering across a range of hierarchical levels. For this purpose, we developed three ecological regionalizations of Switzerland of increasing complexity using agglomerative clustering. Environmental data for our analysis were drawn from a 400 m grid and consisted of estimates of 11 environmental variables for each grid cell describing climate, topography and lithology. Regionalization 1 was defined from the environmental variables which were given equal weights. We used the same variables in Regionalization 2 but weighted and transformed them on the basis of a dissimilarity model that was fitted to land cover composition data derived for a random sample of cells from interpretation of aerial photographs. Regionalization 3 was a further two-stage development of Regionalization 2 where specific classifications, also weighted and transformed using dissimilarity models, were applied to 25 small scale sub-domains' defined by Regionalization 2. Performance was assessed in terms of the discrimination of land cover composition for an independent set of sites using classification strength (CS), which measured the similarity of land cover composition within classes and the dissimilarity between classes. Regionalization 2 performed significantly better than Regionalization 1, but the largest gains in performance, compared to Regionalization 1, occurred at coarse hierarchical levels (i.e., CS did not increase significantly beyond the 25-region level). Regionalization 3 performed better than Regionalization 2 beyond the 25-region level and CS values continued to increase to the 95-region level. The results show that the performance of regionalizations defined by agglomerative clustering are sensitive to variable weighting and transformation. We conclude that large gains in performance can be achieved by training classifications using dissimilarity models. However, these gains are restricted to a narrow range of hierarchical levels because agglomerative clustering is unable to represent the variation in importance of variables at different spatial scales. We suggest that further advances in the numerical definition of hierarchically organized ecological regionalizations will be possible with techniques developed in the field of statistical modeling of the distribution of community composition.

Aim To examine the relationship between diadromy and dispersal ability in New Zealand’s freshwater fish fauna, and how this affects the current environmental and geographic distributions of both diadromous and non-diadromous species. Location New Zealand. Methods Capture data for 15 diadromous and 15 non-diadromous fish species from 13,369 sites throughout New Zealand were analysed to establish features of their geographic ranges. Statistical models were used to determine the main environmental correlates of species’ distributions, and to establish the environmental conditions preferred by each species. Environmental predictors, chosen for their functional relevance, were derived from an extensive GIS database describing New Zealand’s river and stream network. Results In terms of geography, most diadromous species occur in a scattered fashion throughout extensive geographic ranges, and occupy large numbers of catchments of widely varying size. By contrast, most non-diadromous species show relatively high levels of occupancy of smaller geographic ranges, and most are restricted to a few large catchments, particularly in the eastern South Island. In terms of environment, there is marked separation of diadromous from non-diadromous species, with diadromous species generally caught most frequently in low-gradient coastal rivers and streams with warm, maritime climates. With a few notable exceptions, most diadromous species have lower occurrence in river segments that are located above obstacles to upstream migration. Non-diadromous species are usually caught in inland rivers and streams with cool, strongly seasonal climates, typified by a low frequency of high-intensity rainfall events. Main conclusions We interpret the contrasting biogeographies of New Zealand’s diadromous and non-diadromous species as reflecting interaction between their marked differences in dispersal ability and a landscape that is subject to recurrent, often large-scale, natural disturbance. While both groups are likely to be equally susceptible to local, disturbance-driven extinction, the much greater dispersal ability of diadromous species has allowed them to persist over wide geographic ranges. By contrast, the distributions of most non-diadromous species are concentrated in a few large catchments, mostly in regions where less intense natural disturbance regimes are likely to have favoured their survival.

There is strong international agreement on the need for marine protected areas to reverse pervasive human impacts on the oceans’ biodiversity. However, their implementation is often hampered both by legal difficulties in defining reserves in international waters and the patchy nature of data in many offshore waters. We demonstrate the use of recent advances in statistical learning and conservation prioritization to produce MPA scenarios with varying costs and benefits for New Zealand’s Exclusive Economic Zone, based on the analyses of distributions of 96 demersal fish species. MPAs based on our most costeffective scenario would deliver conservation benefits nearly 2.5 times greater than those from equivalent-sized areas recently implemented at the request of fishers, and at lower cost to them. Such results demonstrate the power of quantitative, knowledge-based prioritization approaches, which can be applied at high resolution and at oceanic scales.

