Forest Ecology and Management 262 (2011) 307–316
Contents lists available at ScienceDirect
Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco
Assessing forest vulnerability and the potential distribution of pine beetles under current and future climate scenarios in the Interior West of the US Paul H. Evangelista a,⇑, Sunil Kumar a, Thomas J. Stohlgren b, Nicholas E. Young a a b
Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO 80523, USA US Geological Survey, Fort Collins Science Center, Fort Collins, CO 80526, USA
a r t i c l e
i n f o
Article history: Received 19 October 2010 Received in revised form 14 March 2011 Accepted 24 March 2011 Available online 6 May 2011 Keywords: Bioclim Climate change Maxent Niche models Bark beetles Rocky mountains
a b s t r a c t The aim of our study was to estimate forest vulnerability and potential distribution of three bark beetles (Curculionidae: Scolytinae) under current and projected climate conditions for 2020 and 2050. Our study focused on the mountain pine beetle (Dendroctonus ponderosae), western pine beetle (Dendroctonus brevicomis), and pine engraver (Ips pini). This study was conducted across eight states in the Interior West of the US covering approximately 2.2 million km2 and encompassing about 95% of the Rocky Mountains in the contiguous US. Our analyses relied on aerial surveys of bark beetle outbreaks that occurred between 1991 and 2008. Occurrence points for each species were generated within polygons created from the aerial surveys. Current and projected climate scenarios were acquired from the WorldClim database and represented by 19 bioclimatic variables. We used Maxent modeling technique fit with occurrence points and current climate data to model potential beetle distributions and forest vulnerability. Three available climate models, each having two emission scenarios, were modeled independently and results averaged to produce two predictions for 2020 and two predictions for 2050 for each analysis. Environmental parameters defined by current climate models were then used to predict conditions under future climate scenarios, and changes in different species’ ranges were calculated. Our results suggested that the potential distribution for bark beetles under current climate conditions is extensive, which coincides with infestation trends observed in the last decade. Our results predicted that suitable habitats for the mountain pine beetle and pine engraver beetle will stabilize or decrease under future climate conditions, while habitat for the western pine beetle will continue to increase over time. The greatest increase in habitat area was for the western pine beetle, where one climate model predicted a 27% increase by 2050. In contrast, the predicted habitat of the mountain pine beetle from another climate model suggested a decrease in habitat areas as great as 46% by 2050. Generally, 2020 and 2050 models that tested the three climate scenarios independently had similar trends, though one climate scenario for the western pine beetle produced contrasting results. Ranges for all three species of bark beetles shifted considerably geographically suggesting that some host species may become more vulnerable to beetle attack in the future, while others may have a reduced risk over time. Ó 2011 Elsevier B.V. All rights reserved.
1. Introduction Insect epidemics are sharply increasing in coniferous forests throughout North America causing dramatic impacts to ecosystem processes and disrupting forest-dependent economies. Insects and associated pathogens have resulted in extensive tree mortality that, in turn, has altered forest structure, composition, and function (Raffa et al., 2008). Bark beetles (Curculionidae: Scolytinae) have become especially detrimental to coniferous forests. Although most of North America’s bark beetles are native and play an ⇑ Corresponding author. Present address: Natural Resource Ecology Laboratory, A204 NESB, Colorado State University, Fort Collins, CO 80523-1499, USA. Tel.: +1 970 491 2302; fax: +1 970 491 1965. E-mail address:
[email protected] (P.H. Evangelista). 0378-1127/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.foreco.2011.03.036
