Combining physiological threshold knowledge to species distribution models is key to improving forecasts of the future niche for macroalgae
Abstract
Species distribution models (SDM) are a useful tool for predicting species range shifts in response to global warming. However, they do not explore the mechanisms underlying biological processes, making it difficult to predict shifts outside the environmental gradient where the model was trained. In this study, we combine correlative SDMs and knowledge on physiological limits to provide more robust predictions. The thermal thresholds obtained in growth and survival experiments were used as proxies of the fundamental niches of two foundational marine macrophytes. The geographic projections of these species’ distributions obtained using these thresholds and existing SDMs were similar in areas where the species are either absent-rare or frequent and where their potential and realized niches match, reaching consensus predictions. The cold-temperate foundational seaweed Himanthalia elongata was predicted to become extinct at its southern limit in northern Spain in response to global warming, whereas the occupancy of southern-lusitanic Bifurcaria bifurcata was expected to increase. Combined approaches such as this one may also highlight geographic areas where models disagree potentially due to biotic factors. Physiological thresholds alone tended to over-predict species prevalence, as they cannot identify absences in climatic conditions within the species’ range of physiological tolerance or at the optima. Although SDMs tended to have higher sensitivity than threshold models, they may include regressions that do not reflect causal mechanisms, constraining their predictive power. We present a simple example of how combining correlative and mechanistic knowledge provides a rapid way to gain insight into a species’ niche resulting in consistent predictions and highlighting potential sources of uncertainty in forecasted responses to climate change.
Introduction
Predicting species distribution shifts accurately and rapidly is fundamental to propose adequate conservation strategies for species affected by climate change (Austin, 2002; Guisan et al., 2006). Ambient temperatures and the frequency of heat waves are increasing, driving range shifts across virtually all ecosystems on Earth (e.g., Parmesan, 2006). Common predictive modelling approaches, namely species distribution models (SDMs), are typically based on correlations between distributional and environmental data (e.g., Elith et al., 2010). SDMs have been used to investigate macroecological patterns and guide management policies, such as the selection of protected areas and the prediction of alien species invasions (e.g., Kearney et al., 2008; Recio & Virgós, 2010). However, they do not explore the mechanisms underlying the biological causes of species distributions (Buckley et al., 2010; Elith et al., 2010; Kearney et al., 2010a), making it difficult to apply results outside the environmental gradient where the model was trained. This potentially limits the accuracy of predictions when applied to a species’ response in a particular geographic area and climatic scenario (Araújo & Luoto, 2007; Diamond et al., 2012).
In the last decade, some predictive models have been based on the biological mechanisms underlying the distribution of species, in particular on organisms’ physiological tolerance to temperature (reviewed in Kearney & Porter, 2009; Kearney et al., 2010b). These include the biophysical models of amphibian, reptile and marsupial of Australia (e.g., Kearney et al., 2008), intertidal mollusks and cirripeda (e.g., Helmuth, 1998; Wethey et al., 2011), one butterfly and one lizard (Buckley et al., 2010), and the Japanese serow (Natori & Porter, 2007), among others. These typically take into account the different mechanisms determining the organism's body temperature, as for example heat gains in response to air temperature or to solar radiation (e.g., Natori & Porter, 2007). The different mechanisms are explicitly incorporate in the model in the form of different mathematical equations. Then project the species’ geographic distribution by applying thermal tolerance thresholds for survival or other vital rates. Although promising for forecasting species distributions, this kind of models is uncommon due to the substantial information required for the parameterization of the different equations (Kearney & Porter, 2009; Buckley et al., 2010, 2011) which is seldom available.
The combination of correlative and mechanistic modelling may capture responses not detected by each approach separately and thus recently, some studies have opened the debate on how hybrid models can be constructed (see Buckley et al., 2011 and references therein). The potential applicability of physiological limits to increase the robustness of SDM projections has been suggested (Morin & Lechowicz, 2008; Nogués-Bravo, 2009) but seldom done. For example, thresholds can provide information on unoccupied areas of potential favourable climatic conditions not represented in SDMs or range underfilling, an important matter of discussion (e.g., Petitpierre et al., 2012; Sunday et al., 2012). Physiological thresholds are known for a number of taxa and can be described by ecophysiologists using experiments simulating stress conditions. This approach falls within the family of statistical methods, testing if patterns that occur with an associated probability can be attributed to the predictor variables. Therefore do not incorporate the mechanisms in the form of different equations thus typically resulting in faster and simpler approaches than biophysical modelling. Ecophysiological experiments are parsimonious and may identify tipping points in environmental conditions with small error. Thus, it is surprising that approaches aimed to investigate the physiological mechanisms operating at distributional edges are not commonly used in studies attempting to forecast shifts under climate change (but see Hijmans & Graham, 2006; Austin et al., 2009; Jones et al., 2009, 2010; Pearson et al., 2009; Buckley et al., 2011; Diamond et al., 2012).