SUMMARY 1. Freshwater ecosystems are amongst the most threatened and poorly protected globally. They continue to be degraded through habitat loss, pollution and invading species and conservation measures are urgently needed to halt declining trends in their biodiversity and integrity. 2. During the past decade a suite of decision support tools and computational approaches have been developed for efficient and targeted conservation action in terrestrial or marine ecosystems. These methods may be poorly suited for planning in freshwater systems because connectivity in terrestrial and marine systems is typically modelled in a way unsuitable for rivers, where connectivity has a strong directional component. 3. We modify the conservation prioritization method and software, ZONATION , to account for connectivity in a manner better suited to freshwater ecosystems. Prioritization was performed using subcatchment/catchment-based planning units and connectivity was modified to have directional upstream and downstream components consistent with the ecology of our target species. 4. We demonstrate this modified method for rivers and streams in the southern North Island of New Zealand. Data included predicted occupancy from boosted regression tree models of species distributions for 18 fish species. The study area covered 2.1 million hectares and included 394 first- to fourth order catchment or subcatchment planning units. 5. Realistic modelling of connectivity had a major influence on the areas proposed for conservation. If connectivity was ignored, recommended conservation areas were very fragmented. By contrast, when connectivity was modelled, high priority conservation targets consisted of entire river basins or headwater subcatchments. 6. The proposed method serves as a starting point for the implementation of reserve selection methods in river ecosystems.

Current circumstances - that the majority of species distribution records exist as presence-only data (e.g. from museums and herbaria), and that there is an established need for predictions of species distributions - mean that scientists and conservation managers seek to develop robust methods for using these data. Such methods must, in particular, accommodate the difficulties caused by lack of reliable information about sites where species are absent. Here we test two approaches for overcoming these difficulties, analysing a range of data sets using the technique of multivariate adaptive regression splines (MARS). MARS is closely related to regression techniques such as generalized additive models (GAMs) that are commonly and successfully used in modelling species distributions, but has particular advantages in its analytical speed and the ease of transfer of analysis results to other computational environments such as a Geographic Information System. MARS also has the advantage that it can model multiple responses, meaning that it can combine information from a set of species to determine the dominant environmental drivers of variation in species composition. We use data from 226 species from six regions of the world, and demonstrate the use of MARS for distribution modelling using presence-only data. We test whether (1) the type of data used to represent absence or background and (2) the signal from multiple species affect predictive performance, by evaluating predictions at completely independent sites where genuine presence-absence data were recorded. Models developed with absences inferred from the total set of presence-only sites for a biological group, and using simultaneous analysis of multiple species to inform the choice of predictor variables, performed better than models in which species were analysed singly, or in which pseudo-absences were drawn randomly from the study area. The methods are fast, relatively simple to understand, and useful for situations where data are limited. A tutorial is included.

Multivariate classifications of environmental factors are used as frameworks for conservation management. Although classification performance is likely to be sensitive to choice of input variables, these choices have been subjective in most previous studies. We used the Mantel test on a limited set of sites for which biological data were available to iteratively seek a definition of environmental space (i.e., intersite distances calculated with a set of appropriately transformed and weighted environmental variables) that had maximal correlation with the same sites described in a biological space. The procedure was used to select input variables for a classification of New Zealand's rivers that discriminates variation in fish communities for biodiversity management. The classification performed (i.e., discriminated biological variation) better than classifications with subjectively chosen variables. The inherently linear measures of environmental distance that underlie multivariate environmental classifications mean that they will perform best if they are defined based on variables for which there is a linear variation in the biological community throughout the entire range of the variable. Classification performance will therefore be improved when variables that have nonlinear relationships with biological variation are transformed to make their relationship with biological turnover more linear and when the contributions of environmental factors that have particularly strong relationships with biological variation are increased by weighting. Our results indicate that attention to the manner in which environmental space is defined improves the efficacy of multivariate classification and other techniques in which the environment is used as a surrogate for biological variation.

Quantification of modern pollen rain–vegetation–climate relationships in New Zealand has been complicated by human destruction of at least 75% of the original forest cover since ca. 750 years BP, causing contemporary pollen rain over large areas to bear little resemblance to the pre-human vegetation. We use a pre-deforestation pollen database to circumvent this complication. The relationships between the pre-deforestation pollen assemblages and six climatic variables were explored using principal components analysis and constrained regressions (redundancy analyses). Quantitative estimates of the most significant climate variable (mean annual temperature) were made at seven lowland to montane fossil pollen sites from throughout New Zealand using the modern analogue technique and a transfer function. These showed an initial increase in mean annual temperature after 18 000 cal. yr BP, a cooling at the time of the Antarctic Cold Reversal (centred on 13 500 cal. yr BP) and continuation of warming from ca. 12 000 cal. yr BP across the Younger Dryas chronozone, reaching a Holocene thermal optimum that may have been between 1.5 and 3°C warmer than present and lasted from 9000 to 7000 cal. yr BP depending on the site. Cooling to present-day temperatures was well advanced by 4000–3000 cal. yr BP. Copyright © 2007 John Wiley & Sons, Ltd.

