Balancing conservation priorities for grassland and forest specialist bird communities in agriculturally dominated landscapes

Effective conservation planning often requires difficult decisions when at-risk species inhabit economically valuable landscapes or if the needs of multiple threatened species do not align. In the agriculture-dominated landscape of eastern Ontario and southwestern Quebec, Canada, conflicting habitat requirements exist between threatened grassland birds benefiting from certain agriculture practices and those of a diverse woodland bird community dependent on forest recovery. Using multi-scale species distribution models with Breeding Bird Survey (BBS) data, we assessed habitat suitability for 8 threatened grassland and forest specialists within this region. We also identified landscapes that jointly maximize occurrence of the 8 focal species and diversity of the overall grassland and forest communities. Influential habitat associations differed among species at the territory (200m radius) and landscape level (1km), highlighting the importance of considering multiple spatial scales. Species diversity was maximized when forest or grassland/pasture cover approached 40–50%, indicating a positive response to land cover heterogeneity. We identified species diversity hotspots near Lake Huron, as well as along the shore and southeast of the St. Lawrence River. These areas represent mosaic landscapes, balancing forest patches, wetland, grassland/pasture, and row crops such as corn, soybean, and cereals. Despite drastic landscape changes associated with agroecosystems, we demonstrate that targeted habitat protection and enhancement that prioritizes land cover diversity can maximize protection of bird communities with directly contrasting needs. We highlight multiple pathways to achieve this balance, including forest retention or separating row crops with hedgerows and wooded fence-lines, improving flexibility in conservation approaches.


32
Conservation planning requires decisions on how to allocate limited resources to maximize under SARA (e.g., wood thrush Hylocichla mustelina, Eastern wood-pewee Contopus virens). 66 This scenario leads to a challenging conservation dilemma on how to best manage 67 agroecosystems when the habitat needs for declining, federally listed grassland species conflict 68 with those for threatened forest birds and the broader diversity of the forest bird community. 69 In this study, we focused on an extensive, agricultural-dominated region of eastern 70 Ontario and southwestern Quebec ( Figure 1A) to address this conservation challenge. 71 Specifically, we (1) used a multi-scale species distribution approach to identify geographic areas 72 with the greatest potential to maximize conservation of three targets: i) declining grassland birds, 73 ii) declining forest birds, and iii) total avian diversity. For both declining grassland and forest 74 birds we considered species that are federally listed in Canada as species at risk (SAR), but also 75 species of conservation concern that have yet to be listed. We then assessed (2) which landscape 76 characteristics and spatial scales most influenced occurrence of each species, and thus how to 77 optimize conservation of the three target groups. Finally, (3) we used this information as the 78 basis for a systematic prioritization analysis to formally identify the landscapes and regions that 79 allow us to reach a 25% landscape area protection goal in a manner that maximizes conservation 80 of grassland and forest-dependent species. This goal was chosen to align with Canada's 81 commitment to protect 25% of terrestrial area by 2025. The results from this study will provide 82 critical insights into the types of landscapes we need to manage or protect to jointly benefit at-83 risk forest and grassland specialists despite contrasting habitat needs. The study region in eastern Ontario and southwestern Quebec, Canada, represents a land cover 88 gradient from forest to intensive agriculture (primarily livestock cash crop) that spans 89 approximately 600 km in latitude and 1000 km in longitude ( Figure 1A). This region has 90 experienced extensive habitat modification due to agricultural practices and urbanization, 91 particularly in the south. To the north and west, forests have been cleared for timber and pastoral 92 purposes, yet large sections of native mixed-wood forest, wetland, scrub, and riparian habitats 93 still exist, including in protected areas such as Algonquin Provincial Park. We focused on the 94 distribution of four federally listed species-at-risk (SAR) and four species of concern (SOC), 95 consisting of both grassland and forest specialists (  115 We used a single year of BBS data to match remote-sensing land cover estimates 116 collected in 2018 and limit uncertainty introduced by crop rotation. Count data were converted to 117 occurrence data for each species to use in binomial species distribution models. The BBS 118 sampling methodology, combined with one year of data, does not allow estimates of detection 119 probability. To assess the impact of using uncorrected occurrence data, we used an additional 120 validation test based on occurrence data pooled over 5 years (2014-2018; see section 2.5). Due 121 to a large sample size of 5200 points across 104 routes, we expected issues of detectability to 122 have a minimal effect on estimates of species distribution.  129 For habitat associations, we used 30m resolution land cover data from the 2018 Agriculture and 130 Agri-food Canada annual crop inventory (ISO 19130; Table A1). This data set provides detailed 131 information on up to 71 natural and agricultural land cover classes. For each of a subset of 132 ecologically relevant classes, we calculated the proportional area within a 200m and 1km radius.

133
The resulting 16-and 400-hectare areas allowed us to fit multi-scale species distribution models 134 that incorporated habitat associations at both the territory and landscape scale. 135 We included the following natural land cover classifications in our analyses: water, 136 wetland, shrub, forest, and grassland. Forests consisted of deciduous (broadleaf), coniferous, and 137 mixed-wood forest cover. For agriculture, we separated land covers into pasture, including grass 138 and forages (e.g., hay), and arable crop (hereafter 'crop') following Wilson et al. (2017). Native 139 grassland only represents around 1% of the land cover and therefore it was grouped with pasture.

140
Crops included three main typescorn, soybean, and cerealsas combined they make up >90% 141 of the crop land cover in the region. Finally, we included urban land cover which consists of any 142 human-made structures, such as cities, industrial parks, and roads.