In this study, we provide an example of a simple way to combine knowledge of physiological thresholds with SDMs. Our integrative approach aims to validate (or otherwise) the distribution shift predictions obtained by the SDMs developed in a previous study (Martínez et al., 2012a). Growth and survival responses to a simulated temperature gradient were used as proxies of the species’ fundamental niches (as defined in Soberón & Nakamura, 2009). We did not develop an algorithm relating body temperature and environmental conditions as in biophysical models. Instead, we performed novel manipulative experiments to accurately determine species functional responses to stress gradients by using designs and statistical analyses like those performed in ecophysiological studies (as in Hijmans & Graham, 2006; Austin et al., 2009; Jones et al., 2009). Using thermal thresholds for growth and survival, we reclassified the observed presences and absences of two target species recorded in the same field survey carried out by our research group to construct the SDMs (see Martínez et al., 2012a). We then evaluated the accuracy of the classification using the two methods. Finally, we compared the projections under future climatic scenarios obtained by the thermal thresholds and SDMs.
The model species were two intertidal foundation macroalgae with different biogeographic affinities: Himanthalia elongata is a cold-temperate fucoid abundant in the northwestern corner of the Iberian Peninsula but absent from eastern Cantabrian shores (northern Spain) and towards the centre and south of Portugal; Bifurcaria bifurcata is a southern-lusitanic species distributed along the entire Cantabrian shoreline, the Atlantic coast of Galicia and Portugal except on the southern coast, reappearing in Morocco (Lüning, 1990; Lima et al., 2007). Canopy-forming algae provide food and resources and are the foundation species of rocky shores in coastal marine systems (Dayton, 1975). As intertidal organisms, macroalgae live at the interface of land and sea, enduring atmospheric and oceanic stress conditions frequently close to their physiological tolerance thresholds (see Wethey et al., 2011 and references therein). Thus, they are potentially sensitive indicators of climate-driven changes in the environment. Indeed, a contraction of H. elongata's distribution limit 130 km to the west has been reported in the Cantabrian Sea in the last few years, as part of the decline of several foundational macroalgae in northern Spain (Fernández, 2011; Díez et al., 2012; Lamela-Silvarrey et al., 2012; Duarte et al., 2013). Anecdotal information suggests a higher abundance of B. bifurcata (Pers. Obs). These trends are congruent with the predictions of SDMs under two warming scenarios developed in the above-mentioned study (Martínez et al., 2012a), which forecasted the local extinction of H. elongata in the Cantabrian Sea (northern Spain), and an increase in the occurrence of B. bifurcata. Here, we are attempting to better understand and forecast the fate of the target species, as they play a crucial role in European coastal ecosystems. Furthermore, we aim to provide guidance on how to combine knowledge of physiological thresholds and SDMs for organisms threatened by climate change.
Materials and methods
Ecophysiological experiment set-up
We collected 240 individuals of Himanthalia elongata (L.) S.F. Gray with a mean (±SE) weight of 3.7 ± 1.1 g fresh weight (FW) and 240 of Bifurcaria bifurcata R. Ross weighing 3.0 ± 0.04 g FW during low tide on the 22nd and 24th March 2011 at Redondela (Ria de Vigo, Spain, 42.285973°N, 8.656306°W) and Moledo (northern Portugal, 41.845790°N, 8.867709°W), respectively. Material was placed in a cool box and taken to the laboratory within 2 h of collection. Fronds (480 in total) were kept in a 300 l filtered seawater tank at 15 °C (ambient ocean temperature) until the experiment.