integral role in forest dynamics (Fleming et al., 2002; SanchezMartinez and Wagner, 2002), recent outbreaks are increasing in frequency, severity and extent (Westfall, 2006; Kurz et al., 2008; Raffa et al., 2008). Tree mortality from bark beetle infestations increased from 1.6 to 4 million ha in the US between 2002 and 2003, the largest annual increase recorded (WFLC, 2009). The extent of the infestation has remained high through the present, impacting approximately 3.6 million ha in 2008 (USDA, 2009). Increases in mountain pine beetle (Dendroctonus ponderosae) infestation have been especially noticeable. Between 1990 and 2001, mountain pine beetles infested less than 400,000 ha in the US; however, the extent of the infestation has steadily increased to more than 2.5 million ha by 2008 (USDA, 2009). Outbreaks of the mountain pine beetle have also been problematic in more northern latitudes and at higher elevations. In British Columbia, Canada, mountain
308
P.H. Evangelista et al. / Forest Ecology and Management 262 (2011) 307–316
pine beetles infested 164,000 ha in 1999; by 2004, the outbreak grew to 7 million ha (Ministry of Forests, 2005) and to 9.2 million ha by 2006 (Westfall, 2007). The northern range of mountain pine beetles has been historically limited by climate, as host species occur well beyond climatic thresholds (Carroll et al., 2004). Climate variability over the last few decades has been implicated as the primary cause of the recent expansion of outbreaks in the US and Canada (Hicke et al., 2006; Thomson, 2009). Similar trends have been observed with other herbivorous forest insects, such as the western pine beetle (Dendroctonus brevicomis), pine engravers (Ips pini) and spruce beetle (Dendroctonus rufipennis; Berg et al., 2006; USDA, 2009). Bark beetles have evolved with native forest ecosystems and are important agents in maintaining host species composition, distribution patterns and mixed age-classes (Waring and Pitman, 1985; Oliver, 1995; Malmstrom and Raffa, 2000). Infestations have long been associated with forest stands that are stressed or dying and with host trees that have lower vigor and fewer defense mechanisms (Rudinsky, 1962; Anderson and Anderson, 1968; Berryman, 1972; Larsson et al., 1983; Klepzig et al., 1991; Reid and Robb, 1999). Other studies suggest that infestations are a function of more complex abiotic and biotic relationships between bark beetles and their hosts. These include reproductive and dispersal opportunities for bark beetles (Aukema et al., 2005; Roberston et al., 2009), transport and infection by fungi and other pathogens (Adams et al., 2008; Bleiker et al., 2009), predation and competition of bark beetles (Berryman et al., 1987; Schlyter and Anderbrant, 1993), fire (Breece et al., 2008; Fettig et al., 2008), and management practices (Waring and Pitman, 1985; Hayes et al., 2008). Although the results of these studies help us better understand local dynamics between bark beetles and their hosts, they are difficult to apply across large spatial or temporal scales. Alternatively, life cycles of bark beetles are known to be regulated by climate, which can be useful for predicting population dynamics and potential risk across large landscapes and time spans. Climate, particularly seasonal temperatures, may affect bark beetles during multiple stages of their life cycles. For example, several studies on Ips engravers (Ips spp.) have demonstrated that colder temperatures disrupt egg development (Jonsson et al., 2009) and synchronized flight activities (Aukema et al., 2005), while extended winters reduce the number of successful broods in a given year (Anderbrant, 1989). Similar studies on mountain pine beetles have found that cold temperatures cause significant mortality for emerging adults in the spring (Amman, 1973) and late larval stages in early fall and late winter (Carrol and Safranyik, 2004). Climate also affects host susceptibility by reducing tree vigor and defense mechanisms creating a more favorable environment for beetle development (Mattson and Haack, 1987; Negron et al., 2009). Healthy trees can often repel bark beetle attacks by producing a protective resin; however, under drought conditions, trees may not be capable of producing enough resin to resist attacks (Smith, 1963; Raffa and Berryman, 1987). Host trees can also be weakened by other disturbances that are correlated with climate, such as fire (McHugh et al., 2003), excessive moisture (Kalkstein, 1976), lightning (Anderson and Anderson, 1968), ice (Smith, 2000) and other pathogens (Klepzig et al., 2001). Given the projected short- and long-term changes of global climate, the range of bark beetles and the severity of outbreaks will continue to be modified. The ability to predict spatial and temporal shifts of potential bark beetle habitats, and the mechanisms that drive outbreaks, is critical for assessing risk and developing adaptive management strategies that reduce negative impacts. As previously described, local dynamics between bark beetles and their hosts can be extremely variable; and they tend to be poor predictors across large spatial or temporal scales. Climate is being increasingly used with ecological niche models to spatially define a
species’ range or potential distribution. These methods have been successfully applied to wildlife (Evangelista et al., 2008; Boubli and de Lima, 2009), plants (Pearson et al., 2004; Hijmans and Graham, 2006; Kumar and Stohlgren, 2009), invasive species (Jarnevich and Stohlgren, 2009; Kumar et al., 2009), pathogens (Holt et al., 2009), and insects (Peterson and Nakazawa, 2007). With the availability of future climate scenarios, similar methods can be used to predict a species’ range over time (Pearson et al., 2004; Holt et al., 2009; Jarnevich and Stohlgren, 2009). Our objective was to test 19 bioclimatic variables with Maxent modeling methods to predict the potential distribution of the mountain pine beetle, western pine beetle and pine engraver beetles under current climate conditions. Results from each analysis defined the current climate parameters for each species, and they provided a preliminary estimate of potential insect distributions under future climate scenarios. 2. Methods 2.1. Study area Our study area encompassed the Interior West of the US which is commonly defined by eight states: New Mexico, Arizona, Colorado, Utah, Nevada, Idaho, Wyoming and Montana. These states include approximately 95% of the Rocky Mountains in the contiguous US. The study area covers approximately 2.2 million km2 with elevations ranging from 25 to 4,280 m a.s.l. The majority of the montane and subalpine forests are dominated by coniferous species, composed largely of ponderosa pine (Pinus ponderosa), Douglas-fir (Pseudotsuga menziesii), lodgepole pine (Pinus contorta), whitebark pine (Pinus albicaulis), subalpine fir (Abies lasiocarpa), and Englemann spruce (Picea engelmannii). Eastern margins of the study area are predominantly grasslands of the Great Plains, while the Southwest is generally dominated by semi-arid shrubs, pinyon pine (Pinus edulis), and juniper (Juniperus spp.) woodlands. Climate throughout the Interior West varies regionally. Generally, southern latitudes are warmer and drier, while northern latitudes are cooler and wetter. The US Forest Service manages a significant proportion of the forests of the Interior West. Within our study area, the agency conducts operations under four regional administrative units. The Northern Region (Region 1) includes Montana and northern Idaho. The Rocky Mountain Region (Region 2) includes Colorado and most of Wyoming. The Southwestern Region (Region 3) encompasses New Mexico and Arizona. Finally, the Intermountain West (Region 4) covers Utah, Nevada, southern Idaho and the western margins of Wyoming (see http://www.fs.fed.us/contactus/regions.shtml). 2.2. Bark beetle occurrence data Our analyses relied on aerial surveys conducted by the US Forest Service’s Forest Health Protection Aviation Program. Polygons of infested areas were acquired between 2000 and 2008 for Region 1 (www.fs.fed.us/r1-r4/spf/fhp/aerial/gisdata.html); between 1994 and 2007 for Region 2 (www.fs.fed.us/r2/resources/fhm/aerialsurvey/); between 1998 and 2008 for Region 3 (www.fs.fed.us/r3/ gis/nm_data.shtml); and between 1991 and 2007 for Region 4 (www.fs.fed.us/r1-r4/spf/fhp/aerial/gisdata.html). Data were not available for Arizona at the time of this study. Polygon data representing infestations were converted to 1 km2 grids using ArcGIS (version 9.2; ESRI, Redlands, CA, USA). Occurrence points for infestations of each bark beetle were derived by calculating the centroid of each 1 km2 cell. To reduce spatial autocorrelation effects (Veloz, 2009), we buffered occurrence points by three cells (3 3 km2), removing adjacent occurrence points. We generated minimum
