Net effects of field and landscape scale habitat on insect and bird damage to sunflowers

Agriculture-dominated landscapes harbor significantly diminished biodiversity, but are also areas in which significant gains in biodiversity can be achieved. Planting or retaining woody vegetation along field margins can provide farmers with valuable ecosystem services while simultaneously benefitting biodiversity. However, when crops are damaged by the biodiversity harbored in such vegetation, farmers are reluctant to incorporate field margin habitat onto their land and may even actively remove such habitats, at cost to both farmers and non-target wildlife. We investigated how damage by both insect pests (sunflower moth, Homoeosoma electellum) and avian pests to sunflower (Helianthus annuus) seed crops varied as a function of bird abundance and diversity, as well as by landscape-scale habitat. Surveys for insect damage, avian abundance, and bird damage were carried out over two years in 30 different fields on farms in California’s Sacramento Valley. The mean percentage of moth-damaged sunflowers sampled was nearly four times higher in fields that had bare or weedy margins (23.5%) compared to fields with woody vegetation (5.9%) and decreased in both field types as landscape-scale habitat complexity declined. Birds damaged significantly fewer sunflower seeds (2.7%) than insects, and bird damage was not affected by field margin habitat type, landscape-scale habitat variables, or avian abundance, but was significantly higher along field edges compared to ≥ 50m from the field edge. Avian species richness nearly doubled in fields with woody margin habitat compared to fields with bare/weedy margins in both the breeding season and in fall. These results indicate that the benefits of planting or retaining woody vegetation along sunflower field margins could outweigh the ecosystem disservices related to bird damage, while simultaneously increasing the biodiversity value of intensively farmed agricultural landscapes.

electellum) and avian pests to sunflower (Helianthus annuus) seed crops varied as a function of 23 bird abundance and diversity, as well as by landscape-scale habitat. Surveys for insect damage, 24 avian abundance, and bird damage were carried out over two years in 30 different fields on 25 farms in California's Sacramento Valley. The mean percentage of moth-damaged sunflowers 26 sampled was nearly four times higher in fields that had bare or weedy margins (23.5%) 27 compared to fields with woody vegetation (5.9%) and decreased in both field types as 28 landscape-scale habitat complexity declined. Birds damaged significantly fewer sunflower seeds 29 (2.7%) than insects, and bird damage was not affected by field margin habitat type, landscape-30 scale habitat variables, or avian abundance, but was significantly higher along field edges 31 compared to  50m from the field edge. Avian species richness nearly doubled in fields with 32 woody margin habitat compared to fields with bare/weedy margins in both the breeding 33 season and in fall. These results indicate that the benefits of planting or retaining woody 34 vegetation along sunflower field margins could outweigh the ecosystem disservices related to