On the 24th March, the fronds were placed in 24 outdoor shaded tanks of 20 l filled with filtered seawater. The tanks were set at the following 12 temperatures: 8.3 ± 0.03, 10.0 ± 0.03, 12.3 ± 0.03, 13.9 ± 0.02, 16.0 ± 0.04, 17.9 ± 0.01, 19.9 ± 0.02, 21.8 ± 0.01, 24.7 ± 0.06, 26.0 ± 0.02, 27.9 ± 0.01 and 29.8 ± 0.02 (mean temperature for the entire experiment in °C ±SE, n = 157). Two tanks were set at each mean seawater temperature that was kept constant using seawater chillers and titanium heaters, simultaneously regulated by digital controllers and temperature probes (Aqua Medic ® AT Control System controllers, GmbH, Bissendorf, Germany) with a programmed error of 0.2 °C. Tanks were submersed in large temperature-controlled water baths to ameliorate changes in ambient temperature. Each experimental tank contained a plastic frame with nylon lines, on which ten replicates of each species were hung (20 per temperature) using plastic clips to fix them to the lines. To avoid nutrient limitation, the seawater was enriched every 2 days by adding inorganic N (NaNO3) and P (NaH3PO4) to a final concentration of 50 μm N and 5 μm P, respectively. The tanks were refilled with freshwater every few days to compensate for water evaporation and to maintain ambient salinity levels. Water was agitated using air pumps. The experiment ran for 6 weeks.
Growth and survival
Fronds were weighed to the nearest 0.1 g and the number of surviving individuals was noted, immediately before being transferred to the tanks and then weekly thereafter. Dead individuals were identified by significant tissue necrosis. H. elongata has a two-stage morphology: a small button-like frond or reduced growth is first produced, from which large strap-like fronds bearing reproductive structures are developed. In this study, growth refers to the growth of straps. B. bifurcata fronds show a dichotomous cylindrical morphology throughout their life-cycle. Fronds were vegetative throughout the experiment.
Growth, tissue loss (degrowth) and survival were plotted against the 12 experimental temperatures to determine physiological thresholds (hereafter PTs). Positive growth data (biomass increases) were averaged and mean values were fitted to second-order polynomials to illustrate major trends. Once the growth response reached a stationary-constant state, the temperature with the lowest averaged growth value i.e., close to, but above zero, was identified as the best proxy for the growth threshold of each species. We used logistic regression to explore the relationship between survival and temperature, since it follows a binomial distribution (dead/alive). From this regression, we calculated the temperature threshold associated with a survival probability of 0.5 and approximated this value to the nearest 0.5 (°C) as a proxy of the parameter LC50, indicative of tipping conditions for population survival (e.g., Jones et al., 2009).
Classification maps and metrics
We applied the PTs to the records of the two species that were surveyed from 2004 to 2006 to construct the SDMs (see Martínez et al., 2012a) and classified them as presences or absences in the 198 grid cells (1-km2). A record was classified using PTs as an absence when average August sea surface temperature (SST Aug) for that cell was higher than the PT. If SST Aug was lower than the PT, the record was considered a presence. SST Aug was obtained from images taken by the Advanced Very High Resolution Radiometer (AVHRR) onboard the National Oceanic and Atmospheric Administration (NOAA)-series satellites and processed by the Spanish Ministry of Defense (http://www.inta.es/) for the 2002–2004 period (same data as in Martínez et al., 2012a).
The predictive equations obtained by SDMs using the environmental variables as inputs were also applied to the same cells, i.e., to those with biological data available. We used the favourability function (Real et al., 2006) where favourability values greater than 0.5 indicate the predicted presence of the species in a particular cell (see Martínez et al., 2012a; for details). These predictive equations linking the probability of H. elongata and B. bifurcata occurrence (PH and PB, respectively) to the environmental predictors were: PH = exp(19.40 − 0.61*SST Aug − 0.46*Max Aug + 1.50*Rock)/1 + exp(19.40 − 0.61*SST Aug − 0.46*Max Aug + 1.50*Rock) and PB = exp(−11.12 + 1.16*Rock + 0.58*SST Aug)/1 + exp(−11.12 + 1.16*Rock + 0.58 SST Aug), where SST Aug is average August SST (see above), Max Aug is average maximum air temperature in August, and Rock the presence of suitable rocky substrate. Max Aug was obtained from the Digital Climatic Atlas of the Iberian Peninsula that compiles data measured by meteorological stations from 1951 to 1999 (Ninyerola et al., 2005; http://opengis.uab.es/wms/iberia/espanol/es_model.htm). Rock in each coastal cell (1: presence of sufficient rocky substrate for fixation, 0: absence) was determined through personal observations during the field survey, information obtained from the 1 : 50 000 Spanish Military Cartography and aerial photographs (see Martínez et al., 2012a for further details on the environmental data).