P.H. Evangelista et al. / Forest Ecology and Management 262 (2011) 307–316
convex polygons (MCPs) around all occurrence points for each species to define the area for training the Maxent model and projected the results to the extent of our study area. This was conducted so that background points for the Maxent model were drawn from within the sampling range and to improve model performance (VanDerWal et al., 2009). Occurrence points for the mountain pine beetle totaled 10,775, western pine beetle totaled 6,167, and pine engraver beetle totaled 3,180. 2.3. Environmental variables Current climate conditions were represented by 19 bioclimatic variables acquired from the WorldClim database v1.4 (Hijmans et al., 2006; www.worldclim.org). These data are available at 1 km2 grids and are interpolated from monthly weather station measurements collected between 1950 and 2000 (Hijmans et al., 2005). The WorldClim dataset uses altitude, temperature and precipitation to derive monthly, quarterly and annual climate indices to represent trends (e.g., mean diurnal temperature range), seasonality (e.g., temperature seasonality) and extremes (e.g., maximum temperature of the warmest month) that are biologically relevant (Kumar et al., 2009; for more details, see www.worldclim.org/ bioclim). Future climate scenarios were also acquired from the WorldClim database (www.worldclim.org/futdown.htm). Three climate models were used for the years 2020 and 2050. These were produced from the Canadian Centre for Climate Modeling and Analysis (CCCMA), Hadley Centre Coupled Model v3 (HadCM3) and Commonwealth Scientific and Industrial Research Organization (CSIRO). Each climate model provides two emission scenarios representing conservative (a2a) and liberal (b2a) estimations. Each of the three climate models and the associated emission scenarios were tested independently. In addition, we averaged the results from the CCCMA, HadCM3, and CSIRO future climate models and each of the two emission scenarios, resulting in two model scenarios for 2020 and two for 2050 (i.e., 2020 a2a; 2020 b2a; 2050 a2a; 2050 b2a).
309
criteria (Pearson et al., 2004) to determine the threshold for habitat and non-habitat of each species from the continuous probability of occurrence maps from Maxent. Although the Maxent model is not sensitive to multicollinearity, we ran cross-correlation tests for our independent variables to assist in interpreting our results. Crosscorrelation tests were conducted using SYSTAT version 12 (SYSTAT Software Inc., Chicago, Illinois, USA). 2.5. Hosts species Bark beetles generally target specific host species. Mountain pine beetles infest lodgepole pine, ponderosa pine, and limber pine, and are seldom found in pinyon or bristlecone pine (Leatherman et al., 1999). Western pine beetles almost exclusively infest ponderosa pine (Smith, 1990). Pine engraver beetles are considered more of a generalist species, commonly infesting ponderosa pine and lodgepole pine, and occasionally pinyon pine and limber pine (Kegeley et al., 1997). To identify the range of potential host species in our study area, we used LANDFIRE Vegetation Type land-cover map (http://www.landfire.gov/products_national.php). The LANDFIRE dataset was developed using a compilation of field databases for reference plots with biophysical gradients and Landsat imagery (Rollins, 2009). We combined ecological systems to broadly represent ponderosa pine, lodgepole pine, pinyon-juniper, mixed conifer forests, and limber-bristlecone pine communities. Vegetation codes representing different pine species and communities are described in NatureServe (2009). Ponderosa pine area is the sum of vegetation codes 1031, 1053, 1054, 1117, 1179. Lodgepole pine area is the sum of vegetation codes 1050, 1058, 1167. Pinyon-juniper area is the sum of vegetation codes 1016, 1019, 1025, 1059. Mixed conifer forest area is the sum of vegetation codes 1024, 1026, 1028, 1045, 1047, 1051, 1052, 1061. Limber pine and bristlecone pine area is the sum of vegetation codes 1020, 1049, 1057. Our modified host maps were then used to clip the potential distributions of each bark beetle generated from our climate models.