Introduction 41
In the face of significant losses of both diversity and abundance of avian species (Rosenberg et  birds (Heath et al. 2017). 57 al. 2019). Sunflowers grown for seed are valued at five to ten times that of the commercial oil 104 crops for which they are used (Long et al. 2019), and growers therefore have a low threshold 105 for damage. All sunflower fields in our study were grown for the same seed company and 106 therefore were grown using the same field-management practices. This study was conducted 107 within conventional fields (i.e. non-organic fields), but no growers reported utilizing insecticides 108 on their fields over the duration of this study. 109 The predominant insect pest for sunflowers in North America is the sunflower moth 110 (Homoeosoma electellum). Female sunflower moths lay eggs among the florets of sunflowers in 111 early bloom, and eggs take 2-5 days to hatch. After hatching, larvae remain on the face of 112 flowers for 8 days before boring into the developing seeds where they can cause losses of 30-113 bird damage to sunflowers is often concentrated to the edges nearest to habitat that can act as 116 shelter for birds. For example, in Israel, bird damage within a field was highest in areas close to 117 trees (>5m in height), but increasing the number of trees within a 1-km radius of fields was not 118 associated with higher damage (Schäckermann et al. 2014), suggesting that presence of habitat 119 along edges of crops prone to bird damage is more important than the presence of habitat in 120 the landscape overall. 121 Field-and Landscape-Habitat Complexity woody margin habitat and seven fields with bare or weedy field margins in 2014, and from 12 124 complex fields and 5 simple fields in 2015, for a total of 30 fields sampled. To quantify local 125 (field) habitat complexity, we collected data on the height, width, and number of canopy layers 126 of field margin vegetation at 5 locations along each transect (see Heath et al. 2017 for details). 127 To quantify and incorporate landscape habitat complexity into our study design, we selected 128 fields at varying distances from natural habitat, which in our study area consists mainly of 129 remnant and restored riparian areas ( Figure 1). We used pre-existing habitat data for our study 130 area (CA DWR 2008, Geographic Information Center 2009), and added by hand any trees within 131 800m of each transect that were not included in the existing dataset (e.g. trees lining 132 driveways, trees around homesteads). To calculate the distance to riparian area, we used 133 ArcGIS 10.1 (ESRI 2010) to create a distance raster that encompassed the entire study area by 134 using the Euclidean distance algorithm. We used the riparian vegetation GIS dataset (habitats 135 classified as native riparian, blue oak woodland, valley foothill riparian, fresh emergent 136 wetland, saline emergent wetland, and valley foothill riparian) as the 'source' input for the 137 algorithm and set the output grid cell size to 10 meters. Each field's transect center point was 138 then buffered by 50 meters, and we calculated the distance from each grid cell within the 139 buffer to the nearest riparian vegetation polygon. The mean distance for all cells within each 140 buffer was calculated as the distance value for each field. We also calculated the mean 141 We quantified both bird and insect damage by visually inspecting each of ten sunflowers 161 within each sampling area. Sunflowers were chosen by reaching out to select a plant stalk, so 162 the seed-bearing area of each plant was not seen until after the plant was selected (most 163 sunflowers were at or above head-height for observers). Observers moved a few steps along 164 and between rows to select each new flower. Bird damage was characterized by missing seeds. 165 We were careful to avoid classifying wind-damaged seeds that had been rubbed off of larger continuous areas of the sunflower head, whereas seeds removed by birds were in patchy 168 sections or removed singularly. Wind-damaged seeds were also often seen whole on the 169 ground underneath the plants. Insect damage was characterized by an area of visible frass 170 (insect excrement and webbing) on the surface of multiple sunflower seeds. Seeds under the 171 frass were often shrunken or visibly damaged. All areas that were under frass were classified as 172 To estimate the percent of seeds on each sunflower that were damaged, we used a pre-cut 174 circular piece of galvanized steel chicken-wire that was marked to allow for easy measurement 175 of the flowers. Sunflower heads were classified into different size classes based on the diameter 176 (to the nearest 1.3 cm, or 0.5 inches) of the seed-bearing area on each plant. We then 177 estimated the number of hexagons on the wire (to the nearest ¼ hexagon) that was damaged 178 by birds or damaged by insects on each sunflower head. Using the flower circumference and 179 the known area within each hexagon of our grid, we were then able to calculate the percent of 180 each sunflower head that was damaged by birds, and the total that was damaged by insects. To 181 estimate yield, damage from insects and damage from birds were summed for a total percent 182 damage to each sunflower, since both types of damage result in a direct loss of yield for 183

growers. 184
We sampled from 10 sunflowers at distances from 0m to 200m from the field edge. In 2014, 185 we collected observations of both insect and bird damage from each site at 0, 10, 20, 30, 40, 50, 186 75, 100, 150, and 200m from the field edge. In 2015, we collected observations from each site to close to 0 at distances beyond 50m, and that insect damage was largely unchanged by 189 distance from the field edge (see Figure 2). Estimates for insect and bird damage in 2015 were 190 taken from sunflowers within exclosures and from sunflowers that were approximately 10m 191 from the exclosures (parallel to the field margin), but only data from non-enclosed sunflowers 192 was used in our comparative analysis of insect damage. 193

Bird counts 194
We conducted four bird surveys at each site, two in summer (June 9-July 2) and two in 195 fall (August 5-September 16). All bird surveys were conducted by trained observers and timed 196 to coincide with sunflower bloom in the summer (when sunflower moths typically lay eggs on 197 the flowers), and immediately prior to the seed harvest in the fall. All counts were conducted 198 between dawn and 10am and were not conducted in very cold (<3C) or very hot weather 199 (>24C), in high winds or heavy precipitation. Counts were also re-scheduled if there were any 200 farm workers or machinery in our focal field. We conducted two counts per visit at each field: 201 one to quantify the birds utilizing the field margin habitat, and another to quantify the birds 202 utilizing the field interior. These methods provide relative values for comparing inter-site bird 203 communities. To count birds utilizing field margin habitat, observers walked a 200m transect 204 slowly over 10 minutes, counting all detectable birds by sight or sound within 20m of the field 205 margin, but not within the field itself. To count birds utilizing the field interior, observers 206 returned to the mid-point of the transect, allowed five minutes for birds to settle, and then 207 conducted a 10-minute point count focused only on birds that were observed within the field. 208 We counted all birds detected within each field because each species was assumed to have 209 similar detectability in all fields, since sunflowers were at similar levels of maturation and height at the time of each count, and since fields were all of a similar size. We used different 211 methods for the edge and interior transects to maximize our detection of birds utilizing each 212 type of habitat. While these methods may result in counting the same individual in both 213 habitats on the same visit, this is relevant since birds at our study sites were regularly observed 214 using both the field margin and field interior habitats. 215 216