143
To account for general effects of climate and topography, we extracted breeding season 144 climate variables from the CHELSA Bioclim database and three terrain variables (elevation, 145 slope, aspect) from the Canadian Digital Elevation Model (Table A1). The CHELSA database 146 includes seasonal temperature and precipitation variables averaged from 1901-2016 and thus 147 indicates broad-scale spatial differences in climate. We chose the two most relevant climate 148 variables for the breeding season: mean temperature of the warmest quarter (BIO10), and 149 precipitation of the warmest quarter (BIO18). Across the study region, average breeding season 150 temperature varied by 12.8°C (range: 11.0-23.8°C) and precipitation by 350 mm (170-520mm), 151 with the southwest being warmer and drier on average. From the CDEM, we derived aspect in 152 radians and then converted it from a circular variable using a negative cosine transformation, 153 such that a value of 1 is south, -1 north, and 0 either east or west. conducted along roads, we included urban cover (which includes road surfaces) at the 1km scale 163 only to avoid overpredicting presence along roadways. For forest-dependent species, we tested 164 models that combined all crop classes into one class 'crops' and separated forest cover into its 165 three subclasses: deciduous, coniferous, and mixed-wood. For grassland species, we included 166 forest cover as a general variable and retained individual crop types (cereals, corn, soy) as we 167 expected grassland species to be more responsive to variation in crop type (Table A1). 168 Prior to model fitting, both presence and absence points were thinned to reduce the 169 potential influence of spatial autocorrelation. We calculated pairwise distances among each point 170 and then randomly removed points until no two were within 1km of each other. This process was 171 replicated 10 times for both presences and absences of each species and the single replicate that 172 retained the most combined data points was selected. 173 We used ensemble species distribution models (ESDMs) to predict current habitat    We built SSDMS for: 1) grassland specialist SAR and SOC, 2) forest specialist SAR and 241 SOC, 3) all SAR and SOC species combined, and 4) all species that occur within the region (i.e., 242 total species richness). For all species (4), ensemble models were built for all terrestrial birds 243 observed on the BBS routes, excluding non-native species. We also excluded species observed  (Table A2). For consistency with focal species, the same model structure, binary 246 threshold estimation, and fit evaluation were used (sections 2.4-2.5).

247
Predicted distribution maps were overlaid with those of the 8 focal species and summed 248 to estimate species richness. Variable contribution for species richness was calculated as the 249 average contribution across all species and its associated standard error. We also fit GAMs 250 describing predicted species richness with the land cover, terrain, and climate predictors that 251 were identified as having the greatest variable contribution to evaluate the habitat associations 252 that shape species richness.    (Table A3). Model comparisons among the multi-, fine-, and coarse-scale models 285 clearly indicated an advantage to using a multi-scale approach as it was the best fitting model 286 across all species (Table A3) Table A4). region, yet suitable habitat was still identified within the more intensive agricultural area in the 295 south, particularly near Lake Huron which represents a mosaic of field crops, forest patches, and 296 hedgerows ( Figure 2). Grassland (including pasture, forage, and hay) had the strongest influence 297 on the distribution of bobolink, Eastern meadowlark, and Savannah sparrow, but had little 298 influence on horned lark and killdeer (Table A5). Instead, these latter two species responded 299 positively to row crops, particularly cereals but also soybean for horned lark (Table A5). Forest 300 specialists primarily responded to forest availability, with particularly strong and positive 301 associations with deciduous, and for wood thrush, mixed-wood forest cover (Table A5). The 302 influence of scale also differed among species, highlighting the importance of a multi-scale 303 approach. For example, while grassland species were generally associated with grassland and 304 agriculture land-use at the territory scale (200m radius), horned lark responded most strongly to 305 agricultural land cover at larger spatial scales (1km ; Table A5). difference from the mean species richness at zero. For all associations, the X-axis is proportion of land 348 cover, except for bios 10 (temperature) and bios 18 (precipitation) which are displayed in 1/10 th degrees 349 Celsius and mm, respectively. The shaded band represents the 95% CI and the distribution of points is 350 depicted along the X-axis. 351

Overall species richness 352
Species richness of the overall bird community (76 species) was predicted to reach ~50-60 353 species in multiple mid-latitude areas from east to west ( Figure 5A). We quantified areas with 354 high species richness, defined as containing at least 75% of the total species richness or 75% of 355 the focal species. This demonstrated that greater than average richness for both groups was 356 predicted in regions near Lake Huron, west of Montreal and Ottawa, as well as along the St.

357
Lawrence River ( Figure 5B). These regions contain the necessary land cover, terrain, and climate terrestrial birds with at least 50 observations (76 species), and B) areas with richness greater than the 3 rd 364 quantile (>75% of maximum). For the latter, green indicates areas where species richness is greater than 365 the 3 rd quantile for either the focal species-at-risk (≥ 6 species) or total species richness (≥ 41 species), 366 while yellow depicts areas where high species-at-risk and total richness align. for Eastern meadowlark to 42,992 km 2 for least flycatcher (Table A6). without protection (green). The prioritization selected at least 25% of the distribution for each feature in 382 the most cost-effective manner (i.e., minimum required area). Features were the distribution maps for 383 each species shown in Figure 3 and the upper quartile for total species richness shown in Figure 5A.  As forests are the primary land cover in the region, the potential distribution of forest 417 specialists tended to be considerably greater than that for grassland species. When using a 418 minimum set prioritization with a goal to protect a percentage of each species distribution, the 419 result is a higher area target for forest species. For example, a 25% area goal required only 6,068 feasible; yet these are the regions needed to reach the area targets for grassland species while 427 also accommodating forest species. In these cases, the objective should be to work with 428 landowners to manage landscapes in a manner that maximizes the diversity of both avian groups 429 to the extent possible without compromising agricultural yields.

430
It is important to note that we assessed fundamental habitat suitability, or the potential 431 distribution that each focal species could inhabit, rather than the realized distribution  Data and code will be made available in a public repository upon manuscript acceptance.