Two statistical modelling techniques, generalized additive models (GAM) and multivariate adaptive regression splines (MARS), were used to analyse relationships between the distributions of 15 freshwater fish species and their environment. GAM and MARS models were fitted individually for each species, and a MARS multiresponse model was fitted in which the distributions of all species were analysed simultaneously. Model performance was evaluated using changes in deviance in the fitted models and the area under the receiver operating characteristic curve (ROC), calculated using a bootstrap assessment procedure that simulates predictive performance for independent data. Results indicate little difference between the performance of GAM and MARS models, even when MARS models included interaction terms between predictor variables. Results from MARS models are much more easily incorporated into other analyses than those from GAM models. The strong performance of a MARS multiresponse model, particularly for species of low prevalence, suggests that it may have distinct advantages for the analysis of large datasets. Its identification of a parsimonious set of environmental correlates of community composition, coupled with its ability to robustly model species distributions in relation to those variables, can be seen as converging strongly with the purposes of traditional ordination techniques.

We analysed relationships between demersal fish species richness, environment and trawl characteristics using an extensive collection of trawl data from the oceans around New Zealand. Analyses were carried out using both generalised additive models and boosted regression trees (sometimes referred to as 'stochastic gradient boosting'). Depth was the single most important envi- ronmental predictor of variation in species richness, with highest richness occurring at depths of 900 to 1000 m, and with a broad plateau of moderately high richness between 400 and 1100 m. Richness was higher both in waters with high surface concentrations of chlorophyll a and in zones of mixing of water bodies of contrasting origins. Local variation in temperature was also important, with lower richness occurring in waters that were cooler than expected given their depth. Variables describing trawl length, trawl speed, and cod-end mesh size made a substantial contribution to analysis out- comes, even though functions fitted for trawl distance and cod-end mesh size were constrained to reflect the known performance of trawl gear. Species richness declined with increasing cod-end mesh size and increasing trawl speed, but increased with increasing trawl distance, reaching a plateau once trawl distances exceed about 3 nautical miles. Boosted regression trees provided a powerful analysis tool, giving substantially superior predictive performance to generalized additive models, despite the fitting of interaction terms in the latter.

Summary 1. Relationships between probabilities of occurrence for fifteen diadromous fish species and environmental variables characterising their habitat in fluvial waters were explored using an extensive collection of distributional data from New Zealand rivers and streams. Environmental predictors were chosen for their likely functional relevance, and included variables describing conditions in the stream segment where sampling occurred, downstream factors affecting the ability of fish to move upriver from the sea, and upstream, catchment-scale factors mostly affecting variation in river flows. 2. Analyses were performed using multivariate adaptive regression splines (MARS), a technique that uses piece-wise linear segments to describe non-linear relationships between species and environmental variables. All species were analysed using an option that allows simultaneous analysis of community data to identify the combination of environmental variables best able to predict the occurrence of the component species. Model discrimination was assessed for each species using the area under the receiver operating characteristic curve (ROC) statistic, calculated using a bootstrap procedure that estimates performance when predictions are made to independent data. 3. Environmental predictors having the strongest overall relationships with probabilities of occurrence included distance from the sea, stream size, summer temperature, and catchment-scale drivers of variation in stream flow. Many species were also sensitive to variation in either the average and/or maximum downstream slope, and riparian shade was an important predictor for some species. 4. Analysis results were imported into a Geographic Information System where they were combined with extensive environmental data, allowing spatially explicit predictions of probabilities of occurrence by species to be made for New Zealand's entire river network. This information will provide a valuable context for future conservation management in New Zealand's rivers and streams.

