The debate over the best agricultural practices for biological conservation often focuses on the degree to which agricultural lands should be interspersed with desirable habitats versus protecting lands entirely from production. It is important to understand the benefits agriculture provides for wildlife because it is consuming an increasing proportion of the landscape. We evaluated the nesting ecology of breeding ducks within a mosaic of flood-irrigated conservation areas and agricultural lands in hay production. We assessed how habitat features at two spatial scales across these lands were related to nest site selection, nest density, and nest survival of multiple duck species. Birds selected nest sites with higher visual obstruction, a higher proportion of shrubs around the nest, and less bare ground, but we did not detect evidence of selection per se at larger spatial scales. Nest density was marginally higher along linear features, including irrigation ditches and riparian stretches, but nest survival remained similar across land-use types and habitat features. This system is representative of many agricultural landscapes around the globe and highlights the ways agroecosystems can be managed to maintain habitat suitability for wildlife on working lands.

Study System

We studied a system of flood-irrigated basins in north-central Colorado, USA, to evaluate duck reproductive success across agricultural working lands (Figure 1). The North Platte Basin (hereafter North Park) is a high-elevation (2500 m on average) intermountain basin characterized by sagebrush (Artemesia spp.) steppe and riparian corridors used as sources of water to flood irrigate hay meadows (by diverting water into irrigation ditches). The Intermountain West of North America spans 11 states and is comprised of many of these high-elevation basins associated with river and groundwater-fed wetlands. While many are still associated with flood irrigation, some have predominantly transitioned to sprinkler-based irrigation systems to use water more efficiently (e.g., the San Luis Valley of Colorado). Agricultural production is typically comprised of large cattle ranches that also actively produce high-quality, flood-irrigated hay that is harvested each year. In North Park, harvested meadows consist primarily of Timothy hay (Phleum pretense), and are flooded in May, dried anywhere from July to August, and then harvested from July to September. Because of the short growing season, a single cut of hay each year is typical. The system also has public land parcels along riparian areas that are spared the annual harvest of typical agricultural operations, primarily Arapaho National Wildlife Refuge (NWR). This NWR was created in 1967 to benefit migratory and breeding ducks as mitigation for the conversion of high-quality duck breeding habitat in the Prairie Pothole Region of North America to high-intensity agriculture production in the 1960s and 1970s (Doherty et al. 2018). The NWR flood-irrigates wet meadows that are not cut, and that typically exhibit more diverse vegetation communities than Timothy hay meadows, including forbs, sedges, rushes, and grasses interspersed by small areas of greasewood shrubs (Sarcobatus vermiculatus) and sagebrush. In addition to the NWR, there are also state wildlife areas (SWAs) on which managers flood irrigate to create wetland habitat, as well as waterfowl management areas (WMAs) managed by the Bureau of Land Management (BLM) specifically for breeding ducks. Wetland habitats on the parcels of public land included in the study are comprised of large water storage reservoirs with variable amounts of submerged aquatic vegetation, basin wetlands with rings of emergent vegetation, and irrigated meadows consisting of graminoids and occasionally robust emergent vegetation (e.g., cattails [Typha spp.] and bulrush [Scirpus spp.]).