To construct the SDMs in Martínez et al. (2012a), we used GLMs using the logit link function and allowing from linear, second and third terms, and performed automatic stepwise variable selections based on the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). When each variable was added, we also tested for significance in deviance reduction using an F-test. A hierarchical partitioning analysis was performed to test for the predictive variables with the greatest influence in species occurrence. In addition, we fitted data to GAMs with four smoothing degrees of freedom, though coincident results were omitted to avoid redundancies (see Martínez et al., 2012a).
After classifying the observed presences and absences using PTs and SDMs, we plotted all the points on a map to detect the matches and mismatches in the outputs from the two approaches. We also used these data to calculate prevalence (ratio of observed presences to all records), specificity (ratio of correctly classified absences), sensitivity (ratio of correctly classified presences) and overall accuracy (ratio of correctly classified records) (Fielding & Bell, 1997; Buckley et al., 2010) for particular stretches of coast which showed different model performance.
When applying the PTs, we classified records in sites which lack rocky substratum as absences (Rock category = 0), as this is a required habitat of the target species irrespective of climatic conditions. Records in sandy and muddy sites were correctly classified by SDMs (mostly as absences) and thus did not contribute to explaining differences between methods. Given this, we excluded these records when calculating specificity, sensitivity and accuracy to avoid inflated estimates (that may thus differ from those in Martínez et al., 2012a).
Projections
To predict the distribution of the species under future warming scenarios, we applied the PTs and the equations from SDMs to scenarios of projected temperatures until the year in which the species occupied or disappeared from particular stretches of coastline (from present-day conditions to 2040). This was done using 943 cells (all but those with unsuitable substrate), corresponding to 927 km along the north-western corner of the Iberian Peninsula.
To project future ocean temperatures, we used the same scenario of Martínez et al. (2012a), which is based on regional anomalies reported in the last few decades for northern Spain. We considered an average summer SST increase of 0.53 °C per decade for the Cantabrian Sea (Llope et al., 2006) and 0.50 °C per decade for the Atlantic coast (Cabanas et al., 2003). In the SDM for H. elongata, which includes a term for atmospheric warming, we also applied the corresponding anomaly using the IPCC A1 greenhouse emission scenario in Abanades et al. (2007). For each grid cell, we added the corresponding water thermal anomaly for the considered time period to the SST Aug values of that cell, and the same was done with atmospheric anomalies in the case of H. elongata (as in Martínez et al., 2012a). As projected temperatures were judged to underestimate thermal anomalies compared to field records, we aimed to predict general trends rather than precise patterns. We plotted the outputs on maps for the years when extinctions or colonizations were projected in stretches of coast.
Results
Functional responses to water temperature
As expected, we found that H. elongata was less tolerant to warm conditions than B. bifurcata (compare Figs 1 and 2). The PT for H. elongata was 18 °C with the growth of surviving individuals (37–32%) decreasing during the second week at this temperature. From the second week onwards, the functional responses of growth at different time intervals were similar, suggesting a wide range of optimal temperatures and a marked growth decline at 18 °C for this species (Fig. 1). This value was also the threshold for the 0.5 survival probability i.e., the LC50, from the logistic regression equation linking survival probability to temperature for the same time period (Fig. 3). In the case of B. bifurcata, approximately half of the plants (60–40%) survived at 24.7 °C, but mean growth values of survivors were close to zero in the third week and thereafter (Fig. 2). The LC50 calculated from the logistic regression equation pooling the survival data from the last 3 weeks suggested a slightly lower PT at 23.5 °C (Fig. 3). As the latter temperature was not simulated in the experiments, the PT suggested by growth was preferred over that suggested by mortality in this case, although this selection did not affect projections. At the beginning of the experiment, most H. elongata (95%) and B. bifurcata (95–79%) individuals grew and survived at approximately 5–3 °C above their PTs. This initial transitory phase of tolerance to thermal stress ranged between 1 and 2 weeks in H. elongata and 2 and 3 weeks in B. bifurcata at the temperatures tested (Figs 1-3).
