Habitat Suitability Modeling and Ecological Forecasting of Northern Goshawk Nesting Habitat

Northern Goshawks (Accipiter gentilis) are dynamic forest raptors often used as a management indicator species in forest management planning. Despite their use as indicators, there is a limited understanding of habitat requirements for goshawks, especially in dry forested landscapes as climate change occurs. We examined goshawk nesting habitat on the Helena-Lewis and Clark National Forest (HLCNF) in central Montana to further understand climate change on goshawk nesting habitat. The HLCNF receives low annual precipitation, making this forest useful to understand climate change in dry forest in structure and ecology. We measured characteristics of nest sites throughout the HLCNF and used principle component analysis to predict important variables for nest trees and nesting habitat for goshawks in this landscape. We also used remote sensing and GIS to create a composite map predicting areas of high, medium and low suitability for nesting habitat for goshawks in the HLCNF. We used the composite map to predict potential impacts of mountain pine beetle (Dendroctonus ponderosae), fire, and climate change to Northern Goshawk habitat in the HLCNF. The best indicators of a chosen nest tree were diameter at breast height, tree height, and tree status, which can be considered indicative of trees preferred for nesting by goshawks. The best indicators for nest forest were canopy closure, slope angle, and slope position, which indicate habitat areas of high suitability for goshawk nesting. Two forecast models where developed based on climate change predictions. After overlapping a projected mountain pine beetle risk map and fire frequency risk map onto the modeled of high suitability goshawk nesting habitat, we calculated that 52 % of modeled high suitability habitat may be at risk for mountain pine beetle blight and 66 % high suitability habitat at risk for frequent wildfires, which will reduce goshawk habitat quality.


Introduction
Field Collection 126 We used the protocol outlined by the Northern Goshawk Inventory and Monitoring  132 We also broadcasted previously taped goshawk calls on a digital amplifier at variable decibels  140 We recorded values for a list of variables for each nest site pertaining to the physical 141 habitat characteristics of both the tree containing the nest and the surrounding area. We recorded 142 the species of nest tree, status of the nest tree (dead or alive), diameter at breast height (DBH), 143 nest height, and nest tree height. For the area surrounding the nest site, we recorded placement of 144 the nest (elevation, nest orientation, slope aspect, slope angle, slope position (summit, shoulder, 145 footslope, toeslope)). As well as habitat characteristics (age class of the trees, overstory canopy 146 closure, evidence of previous timber cutting (yes or no), distance to a water source, tree species present within the immediate area around the nest (10 m radius around the nest), and vegetation 148 cover within the immediate area around the nest). including tree species, status (dead or alive), DBH, nest height, and nest-tree height. Tree status 155 was coded as a dichotomous variable with zero indicating the nest tree was dead and one 156 indicating the nest tree was alive. Tree species were coded by rank with the most common nest 157 tree species having the higher number (Douglas fir coded as 4, lodgepole pine coded as 3, 158 quaking aspen coded as 2, ponderosa pine coded as 1). All other variables were entered as 159 continuous variables-DBH, nest height, nest-tree height. 160 We used six variables to assess characteristics of habitat surrounding nest sites including 161 elevation, canopy closure, overstory canopy species, slope aspect, slope angle, age class of trees, 162 and slope position. Age class of the trees was coded by rank with the most common age class 163 present at nest stands, mature (coded 2) and a mixture of mature and pole trees (coded 1). Age  Analysis (FIA) programs survey, analyze, and produce geospatial data layers for various forest 178 attributes. We used the VMap product to obtain vegetation metrics in vector format for tree 179 canopy cover-class size. Tree canopy cover is classified by percent canopy cover. We derived 180 basal area, in m 2 /ha, from the FIA product at 250 m resolution. We also derived vegetation  Using USGS Earth Explorer, we acquired Shuttle Radar Topography Mission (SRTM) 1 185 arc-second (30 m) void-filled data, which were used to derive terrain elevation, slope, and aspect. 186 We acquired Landsat 8 Operational Land Imager (OLI) data from the USGS Earth Resources   of the variables that will be used for running each model (Table 1). Any two variables with a 248 correlation coefficient greater than 0.7 were flagged. The variables tree size and tree canopy 249 were highly correlated at 0.987, therefore tree size was excluded and the remaining eight 250 variables were used in each model.

251
Once the 3 models were run using the 8 environmental variables, each model was

292
Nest-Tree Characteristics 293 We located 190 nest sites. Many of these nests were in the same territories and some of 294 the nest were old and falling apart. We analyzed 23 total goshawk nests, both active regular 295 territory (n = 15) and inactive regular territory (n = 7 active the year before during surveys; 296 unknown: n = 1 irregular territory and has been abandoned); however, four of these sites were 297 present on private property, so we were unable to take all descriptive measurements at these  Table   304 2). Nest trees ranged from 18 to 78 cm in diameter, but the average nest trees measured from 25   (Table 5). One-hundred-three 357 nests coincided with high suitability areas and 66 nests were located in medium suitability areas.

358
There is some uncertainty in the habitat suitability map because not all models are perfect and 359 that 21 nests were found in areas modeled as low suitability. Not all nests were active, and some 360 of the 21 nest areas may have been subject to habitat disturbances between the last known active 361 year and present, which could not be accounted for in our model. Another possible explanation is 362 that 2 nd year nesters will sometime choose poor locations for nesting.

363
After overlapping the reclassified 2027 projected mountain pine beetle risk map onto the 364 modeled best available goshawk habitat, we calculated that 52% of modeled high suitability 365 habitat may be at risk for mountain pine beetle blight (Fig. 8). We overlaid the reclassified fire 366 frequency risk map and the modeled high suitability nesting habitat map using the "intersect" 367 function and identified 66% of high suitability habitat at risk for frequent wildfires, possibly 368 driven by drying conditions from climate change (Fig. 8). Through vegetation analysis and habitat modeling, we were able to determine that high 374 suitability goshawk nesting habitat is some of the most underrepresented habitat area available in 375 the HLCNF. We found that DBH, tree height, and tree status (alive or dead) were the best 376 predictors of which nest trees goshawks would use for nesting, whereas canopy closure, slope suitability goshawk nesting habitat with the mountain pine beetle risk map and calculated that 52% of modeled high suitability habitat may be at risk for mountain pine beetle blight (Fig. 8). 405 Continued warming and decreased precipitation is expected to put even greater amounts of 406 habitat at risk in the future as there will likely be an increase in beetle infestation in areas that are 407 not currently affected (Kurz et al. 2008).

408
Warmer and drier climate will also create conditions that make forested areas more 409 susceptible to wildfires and possibly increase the frequency of fires. We observed the reclassified 410 fire frequency risk map and overlaid the modeled high suitability habitat with this map and found 411 that 66% of modeled high suitability habitat may be at risk for frequent wildfires (Fig. 8). It is, 412 however, not guaranteed that wildfires will occur at the modeled high suitability goshawk nest 413 sites because wildfires effect the whole landscape not just high suitability goshawk nest sites.

414
Regardless, warming trends will possibly contribute to continued goshawk habitat loss in the 415 HLCNF.

416
The limiting factor that could not be accounted for in this project was exact location of 417 the nesting habitat that goshawks prefer to validate the suitability model. Although we utilized 418 eight environmental variables that influence goshawk nesting habitat in our habitat modeling, 419 there is some uncertainty in our results. We were not able to incorporate an important 420 requirement for nesting habitat in our modeling; goshawks prefer to nest in forested areas that 421 have a closed canopy and an open understory. We did use a tree-canopy percent-cover dataset in 422 our modeling, but this was a generalized dataset and did not contain sufficient understory