Data Collection and Processing

We searched systematically for duck nests from 20 April until 1 August 2018-2023. Study sites included five private ranches on which agricultural production was predominantly focused on cattle and hay. Additionally, we included Arapaho NWR, Lake John SWA, and Hebron WMA, which are multi-use parcels of public land spared from extractive agricultural production but subject to light cattle grazing. We searched randomly selected nest plots across land-use types in addition to searching opportunistically between plots. After overlaying a grid with 8-ha grid cells on the wet meadows of Arapaho NWR using a geographic information system (GIS; Esri ArcGIS Pro 2.8.0), we randomly selected 16 square plots to sample portions of the large expanses of the irrigated meadow. However, plots on private lands followed the natural boundaries of hay meadows, which were often smaller and more easily definable (Figure 2). As a result, plots on private land varied in size and number, but we still delineated them based on landscape features in a GIS and randomly selected a subset to search each year. Access to ranches also varied across years, which altered the number of plots we could search. The number of plots we searched on private ranches varied from five during a pilot year to 131, and plots ranged in size from 0.14-35.83 ha, averaging 6.44 ha. Additionally, we randomly selected 500-m length sections of riparian areas (n=40) and irrigation ditches (n=25) across the study area, searching within a 200-m buffer of the edges, and systematically searched the perimeter of all basin wetlands out to a radius of 200 m. We display an example ranch in Figure 2, which shows the layout of selected plots of several wetland habitats. We report the total area (ha) of each habitat type in the study area in Table 1 alongside the area of each habitat in our sampling frame, including land associated with accessible ranches and focal parcels of public land. Finally, we report the area within that sampling frame that we searched annually to illustrate which habitats were represented in our search plots relative to the area available. We searched plots 1-5 times per year and used a combination of rope drags (on foot; Higgins et al. 1969) and systematic foot searches to flush laying and incubating hens off of the nests, marking the location with a global positioning system (GPS) device. We recorded search effort each year (date searched and the number of people searching a plot) and used a GIS framework to compute the area in ha of each plot, whether the plot contained or its centroid was within < 200 m of a basin wetland, and the composition of rasterized habitat classes within each plot based on the 2021 National Land Cover Database (NLCD) layer. We identified the species incubating each nest as the hen flushed and used the size and color of the eggs to verify the identification. We candled several eggs in each nest to calculate the nest initiation date by backdating from the date the nest was located based on the embryonic stage of development and the number of eggs in the nest (Klett et al. 1986). As incubating hens typically cover their eggs with down feathers upon leaving the nest, we also covered eggs after each nest visit and placed two pieces of grass across the top of the nest in an “x” shape to determine whether the hen returned to the nest or abandoned after disturbance. We monitored each nest approximately every five to seven days, noting its incubation status, hen presence or absence, full clutch size, and ultimately nest fate.

Regardless of whether a nest failed (i.e., all eggs were eaten by a predator or abandoned by the hen) or was successful (i.e., at least one egg hatched), we conducted vegetation surveys on the estimated or actual hatch date (McConnell et al. 2017). We calculated the hatch date based on the stage of embryonic development of the eggs during each nest visit and the average incubation time for each species. For successful nests, we conducted surveys the day after ducklings left the nest. Vegetation surveys occurred at the nest bowl and at four randomly selected points within a 200-m radius of the nest bowl to evaluate fine-scale (i.e., third-order; Johnson 1980, Eichholz and Elmberg 2014, Kaminski and Elmberg 2014) metrics of habitat selection. Surveys included visual estimation of percent cover within a 1-m Daubenmire frame (Daubenmire 1959). We estimated the percent cover of bare ground, litter (dead vegetation from the previous growing season), water, grasses, forbs, shrubs, sedges, and rushes, and we allowed the total percent cover to sum to more than 100% because the vegetation was often layered vertically. We also assigned each nest to a categorical habitat type at the time of measurement and measured visual obstruction by noting the lowest decimeter visible on a 1-m Robel pole from each cardinal direction and averaged the four values (Robel et al. 1970). Habitat types were classified based on the dominant vegetation within 200 m of the nest and included riparian, shrub-scrub, emergent marsh (dominated by robust vegetation like cattails), graminoid meadow, graminoid meadow interspersed by shrubs, Timothy hay meadow, and irrigation ditch, which was used when a nest was within 3 m of the inner channel of an irrigation ditch. We separated graminoid meadows from graminoid meadows interspersed with shrubs because shrubs may provide perches for avian predators from which duck nests may be more easily located (Thompson et al. 2012, Coates et al. 2021, Peterson et al. 2022).

We measured broad-scale habitat characteristics using a GIS to evaluate the drivers of nest site selection at a larger scale. We created ~10000 random points across the study area (i.e., within the sampling frame indicated by the delineated boundaries in Figure 1) in all habitats where we consistently searched for nests. We calculated the distance of each random point and nest site to the nearest irrigation ditch, river, open water (i.e., ponds, marshes, or reservoirs), road, harvested hay meadow, and uncut irrigated meadow using NLCD land cover types. We refer to this as the “patch scale.” We also created circular buffers around each nest point and random point with a radius of 200 m. This allowed us to tabulate the number of 30 x 30 m pixels within each 200-m radius buffer plot assigned to each NLCD land cover category, and from that the area in ha of each habitat. We refer to this as “landscape composition.”  We could not include the year as a random effect because the GIS layer was not year-specific, but changes in the patch-level GIS variables used were minimal over the study period. This differing random effect structure precluded the comparison of models of fine-scale vegetation metrics to models of patch-scale metrics, so we considered nest site selection at the two scales separately.