The thermal threshold obtained for H. elongata in this study was similar to that suggested by the relationship between the percentage of presences observed in the field and the averaged maximum August SST (SST Aug) found in Martínez et al. (2012a). The species was absent from locations with an average SST Aug higher than 19 °C (see Martínez et al., 2012a; Fig. 4). Regarding B. bifurcata, our results show that it does not encounter lethal conditions in the studied area, since the highest average SST Aug was 20.4 °C (Fig. 4), which is far below the thermal threshold for this species. However, this species was absent or less frequent in sites with average values between 14.3 and 17 °C, particularly in the cold area around Cape Finisterre to Ría de Arosa, even though these temperatures are within its optimal range for growth and survival (Figs 2 and 3, Martínez et al., 2012a).
Classification accuracy applying PTs and SDMs
We differentiated three geographical stretches of coast according to the accuracy of models (correct and incorrect classification of predicted values) in the case of H. elongata. In the area with reduced presence of northern Spain with warm ocean conditions (from Llanes in Asturias to Nois in Galicia, 9 presences and 32 absences in rocky sites, prevalence = 0.22), absences were correctly classified by the two approaches (Fig. 4a, outer coastline). Thus, specificity was high (0.88) using both PT and SDM. The few presences reported on the field surveys of 2004–2006 at this contracting edge were mostly classified as absences (Fig. 4a, inner coastline), and sensitivity was low using SDM (0.11) but not so low using PT (0.33). This slightly raised the accuracy of the mechanistic approach (0.76 and 0.71 using PT and SDM, respectively).
The second geographical grouping was the area from Nois to the proximities of Cape Finisterre (N to Ría de Noya), where most presences of the species were observed (38 presences) but absences were also numerous (23 absences in rocky sites, prevalence = 0.62, Fig. 4a). Models showed the same accuracy (0.66), sensitivity (1) and specificity (0.09). Both approaches correctly classified all presences, but they did not correctly classify the absences in this area of favourable ocean and air temperatures (Fig. 4a).
Finally, in the Atlantic Rías where H. elongata was sparsely distributed (47 absences and 20 presences in rocky sites, prevalence = 0.30), the models obtained different results. PT correctly classified all presences but none of the absences (sensitivity = 1, specificity = 0), since satellite data suggested optimal ocean temperatures and below the PT for this species (between 14.3 and 17 °C, Fig. 1). In contrast, SDM performed poorly for presences (sensitivity = 0.4) and relatively well in predicting absences (specificity = 0.74). The latter was related to the inclusion of a negative term for air temperature in the SDM equation which is high in these embayments (Fig. 4a, inner coastline). This resulted in greater accuracy in this geographic area when using SDM (0.64) than when using PT (0.3).
For B. bifurcata, which occurs throughout the entire study area, PT classified all records in rocky sites as presences (sensitivity = 1, specificity = 0, accuracy = 0.46), suggesting no differences on a regional scale (Fig. 4b). The SDM correctly classified presences in all areas except in the cold ocean area around Cape Finisterre to Ría de Arosa (Fig. 4b, inner coastline) (sensitivity for SDM using all records from rocky sites = 0.76). However, the SDM correctly classified the absences in this stretch of coast and incorrectly elsewhere (specificity = 0.51, accuracy = 0.62) (Fig. 4b, outer coastline).
Projections
In the low prevalence area of H. elongata in northern Spain (from Llanes in Asturias to Nois in Galicia), the two models predicted this species’ local extinction at different projected temperatures. The SDM predicted that the species would become extinct at an increase in mean August SST of 1 °C and at an air anomaly of 1 °C (projected conditions for 2023 (Figure S1a), while the PT, which does not includes a term for atmospheric warming, predicted this species’ extinction in this area at an ocean thermal anomaly of 1.9 °C (2039, Figure S1b). Both methods, particularly the physiological approach, suggested less risk of extinction in the cold region around Cape Finisterre (Figure S1). In this geographic area is found the center of the seasonal upwelling of sub-superficial water of low temperature (Gómez-Gesteira et al., 2008). In the Atlantic Rías the SDM, that included a term for atmospheric warming, predicted that H. elongata would become extinct earlier than the PT (Figure S1). Although both approaches predicted the extinction of the species in the Cantabrian Sea at a fast pace, the projections differed in their precise patterns (above). In general, PTs predicted a larger area of presence of the species (Figure S3).