2.4. Model and analyses 3. Results We conducted our analyses using the Maxent software v.3.3.1 (http://www.cs.princeton.edu/~schapire/maxent/), which is a general-purpose niche modeling algorithm for estimating probability of distributions based on the principle of maximum entropy (Phillips et al., 2004, 2006). Maxent uses presence-only data to identify environmental conditions based on the independent variables to predict a species’ distribution excluding all conditions that are unfounded or undefined. The model is nonlinear, nonparametric, and not sensitive to multicollinearity. Results from the Maxent model include two model evaluations, the area under the receiving operating characteristic (ROC) curve (AUC) and jackknife testing provides the percent contribution of each independent variable used in the model. For each independent and the averaged climate analysis, we used 70% of the presence points to train models for each of the bark beetles. The remaining 30% were withheld for model validation. Each analysis consisted of ten replicates using a different set of randomly drawn presence points for training and validating the model. The model outputs were averaged across the ten replicates for a single model, and model AUC values and percent variable contributions were recorded. To estimate the current habitat distribution of bark beetles and forest vulnerability for each species, a threshold was determined to define habitat and non-habitat based on the results from Maxent models run using current 19 bioclimatic variables. The choice of a threshold value is important because model results and outputs will vary based on the threshold applied. We used 95% sensitivity
3.1. Mountain pine beetle Suitable habitat for the mountain pine beetle covered approximately 244,800 km2 with a noticeable concentration in the northwestern states of Region 1 and the central Rocky Mountains of Region 2 (Fig. 1). Under current climate scenarios for the mountain pine beetle, the Maxent model performed reasonably well having an average test AUC of 0.898 (±0.003). Precipitation of the warmest quarter (Bio18) was the best predictor with an average contribution of 65.3% (Table 1). The second leading predictor was the mean temperature of the warmest quarter (Bio10) with an average contribution of 8.7%. The Bio10 variable was also highly correlated with two other variables: annual mean temperature (Bio1) and maximum temperature of the warmest month (Bio5). The Maxent model’s internal jackknife test of variable importance indicated that precipitation of the warmest quarter (Bio18) and annual precipitation (Bio12) were the two most important predictors for mountain pine beetle range. All of the independent models using future CCCMA, HadCM3 and CSIRO climate models predicted decreasing habitat suitability for the mountain pine beetle (Fig. A.1). For 2020, suitable habitat ranged from 208,400 km2 (CCCMA, b2a; 15%) to 158,600 km2 (HadCM3, b2a; 35%). Suitable habitat for 2050 ranged from 226,700 km2 (CCCMA, b2a; 7%) to 132,700 km2 (HadCM3, b2a; 46%). Areas of suitable habitat for the mountain pine beetle by host species were derived from the LANDFIRE Vegetation Type
310
P.H. Evangelista et al. / Forest Ecology and Management 262 (2011) 307–316
Fig. 1. The predicted range of the mountain pine beetle under current (a) and future climate scenarios. Climatic extents were reduced to areas dominated by host species as defined by the LANDFIRE Vegetation Type land-cover map. Two emission scenarios, a2a and b2a, were modeled for the years 2020 (b and c) and 2050 (d and e), respectively. The green areas represent the potential distribution extent under current climate conditions, red areas represent distribution increases, and gray areas represent distribution decreases. Average test AUC was 0.898 (±0.003). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
land-cover map and are provided in Table A.1. These include areas of the model results for the three independent climate scenarios and for the averaged results, including a2a and b2a emission scenarios.
Averaged models for all three climate scenarios predicted that the total area of mountain pine beetle habitat will decrease by16% (Fig. 1, Table 2) for 2020 under the a2a emission scenario. Approximately 74% of the current habitat will remain suitable,
P.H. Evangelista et al. / Forest Ecology and Management 262 (2011) 307–316 Table 1 The top four predictors and their average (from 10 replicates) percent contribution reported from the Maxent model for the three bark beetles. Predictor variables
Contribution (%)