Increasing use is being made in conservation management of statistical models that couple extensive collections of species and environmental data to make predictions of the geographic distributions of species. While the relationships fitted between a species and its environment are relatively transparent for many of these modeling techniques, others are more ‘black box’ in character, only producing geographic predictions and providing minimal or untraditional summaries of the fitted relationships on which these predictions are based. This in turn prevents robust evaluation of the ecological sensibility of such models, a necessary process if model predictions are to be treated with confidence. Here we propose a new but simple method for visualizing modeled responses that can be implemented with any modeling method, and demonstrate its application using five common methods applied to the prediction of an Australian tree species. This is achieved by insetting an “evaluation strip” into the spatial data layers, which, after predictions have been made, can be clipped out and used for creating plots of the modelled responses. We present findings of the application strip for algorithms GLMs, GAMs, CLIM, DOMAIN and MARS. Evaluation strips can be constructed to investigate either uni-variate responses, or the simultaneous variation in predicted values in relation to two variables. The latter option is particularly useful for evaluating responses in models that allow the fitting of complex interaction terms.

A broad-scale, small-fish survey was carried out in northern New Zealand inshore waters using beach seines. The survey covered 30 estuaries spanning ca 1000 km of coastline and three degrees of latitude. Correspondence analysis and cluster analysis were used to identify assemblages, and Generalized Additive Models (GAMs) were used to model the abundance and occurrence of individual species. We aimed to assess the utility of these models for making predictions. The results were mixed. Descriptive models of fish abundance performed well for four out of 12 species; for most other species, and species richness, the models described the data well but performed poorly to moderately under cross validation. Predictive models of fish abundance usually performed worse than descriptive models, but appeared reasonable for four species. Presence–absence models performed better overall than abundance models: descriptive models showed good performance for all 12 species, and predictive models performed well for eight species. For an independent data set, the models successfully predicted occurrence for five species. Water clarity, salinity and the amount of freshwater inflow were important predictor variables. Despite the limitations of our GAMs, they should be useful for planning intensive process-based research, and for guiding the management of human activities that impinge on coastal marine environments.

Successful biodiversity management, including the selection and subsequent management of protected areas, depends in large measure on classifications showing land areas with similar ecosystem character. In contrast to widely used, qualitative land-classification techniques, we used a numerical classification of explicit spatial layers describing aspects of New Zealand's climate and landforms. We chose input variables for their strong functional links with major physiological processes of trees and high statistical correlations with geographic distributions of individual tree species as determined from previous studies. Higher-level divisions of the resulting classification were dominated by macroclimatic variation associated with change in both latitude and orographic protection provided by New Zealand's main mountain ranges, but variation in landform became more important at finer scales of classification. Classification units showed marked variation in the proportional extent of both indigenous vegetation cover and land set aside for conservation purposes. Indigenous ecosystem remnants of the highest priority for increased protection occurred in warm, lowland domains, particularly in drier environments, where both indigenous cover and protected areas are of minimal geographic extent. Such results underline the considerable potential of an environmental classification to provide a landscape context for systematic conservation management, particularly in environments where the natural ecosystem pattern has been severely modified by human activity.

The utility of explicit spatial predictions for biodiversity assessment is investigated with New Zealand fern flora. Distributions of 43 species were modelled from climatic and landform variables and predicted across New Zealand using generalised additive models (GAM). An original package of functions called generalised regression analysis and spatial prediction (GRASP) was developed to perform the analyses. On average, for the 43 models, the contributions of environmental variables indicate that mean annual temperature is the most important factor at this broad regional scale. Both annual solar radiation and its seasonality had higher correlations than temperature seasonality. Measures of water availability such as ratio of rainfall to potential evapotranspiration, air saturation deficit and soil water deficit presented significant contributions. Lithology was a better predictor than slope and drainage. These results are similar to those obtained from analyses of the distributions of New Zealand tree species and are consistent with the hypothesis that both tree and fern diversity are highest on sites conducive to high productivity. In order to identify hotspots of fern diversity, spatial predictions of individual species were summed up. The resulting map gave a very similar result to the direct prediction of their corresponding richness (number of species by plot out of 43 spp.). As a consequence, and where individual species models were not all available, the number of species within different species assemblages was directly modelled. Predicted richness hotspots of total species (out of 122 spp.), selected species (out of 43 and 21 spp.) and common species (out of 23 spp.) present very similar spatial patterns and are highly correlated. Richness of uncommon species (out of 39 spp.) was also accurately predicted, but presented a different spatial pattern. The number of rare species (out of 60 spp.) was not correctly modelled. Even though the lack of data for rare species clearly limits the application of this approach, fern community composition of more common species can be partially reconstructed from individual species predictions. This case study offers therefore a consistent approach not only for biodiversity hotspots identification, but also for setting targets to biodiversity assessment and restoration programs.