Bifurcaria bifurcata was predicted by the mechanistic approach to spread to all rocky sites under present-day and future climatic scenarios, since the PT of this species is higher than current and predicted temperatures. The SDM also predicted that this species will expand to previously unoccupied sites (Figure S2), but suggested a progressive colonization as ocean temperature increases. Thus, all northern warm sites (from Llanes to Malpica) would be colonized at an ocean temperature increase of 0.8 °C (2020, Figure S2a), corresponding to a stretch of coast of 816 km (Figure S3). All rocky sites south to Ría de Noya would be colonized at a projected anomaly of 1.6 °C on the Atlantic coast (2036, Figure S2b, 927 km in Figure S3). The coldest area around Cape Finisterre (i.e. the entire study area of 943 km in Figure S3) would be colonized at an increase of 2.3 °C in the Atlantic (2049, Figure S2c).
Discussion
By combining knowledge on physiological thresholds (PTs) and SDMs, we gained information of the species’ niches which supported the overall predicted trends. Results from both approaches coincided in predicting the local extinction of cold-temperate Himanthalia elongata at its southern distribution limit in northern Spain and the increase in occupancy of southern-lusitanic Bifurcaria bifurcata in response to global warming (Martínez et al., 2012a,b). Projections of the modelling methods were similar in areas where the species is either absent-rare or frequent (high presence) i.e., where potential and realized niches match (as reviewed in Soberón & Nakamura, 2009). This can be expected at contracting distributional edges, where environmental conditions are close to species’ physiological tolerance thresholds. Thus, information on physiological limits may support the trends forecasted by SDMs in ecological systems subjected to climate change. However, some incongruence is evident between approaches in areas of intermediate presence. In general, models based on PTs tend to over-predict the prevalence of the species, as they cannot identify absences in climatic conditions within the range of the species’ physiological tolerance. SDMs may implicitly capture the mechanisms driving such absences, such as low competitive potential of the focal species under particular environmental conditions, but this may or may not extrapolate to other geographic areas or times. By overlaying physiological information and distributional modelling, we may provide further support to projections under climate change if approaches coincide, or highlight potential flawed projections in areas where they do not. The latter may suggest the importance of biotic mechanisms not considered in the models (Kearney, 2006; Buckley et al., 2010, 2011). Although the use of physiological limits has been claimed to be a valid approach (Morin & Lechowicz, 2008; Nogués-Bravo, 2009), this study is one of the few studies combining these two sources of information in predictions (but see Hijmans & Graham, 2006; Austin et al., 2009; Jones et al., 2009; Buckley et al., 2011; Diamond et al., 2012). In the paragraphs that follow, we discuss patterns of concordance, as well as the reasons for differences between physiological and correlative approaches, using simple case examples.
Thermal niches and present-day distributions
Lethal and sublethal thermal conditions were associated with PTs using straightforward experiments simulating stress. These empirical PTs were congruent with the correlative response functions behind SDMs (Martínez et al., 2012a), an uncommon outcome in the few published studies relating physiology and SDMs (but see Austin et al., 2009). Perhaps the main reason for this concordance is the use of realistic environmental data similar to those measured in the field in the SDMs. H. elongata's distributional edge in the Cantabrian Sea seems to be determined by water temperatures above the physiological threshold of about 18 °C. This value is similar to that obtained using the correlative approach (i.e., the relationship between the percentage of observed presences and the averaged maximum August SST- SST Aug). It also lies within the temperature range traditionally used to explain the southern distribution limits of this and cold-temperate macroalgae in the Iberian Peninsula (20 ± 3 °C represented by the oceanic August isotherm, reviewed in Lüning, 1990). Sea water temperature was found to be the only significant factor driving the growth and physiology of this species in manipulative stress experiments comparing harsh vs. mild conditions of water temperature, air temperature, solar radiation and humidity (using the prototype in Martínez et al., 2012b; unpublished data). Although the equation from the SDM for this species included a term that accommodates atmospheric warming, this is not relevant in the eastern section of the Cantabrian Sea where ocean temperatures over 18 °C have become common in the last decade (e.g., Hobson et al., 2008). This situation differs from that on the Atlantic coast with persistent upwelling events of cold subsuperficial water (Gómez-Gesteira et al., 2008), and an increase of air temperatures with latitude. Results reinforce our conclusions that the recently observed dramatic contraction of H. elongata to the northwest corner of the Iberian Peninsula is caused by persistent ocean warming, whereas its persistence on the Atlantic coast is due to lower sea temperatures (Martínez et al., 2012a,b). Ocean warming has been linked to the decline of macroalgae of ecological importance in Northern Spain (Fernández, 2011; Díez et al., 2012; Lamela-Silvarrey et al., 2012; Duarte et al., 2013).