Mountain pine beetle Precipitation of the warmest quarter (Bio18) Mean temperature of the warmest quarter (Bio10) Precipitation seasonality (Bio15) Precipitation of driest quarter (Bio17)
65.3 8.7 6.9 5.4
Western pine beetle Annual mean precipitation (Bio12) Precipitation of wettest month (Bio13) Precipitation of the wettest quarter (Bio16) Precipitation seasonality (Bio15)
39.5 23.1 8.0 6.9
Pine engraver beetle Annual mean precipitation (Bio12) Annual mean temperature (Bio1) Precipitation of wettest month (Bio13) Precipitation seasonality (Bio15)
37.8 14.2 13.0 10.6
Table 2 The predicted area (km2) of suitable habitat for mountain pine beetle, western pine beetle and pine engraver beetle averaged from three different climate models; CCCMA, HadCM3 and CSIRO. Predicted changes in area for 2020 and 2050 show the calculated areas of stable, increasing and decreasing habitat compared to current climate models. Calculations include two emission scenarios (i.e. a2a, b2a) and percent changes (%) reported in parentheses. Total area Mountain pine beetle 2020 a2 205,300 ( 2020 b2 179,100 ( 2050 a2 168,800 ( 2050 b2 176,800 (
Stable 16) 27) 31) 28)
182,300 156,400 141,400 139,300
Increase
Decrease
(74) (64) (58) (57)
23,100 22,700 27,400 37,400
(9) (9) (11) (15)
62,500 ( 26) 88,400 ( 36) 103,400 ( 42) 105,500 ( 43)
Western pine beetle 2020 a2 137,300 2020 b2 138,700 2050 a2 134,200 2050 b2 141,800
(8) (9) (5) (11)
112,800 (89) 112,200 (88) 99,000 (78) 106,000 (83)
24,500 26,500 35,200 35,800
(19) (21) (28) (28)
14,600 15,200 28,500 21,400
( ( ( (
11) 12) 22) 17)
Pine engraver beetle 2020 a2 204,100 2020 b2 197,500 2050 a2 179,400 2050 b2 187,900
( ( ( (
161,400 151,400 118,600 124,000
42,700 46,100 60,800 64,000
(20) (22) (29) (30)
49,800 59,800 92,600 87,300
( ( ( (
24) 28) 44) 41)
3) 7) 15) 11)
(76) (72) (56) (59)
while new habitats are predicted to increase by 9% and previous habitats will decrease by 26%. Averaged models for 2020 b2a climate scenarios predict that total mountain pine beetle habitat will decrease by 27%. Of this area, about 64% will remain suitable from the current climate prediction. New habitat is predicted to increase by 9% and previous suitable habitat will decrease 36%. Under the 2050 averaged climate scenarios, a2a and b2a models predicted a decrease in total mountain pine beetle habitat from our current habitat model (Table 2). Total habitat areas for the a2a and b2a decreased by 28% and 28%, respectively. Models for 2050 a2a show that approximately 58% will remain suitable habitat from our current habitat model, while new habitat was predicted to increase by 11% and previous habitat will decrease by 42%. Models for 2050 b2a predict that 57% will remain suitable habitat. New habitat is expected to increase by 15% and previous habitat will decrease by 43%. Model results for the current and averaged climate scenarios prior to clipping to the LANDFIRE Vegetation Type landcover map are presented in Fig. B.1. 3.2. Western pine beetle Suitable habitat for the western pine beetle under current climate scenarios covered 127,400 km2, with the largest extent
311
occurring in Region 1 and fragmented habitats in the southern Rocky Mountains of Region 2 (Fig. 2). The average test AUC for the models was 0.938 (±0.002). Annual mean precipitation (Bio12) and precipitation of the wettest month (Bio13) were the leading predictors, with average contributions of 39.5% and 23.1%, respectively (Table 1). Annual mean precipitation was highly correlated to the precipitation of the coldest quarter (Bio19) and the precipitation of the wettest month was highly correlated to the precipitation of the wettest month (Bio16). The Maxent model’s internal jackknife test of variable importance showed that precipitation of the wettest quarter (Bio16) and precipitation of the wettest month (Bio13) were the two most important predictors for western pine beetle range at this spatial scale. Independent models using future climate scenarios for the western pine beetle had some variation in the results (Fig. A.2). The CCCMA and CSIRO models had similar predictive trends suggesting that suitable habitat will steadily increase over time. Increases in the area of suitable habitat for 2020 only ranged from 142,400 km2 (CCCMA, b2a; +12%) to 146,600 (CSIRO, b2a; +15%), while 2050 predictions range from 148,000 km2 (CSIRO, a2a; +16%) to 161,500 km2 (CCCMA, b2a; +27%). In contrast, the HadCM3 model predicted that suitable habitat for the western pine beetle will remain stable or decrease in time. The area of suitable habitat from the 2020 HadCM3 model ranged from 127,000 km2 (b2a;