We discuss a paradigm for informed ecosystem management that provides a quantitative and rigorous foundation for informing conservation decisions and sustainable ecosystem management. Information pyramids incorporate conceptual and technological advances in ecosystem depiction and provide a framework for the integration and generalization of raw data into forms that are spatially extensive and at the appropriate level of generalization for a particular use. The basic tenets of the pyramid are: (1) Higher levels of the pyramid are entirely derived from a foundation of underlying data. (2) The process of generalization and integration upward should be objective and explicit. (3) Pyramids for different purposes often overlap, with common data and common methods for integration. (4) All levels of the pyramid should be developed together, including base data, methods and kinds of integration, and algorithms for using the information for planning and decision-making. Information pyramids are a powerful approach to organizing research science, and provide a mechanism by which research, data collection, storage and generalization can be focused on conservation outcomes. Common data and methods lead to increased efficiency, while also allowing for separate disciplines and programs. A case study of an integrated pyramid from New Zealand is discussed, which illustrates the characteristics of information pyramids. Components of this pyramid are discussed that provide examples of integration and generalization at various levels of the pyramid, from base data, to derived data, to spatial predictions and classifications, to a method of integrating this information into conservation decisions.

Competitive interactions between New Zealand's four Nothofagus or southern beech species were analysed using an extensive dataset describing the composition of natural forests, supplemented by environmental estimates describing both climate and landform. Using multiple regression models of progressively increasing complexity, the analysis first accounted for variation in tree abundance attributable to both environment and regional-scale distributional disjunctions of likely historic origin. Intra-generic competition, expressed as variation in tree abundance dependent on the presence or absence of each congener, was then assessed by adding (1) simple terms to assess the magnitude of gross changes in abundance, and (2) interaction terms to assess variation in abundance along the dominant temperature gradient given different competitive contexts. Results indicate the presence of substantial intra-generic interactions, with simple interaction terms giving marginal increases in explained deviance equal to that explained by initial regressions using environment alone. Addition of interaction terms brought about smaller improvements in model fit, but confirm that variation in abundance along the dominant annual temperature gradient is strongly influenced by the competitive context provided by the remaining congeners. Such results are consistent with current understanding of the niche concept, and underline the difficulty inherent in using current species limits to predict likely changes in species distributions consequent on global warming.

New Zealand's four broad-leaved evergreen tree species from the genus Nothofagus all show pronounced distributional disjunctions, independent of environmental factors known to influence tree distributions. Here, we use these disjunctions as the basis for a natural removal experiment to investigate competitive interactions between Nothofagus and a range of other widespread conifer and broad-leaved tree species. We first model the abundance of non-Nothofagus species as a function of environment, using Generalized Additive Models (GAMs) and an extensive data set sampling much of New Zealand's remaining old-growth forests. We then assess the effects of competitive interaction with Nothofagus by adding statistical terms describing (1) Nothofagus abundance, and (2) interactions between Nothofagus abundance and annual temperature, the dominant environmental gradient. Results indicate substantial reductions in the abundance of many species as Nothofagus abundance increases. The magnitude of this reduction varies with position along the dominant environmental gradient; species overlapping most strongly with Nothofagus are generally most sensitive to increases in Nothofagus abundance. In addition, both the shapes of species responses to mean annual temperature and the positions of their optima change as Nothofagus abundance increases. This demonstration of competition using community compositional data has implications both for vegetation theory and for prediction of the likely impacts of global warming on New Zealand's forest pattern.

Reversion of agricultural land to native woody vegetation can sequester carbon (C), influencing regional and global C budgets. We examined whole-ecosystem differences in C and nitrogen (N) storage and distribution, and sapwood - leaf area relationships in a scrubland vegetation chronosequence in New Zealand dominated by manuka (Leptospermum scoparium J.R. et G. Forst) and kanuka (Kunzea ericoides var. ericoides (A. Rich.) J. Thompson). At 25 years, manuka dominated, and vegetation C was 6.5 kg C·m-2. In the 55-year-old stand, stem density was similar for the two species, and vegetation C storage was 15.1 kg C·m-2, similar to the 35-year-old stand (p = 0.9). Foliar biomass comprised 3-5% of vegetation C stock but contained 26%-37% of vegetation N. Root biomass was 10-15% of total and varied little with stand age. The sapwood - leaf area relationship differed significantly for the two species (p < 0.05). Mineral soil C and N (to 0.30 m) did not vary with stand age, but forest floor C and N were highest in the 55-year-old stand (2 kg C·m-2; p < 0.01). Soil and forest floor C/N ratios were significantly higher in the 35-year-old stand (p < 0.04), possibly because of high interspecific competition for N. While the sampling intensity was too limited to allow spatial extrapolation, our results suggest that carbon accumulation in this scrubland is rapid and similar to plantation forests, suggesting that land abandonment could significantly impact New Zealand's C budget.