For B. bifurcata, there was again concordance among the ecophysiological experiment, the correlative approach and previous literature, suggesting that the physiological threshold of this species reaches higher temperatures than those in the studied area (reviewed in Lüning, 1990). These results agree with the high prevalence of this species in the warm area around Cape Peñas and towards the eastern Cantabrian Sea. The expansion of its distributional range in the English Channel (Mieszkowska et al., 2006) and towards southern Portugal (Lima et al., 2007) has been linked to ocean warming.
However, some discrepancies were found between the physiological and the correlative response functions suggested by experiments and SDMs, respectively. The most surprising discrepancy was the low observed prevalence of B. bifurcata around Cape Finisterre where temperatures were within the species’ physiological optimum. The importance of other physical stressors was neglected, as they were excluded as predictors in the statistical modelling (namely, minimum water temperature, maximum and minimum air temperatures, presence of rocky substrate, waviness and cloudiness, Martínez et al., 2012a). We thus hypothesize that this southern-lusitanic species may be less competitive than other cold-temperate algae, in particular in this geographic area of cold ocean conditions. For instance, B. bifurcata was suggested to be outcompeted by other fucoids in northern Spain (Cantabrian Sea) in the 1990s, a period of expansion for cold-temperate species (Arrontes, 2002). This example of potential niche underfilling, made apparent through the integrated approach adopted in this study, suggests the potential importance of biotic factors in driving the geographic distributions of species.
Concordance of models and predictions
Physiological thresholds and SDMs performed well and obtained similar classifications in areas where the species are either absent-rare or frequent (high presence). H. elongata absences in the low presence area of the Cantabrian Sea were accurately classified by both models, particularly in the eastern section which has maximum ocean temperatures in the studied area. The models also showed good performance in the area around Cape Ortegal to Cape Finisterre, where this fucoid is dominant and the centre of the seasonal upwelling of subsuperficial water of low temperature is found (Gómez-Gesteira et al., 2008). This area of cold ocean conditions falls within H. elongata's potential niche, while warm sea conditions in the Cantabrian Sea are outside its physiological bounds. For B. bifurcata, which showed greater tolerance to warm conditions and is spread throughout the study area, presences were, in general, correctly classified by both models. The two models performed fairly well when the species’ distributions reflected physiological tolerance to climatic conditions, and thus potential and realized niches matched.
Predictions reflect the general similarity in the classification of the models, and the trends forecasted by the two approaches were congruent. Both models predicted that H. elongata will become extinct in the Cantabrian Sea in response to ocean warming, whereas B. bifurcata will spread to unoccupied localities, potentially taking advantage of the decline of other fucoids, as noted in Martínez et al. (2012a). These predictions are consistent with evidence of the current distributional shift of the species in Europe (see above).