Statistical Analyses 217
Because the variables describing field margin habitat (height, width, and number of 218 vegetation layers) were highly correlated, we used a Principle Components Analysis (PCA) to 219 reduce these into two orthogonal axes that explained over 95.5% of the variance among them. 220 The two axes, PC1 and PC2, were included as predictor variables in our candidate models for 221 sunflower damage and for bird abundance and richness. PC1 explained 86.2% of the variability 222 among habitat variables and was negatively associated with all three variables, whereas PC2 223 was positively associated with habitat width and height, and negatively associated with habitat 224 layers. Therefore, if PC1 is a positive predictor of damage, we would expect less damage at sites 225 with habitat that is taller, wider and has more layers (because of the inverse relationship). If 226 PC2 is a positive predictor of damage, we would expect less damage at sites with more habitat 227 layers and more damage at sites with taller/wider habitat. We also found collinearity among 228 the predictor variables for landscape-scale habitat complexity, so constructed separate models 229 for each landscape-scale habitat complexity variable. Model selection revealed that the 230 variable for mean distance to natural habitat was most parsimonious in our sunflower damage 231 models (Tables S1-2), so we present the results from that model in the main text of this paper. For both damage categories, we used generalized linear models with a negative 237 binomial family of errors to analyze our data on percent damage to sunflowers in R v.3.3.1 . 238 Sunflower moth damage and bird damage were analyzed in separate models. For our bird 239 abundance and richness data, we ran eight separate linear regressions for avian species 240 richness and abundance along the field edge and within the field interior for data collected in 241 summer and in fall. For all analyses, we included as predictor variables in our maximal models 242 the continuous variables for the distance from the nearest riparian habitat, PC1, and PC2, as 243 well as the categorical variable for whether the field had a weedy or bare edge (simple edge 244 habitat) or had woody field margin habitat (complex edge habitat). We simplified the maximal 245 models by removing interactions, then main effects, until no further reduction in residual 246 deviance (measured using Akaike's Information Criterion) was obtained. For all regression 247 analyses, we considered candidate models with ΔAIC  2 and chose the most parsimonious 248 model. 249

Economic Estimates 250
We used published data on the range and mean sunflower yields and economic value for the

Sunflower damage 266
Sunflower moth damage was almost four-times higher at sites with bare or weedy field margin 267 habitat (23.46 ± 1.41%) compared to sites with woody vegetation along field margin habitat 268 (5.89 ± 1.16%; z= 7.12, p < 0.001). There was a slight decrease in sunflower moth damage as 269 with PC1. Since PC1 was negatively associated with all three measures of field margin habitat 301 complexity (habitat height, width, and number of canopy layers), our results predict that as 302 field margin habitat becomes more complex, avian richness and abundance along field edges 303 increased. For summer field interiors, avian species richness was uncorrelated with PC1 (t = -304 1.83, p = 0.08, Figure 3c). None of our predictor variables were retained in the model for avian 305 abundance within the field interior in summer. 306 In the fall, field edge richness, field edge abundance, and field interior abundance were 307 all positively associated with increasing field margin habitat complexity. Field edge avian 308 species richness was negatively associated with PC1 (t = -9.82, p <0.001, Figure 3e), PC2 (t = -309 2.80, p <0.01) and average distance to nearest riparian habitat (t = -2.30, p = 0.03). Avian 310 abundance at the field edge in fall was negatively associated with PC1 (t = -23.40, p <0.001, 311 Figure 3f). Avian species richness in field interiors during the fall was not correlated with edge 312 complexity or distance to riparian habitat. Avian abundance was significantly higher at sites 313 with weedy/bare edges, compared to sites with woody vegetation (mean of 109 more birds at 314 simple sites; t = 2.33, p = 0.03), but increased in both bare/weedy and woody vegetation field 315 margin types with increasing field margin habitat complexity ( negatively association with PC1; t 316 = -2.31, p = 0.03, Figure 3h). 317 at sites without field margin vegetation, while bird damage was not driven by field margin 322 habitat. Furthermore, within sunflower fields across all distances from the field margin, 323 sunflower moth damage was significantly higher than bird damage, and was therefore the main 324 source of yield loss for sunflower growers in our area. The pest control service benefits that 325 farmers receive from field margin vegetation therefore outweigh the potential ecological 326 disservices associated with bird damage to sunflowers. In fact, bird damage at our 30 fields was 327 similar across sites with and without field margin habitat. Our results also indicate a clear 328 benefit for biodiversity, with significantly higher species richness and avian abundance along 329 field edges that had woody habitat. Combined, these results support the assertion that Our model selection process revealed that the distance to nearest natural vegetation 357 was the strongest of our landscape-scale predictors of sunflower damage (Table S2) We thank the landowners and growers who provided access and information for this study 411   habitats; and c) percent seeds damaged by birds as a function of the distance of sampling 606 points within each field from the nearest field margin. The mean economic yield per hectare is 607 shown as a secondary y-axis and applies to all three panels. 608 damage to sunflower seeds using the distance to nearest natural habitat as a measure of 625 landscape-scale habitat complexity. A principal components analysis was used to consolidate 626 local habitat complexity variables into two orthogonal axes (PC1 and PC2). Field margins for 627 each site were categorically defined based on the presence or absence of woody vegetation 628 along the field margin. The 'Distance into Field' measure is the number of meters within the 629 field for each sampling location from the nearest field edge.