Correlations between environment and tree alpha-diversity in New Zealand's primary forests were examined using an extensive quantitative dataset (14 540 plots). Generalised additive models were used to examine relationships between species richness and temperature, solar radiation, root-zone moisture deficit, relative humidity, lithology, drainage, and plot size for all trees (112 species), and separately for broadleaved trees (88 species), conifers (17), and the genus Nothofagus (4). Diversity both for all tree species and for broadleaved trees was predicted to be highest on sites with high temperatures, high solar radiation, and high soil and atmospheric moisture, and on sedimentary and basaltic substrates. Highest conifer diversity was predicted on sites with intermediate temperatures, low solar radiation, high root-zone and atmospheric moisture. and rhyolitic and Quaternary substrates, particularly where drainage was impeded. Highest Nothofagus diversity, was predicted for sites combining low temperatures, high solar radiation. high root-zone moisture but low atmospheric moisture, and on granitic substrates. Differences in diversity between the species groups on different lithologies are interpreted as reflecting both the effects of variation in large-scale disturbance histories, and the effects of confounding environmental factors associated with particular substrates. There were also significant interactions between species groups: both broadleaved tree and conifer richness were predicted to be lower on sites where one or more Nothofagus spp. - all of which have marked patchiness in their distribution - are present. Although these results are consistent with the hypothesis that tree diversity is highest on sites conducive to high productivity, history is also indicated as an important determinant of tree diversity in New Zealand.

Past explanations of the large disjunctions in the distribution of New Zealand's four Nothofagus species have emphasized displacement during glacial cycles followed by slow re-occupation of suitable sites, or the effects of plate tectonics coupled with ecological and/or environmental limitations to further spread. In this study the degree of equilibrium between Nothofagus distribution and environment was compared with that of other widespread tree species by statistical analysis. Generalized additive regression models were used to relate species distribution data to estimates of temperature, solar radiation, soil water deficit, atmospheric humidity, lithology and drainage. For each species, the amount of spatial patterning remaining unexplained by environment was assessed by adding a variable describing species presence/absence on adjacent plots. Results indicate that Nothofagus species occur more frequently in environments suboptimal for tree growth, i.e. having various combinations of cool temperatures, low winter solar radiation, high root-zone water deficit, low humidity, and infertile granitic substrates. Despite these demonstrated preferences, they exhibit substantially more spatial clustering which is unexplained by environment, than most other widespread tree species. Predictions formed from regressions using environment alone confirm that several major Nothofagus disjunctions are not explicable in terms of the environmental factors used in this analysis, but more likely reflect the effects of historic displacement coupled with slowness to invade forest dominated by more rapidly dispersing endomycorrhizal species. The technique used in this study for detecting residual spatial autocorrelation after fitting explanatory variables has potentially wide application in other studies where either regression or ordination techniques are used for analysis of compositional data.

The introduced brushtail possum Trichosurus vulpecula Kerr is recognised as the primary agent of defoliation and stand-level dieback in New Zealand broadleaved forests (except Nothofagus). The distribution and magnitude of canopy collapse of forest in 33 500 ha of the southern Ruahine Range, New Zealand was mapped from 1995 polychrome aerial photographs. Relationships between canopy collapse and forest type, altitude, aspect and slope were analysed using generalised additive models. Canopy composition was the strongest factor predicting the extent of collapse and modification: broadleaved-conifer forest is most affected, with 68–87% of the area of six such forest types replaced by scrub-low forest and tree-fernland; and surprisingly, three Nothofagus-dominated types have up to 28% of their former area now in shrubland or tussock grassland. The susceptibility to collapse of Nothofagus forest was positively correlated with distance to non-Nothofagus forest types. Subalpine scrub, which is dominated by species not favoured by possums, has increased in area by 32%, replacing former upper montane forest. Physiographic factors were much less important in multiple regressions. However, in general terms, forests on steeper slopes, in the upper montane-subalpine zone, and on warm westerly and northerly aspects were more susceptible to collapse and modification than elsewhere. It is postulated that possums were the primary agent responsible for collapse of non-Nothofagus broadleaved forest, and that red deer Cervus elaphus and goats Capra hircus were responsible for inhibiting canopy replacement by eliminating regeneration in forest understoreys. Secondary effects, such as outbreaks of defoliating insects and mechanical damage from wind also contributed to canopy collapse and were possibly triggered by possums opening up the canopy. Collapse of Nothofagus canopies possibly results from a breakdown in the recruitment phase of stand turnover, by the smothering effects of deer-induced shrubs and tussock grasses on seedlings and saplings.