Uncertainties of models and predictions
PTs cannot predict absences under favourable climatic conditions and thus tend to over-predict species ranges. For B. bifurcata, absences were erroneously classified as presences by PT in all areas (i.e., null sensitivity). This resulted in an unrealistic forecast map, which suggested this species’ expansion along the entire coast (943 km) under present-day climatic conditions. A general trend to over-predict species ranges has been shown for mechanistic proxies of fundamental niches of species that apply thermal thresholds (Buckley et al., 2010). On the other hand, SDMs seemed to perform better assigning absences in areas of intermediate to low prevalence as, for example, in the area of low prevalence of B. bifurcata around Cape Finisterre. As previously mentioned, the classification of observed records in this area is based on a regression with temperature that contradicts physiological response. Potential biotic interactions here would be implicitly captured in the relationship with ocean temperature included in the SDM increasing specificity (ratio of correctly classified absences) and thus the accuracy for this stretch of coast. However, this relationship do not necessarily extrapolates to northern geographic areas with different algal communities or to future scenarios of declining fucoids in northern Spain. Furthermore, the SDMs failed to classify absences elsewhere, increasing the uncertainty of the forecasted colonization pattern obtained using SDMs for B. bifurcata (Figure S2 and S3). There was a consensus between the modelling methods regarding this species’ predicted expansion to previously unoccupied localities but not regarding its precise colonization pattern, which differed in the forecast maps obtained using the two approaches. This example illustrates the well known issue of how predictions using climatic and nonclimatic physical predictors may be less reliable if biotic factors are important, irrespective of the modelling approach used (Araújo & Luoto, 2007; Kearney & Porter, 2009; Elith et al., 2010). Nevertheless, combining mechanistic knowledge and correlative approaches may unravel mismatches between potential and realized niches and situations of range underfilling, identifying geographic areas with relevant biotic factors, thereby helping to make more robust predictions and new hypotheses regarding species distributions (Kearney, 2006; Buckley et al., 2010).
The area of uncertainty for H. elongata corresponds to the Atlantic Rías. In this area of intermediate prevalence within the species’ distributional ranges, the forecast maps based on PTs showed larger stretches of coast occupied by the species than those from SDMs. Apart from the limitations of PTs in predicting absences and the contribution of potential biotic interactions as discussed above, the discrepancy between approaches seems to be related to the inclusion of a term in the SDM equation accounting for atmospheric thermal stress during low tide. This suggests the importance of a second climatic factor not considered by the PT on the Atlantic coast, where low tide stress increases with latitude potentially altering the distribution of intertidal cold-temperate macroalgae (Martínez et al., 2012a,b), while sea temperature is lower due to the upwellings. However, preliminary experiments on tolerance to atmospheric thermal stress (Martínez et al., unpublished data) indicate that thermal thresholds are higher than the highest mean maximum values of the meteorological data used in the SDMs (average maximum air temperature in August of 27 °C from the Digital Climatic Atlas of the Iberian Peninsula). Averaged meteorological data tend to underestimate maximum air temperature in the rocky intertidal (Helmuth & Hofmann, 2001), making integrative approaches difficult since mechanistic models perform poorly when environmental predictors are estimated with error (Buckley et al., 2010).
Concluding remarks
Shifts in species distributions have created a need for rapid and accurate predictive modelling. Threshold models are recommended if species are known to be restricted by a particular environmental stressor (Buckley et al., 2010), as in the contracting distributional edges under global warming. The simple approach proposed here experimentally detects physiological limits for ectotherms and combines this knowledge with SDMs to help explain and predict distributional shifts in response to climate change. In comparison with biophysical models that incorporate as different equations the biological responses behind the geographic distribution of species, physiological models are potentially of less power to explain and predict trends. However, they can statistically identify tipping points in environmental conditions with small error without the substantial information and time required to parameterize biophysical models. This is perhaps why probabilistic approaches are preferred in most ecological studies.
Regarding whether physiological limits or SDMs perform better, we may distinguish two situations. Projections of the modelling methods were similar in areas where the species is either absent-rare or frequent. In areas of intermediate presence, where biotic interactions are feasible, models based on PTs tend to over-predict the prevalence of the species because cannot identify absences in climatic conditions within the range of the species’ physiological tolerance. SDMs based on predictors of ecological relevance for species tended to have higher sensitivity than threshold models in these areas. However, caution should be taken in not including predictor variables that do not represent causal mechanisms, as they potentially constrain the predictive power of SDMs (Austin, 2002; Elith et al., 2010). Both approaches may give flawed projections if do not account for biotic factors. A combination of physiological knowledge and SDMs can highlight this problem, providing a simple and rapid way to gain insight into species’ niches. This is helpful in formulating more robust predictions, when both approaches agree, and in formulating new hypotheses on potential reasons of disagreement, as exemplified in this study.
Acknowledgements
We thank M. A. Olalla-Tárraga for his comments on the manuscript, and M. A. L. De Hond for linguistic assistance. This research was supported by projects CGL2007–66095, HP2007–0081 and CGL2010–19301 funded by the Spanish Ministry of Science, and PTDC/MAR/105147/2008 funded by the FCT (Portugal).