A dataset of some 10 000 plots was used to describe the climatic relationships of 33 widespread New Zealand tree species. Estimates of mean annual temperature, temperature seasonality, mean annual solar radiation, and moisture balance were derived from mathematical surfaces fitted to climate station data. Plots were also categorized into five lithological classes and three drainage classes. Generalized additive models were used to examine species/environment relationships. Mean annual temperature and mean annual solar radiation are most strongly correlated with current tree distributions, followed by moisture balance, temperature seasonality, lithology, and drainage. Most broad-leaved tree species other than Nothofagus spp. reach their greatest levels of occurrence in warm, moist environments with high solar radiation. In contrast, Nothofagus spp. generally reach their greatest levels of occurrence in cooler and/or lower insolation environments, and all have lower levels of occurrence on rhyolitic substrates which have resulted from large-scale geomorphic disturbance, mostly over the past few thousand years. Although coniferous species have widely differing climatic optima, many are biased towards lithological classes characterized either by large-scale geomorphic disturbance or harsh edaphic conditions. The relevance of these results to particular synecological questions is briefly discussed. Continuing adjustments in the range of slow-dispersing Nothofagus spp. are strongly suggested, and the climatic suitability of extensive rhyolitic basins in the central North Island, from which these species are largely absent, is confirmed.

International audience; Ecological regionalizations define geographic regions exhibiting relative homogeneity in ecological (i.e., environmental and biotic) characteristics. Multivariate clustering methods have been used to define ecological regions based on subjectively chosen environmental variables. We developed and tested three procedures for defining ecological regions based on spatial modeling of a multivariate target pattern that is represented by compositional dissimilarities between locations (e.g., taxonomic dissimilarities). The procedures use a training dataset' representing the target pattern and models this as a function of environmental variables. The model is then extrapolated to the entire domain of interest. Environmental data for our analysis were drawn from a 400 m grid covering all of Switzerland and consisted of 12 variables describing climate, topography and lithology. Our target patterns comprised land cover composition of each grid cell that was derived from interpretation of aerial photographs. For Regionalization 1 we used conventional cluster analysis of the environmental variables to define 60 hierarchically organized levels comprising from 5 to 300 regions. Regionalization 1 provided a base-case for comparison with the model-based regionalizations. Regionalization 2, 3 and 4 also comprised 60 hierarchically organized levels and were derived by modeling land cover composition for 4000 randomly selected training' cells. Regionalization 2 was based on cluster analysis of environmental variables that were transformed based on a Generalized Dissimilarity Model (GDM). Regionalization 3 and 4 were defined by clustering the training cells based on their land cover composition followed by predictive modeling of the distribution of the land cover clusters using Classification and Regression Tree (CART) and Random Forest (RF) models. Independent test data (i.e. not used to train the models) were used to test the discrimination of land cover composition at all hierarchical levels of the regionalizations using the classification strength (CS) statistic. CS for all the model-based regionalizations was significantly higher than for Regionalization 1. Regionalization 3 and 4 performed significantly better than Regionalization 2 at finer hierarchical levels (many regions) and Regionalization 4 performed significantly better than Regionalization 3 for coarse levels of detail (few regions). Compositional modeling can significantly increase the performance of numerically defined ecological regionalizations. CART and RF-based models appear to produce stronger regionalizations because discriminating variables are able to change at each hierarchic level.

Most methods for modeling species distributions from occurrence records require additional data representing the range of environmental conditions in the modeled region. These data, called background or pseudo-absence data, are usually drawn at random from the entire region, whereas occurrence collection is often spatially biased toward easily accessed areas. Since the spatial bias generally results in environmental bias, the difference between occurrence collection and background sampling may lead to inaccurate models. To correct the estimation, we propose choosing background data with the same bias as occurrence data. We investigate theoretical and practical implications of this approach. Accurate information about spatial bias is usually lacking, so explicit biased sampling of background sites may not be possible. However, it is likely that an entire target group of species observed by similar methods will share similar bias. We therefore explore the use of all occurrences within a target group as biased background data. We compare model performance using target-group background and randomly sampled background on a comprehensive collection of data for 226 species from diverse regions of the world. We find that target-group background improves average performance for all the modeling methods we consider, with the choice of background data having as large an effect on predictive performance as the choice of modeling method. The performance improvement due to target-group background is greatest when there is strong bias in the target-group presence records. Our approach applies to regression-based modeling methods that have been adapted for use with occurrence data, such as generalized linear or additive models and boosted regression trees, and to Maxent, a probability density estimation method. We argue that increased awareness of the implications of spatial bias in surveys, and possible modeling remedies, will substantially improve predictions of species distributions.

International audience; Hydrologic variability is important in sustaining a variety of ecological processes in streams and rivers. Natural flow regime classifications group streams and rivers that are relatively homogeneous with respect to flow variability and have been promoted as a method of defining units for management of river flows. Although there has been considerable interest in classifying natural flow regimes, there has been less emphasis given to developing accurate methods of extrapolating these classifications to locations without flow data. We developed a method of mapping flow regime classes using boosted regression trees (BRT) that automatically fits non-linear functions and interactions between explanatory variables of flow regimes, both of which can be expected when comparing responses between complex systems such as watersheds. A natural flow regimes classification of continental France was developed from cluster analysis of 157 hydrological indices derived from 763 gauging stations representing unmodified flows. BRT models were used to predict the likelihood of gauging stations belonging to each class based on the watershed characteristics. These models were used to extrapolate the natural flow regime classification to all segments of a national river network. The performance of the BRT models were compared with other methods of assigning locations to flow regime classes, including the use of geographically contiguous regions, linear discriminant analysis (LDA) and classification and regression trees (CART). The fitted misclassification rate (associated with model fits) for assignment based on the BRT models was 13% whereas the fitted misclassification rates for geographically contiguous regions, LDA and CART were 52%, 44% and 39%, respectively. A predictive misclassification rate (calculated for new cases) was estimated for assignments based on the BRT, LDA and CART models using cross validation analysis. For assignment based on the BRT models, the mean predictive misclassification rate was 37% whereas for LDA and CART it was 45% and 64%, respectively. Our method of mapping flow regimes could increase the confidence in decisions associated with setting environmental flows and the ability to undertake broad-scale ecohydrological research.

In ecological modeling of the habitat of a species, it can be prohibitively expensive to determine species absence. Presence-only data consist of a sample of locations with observed presences and a separate group of locations sampled from the full landscape, with unknown presences. We propose an expectation–maximization algorithm to estimate the underlying presence–absence logistic model for presence-only data. This algorithm can be used with any off-the-shelf logistic model. For models with stepwise fitting procedures, such as boosted trees, the fitting process can be accelerated by interleaving expectation steps within the procedure. Preliminary analyses based on sampling from presence–absence records of fish in New Zealand rivers illustrate that this new procedure can reduce both deviance and the shrinkage of marginal effect estimates that occur in the naive model often used in practice. Finally, it is shown that the population prevalence of a species is only identifiable when there is some unrealistic constraint on the structure of the logistic model. In practice, it is strongly recommended that an estimate of population prevalence be provided.

Summary 1. Ecologists use statistical models for both explanation and prediction, and need techniques that are flexible enough to express typical features of their data, such as nonlinearities and interactions. 2. This study provides a working guide to boosted regression trees (BRT), an ensemble method for fitting statistical models that differs fundamentally from conventional techniques that aim to fit a single parsimonious model. Boosted regression trees combine the strengths of two algorithms: regression trees (models that relate a response to their predictors by recursive binary splits) and boosting (an adaptive method for combining many simple models to give improved predictive performance). The final BRT model can be understood as an additive regression model in which individual terms are simple trees, fitted in a forward, stagewise fashion. 3. Boosted regression trees incorporate important advantages of tree-based methods, handling different types of predictor variables and accommodating missing data. They have no need for prior data transformation or elimination of outliers, can fit complex nonlinear relationships, and automatically handle interaction effects between predictors. Fitting multiple trees in BRT overcomes the biggest drawback of single tree models: their relatively poor predictive performance. Although BRT models are complex, they can be summarized in ways that give powerful ecological insight, and their predictive performance is superior to most traditional modelling methods. 4. The unique features of BRT raise a number of practical issues in model fitting. We demonstrate the practicalities and advantages of using BRT through a distributional analysis of the short-finned eel ( Anguilla australis Richardson), a native freshwater fish of New Zealand. We use a data set of over 13 000 sites to illustrate effects of several settings, and then fit and interpret a model using a subset of the data. We provide code and a tutorial to enable the wider use of BRT by ecologists.