Because of the importance of the Central Valley of California, USA, to shorebirds along the Pacific Flyway, conservation investments have been made in the area’s agricultural fields and managed wetlands. Increasingly, landowner incentive programs are being used to deliver shorebird habitat, presenting opportunities to answer remaining questions about which implementation strategies and field management practices are most effective at attracting birds. To provide management guidance for these investments, we collected and analyzed 5 years of data (2014-2018) on shorebird abundance in flooded rice fields enrolled in a dynamic habitat incentive program. The program incentivized flooding in fallow and post-harvest rice fields in seasons when habitat is relatively sparse, specifically the early fall and late spring. Across nearly 9,000 field observations over 5 years, we explored the relationship between abundance and density (birds/ha) of shorebirds and vegetation cover, soil clay, landscape-level flooding, and local flood timing, duration, and depth. We observed more shorebirds in fields that were approximately 50% flooded, had water depths of 5-10 cm, and had minimal rice straw or stubble cover, with strong or very strong evidence for each of these relationships. We found that the timing of habitat provisioning was important, with moderate evidence that earlier fall flooding and strong evidence that the duration of fall flooding was associated with higher shorebird density. We observed lower shorebird densities in locations with ample flooded rice habitat in surrounding areas, potentially because shorebirds spread out across the landscape. We found very strong evidence that flooding consistency, either at a site that was continually flooded over many months or a site that had been flooded in previous years, was associated with higher shorebird density. Soil clay content was associated with decreased observed shorebird density, potentially through its influence on the ability of shorebirds to forage for invertebrate prey. These results suggest best practices for shorebird habitat creation in agricultural landscapes, providing important information for conservation and population recovery.

Bird surveys

We monitored enrollment fields distributed across the Sacramento Valley (Figure 1) following the sampling design and survey protocol described in Golet et al. (2018). Briefly, we selected a set of spatially balanced random survey point locations along the edges of fields using the generalized random tessellation stratified (GRTS) sampling methodology (Stevens and Olsen 2004). The rice fields were composed of multiple checks (or patties), and we established only 1 survey point in any given check. From these points, we conducted surveys of semi-circular areas within the fields that had a fixed radius of 200 m, or less when they intersected earthen levees or field boundaries (Reiter et al. 2011, Strum et al. 2013). While the number of survey areas within an enrollment field depended upon its size, we tried to ensure a sampling density of 1 survey point per 12 ha.

Trained biologists conducted surveys and established a common understanding of all data entry queries during group training sessions led by a senior biologist in the field at the beginning of each year. In addition, surveyors participated in online training (from eBird, https://ebird.org, and Sibley Guides, https://www.sibleyguides.com) to hone flock counting skills. Many of the biologists that collected the data worked on the project for multiple years, which brought consistency to data collection. For at least 2 minutes, and as long as necessary to count all birds present, surveyors scanned the survey areas with binoculars and spotting scopes to record all visually detectable waterbirds. Surveyors endeavored to count all birds in the water and on the field edges, including those that flew off or landed during counting. Except for aerial foragers, we did not record birds that flew by. Monitoring started when enrollment started. We monitored enrollment fields every 5 to 7 days during the enrollment period, and continued until the fields were no longer shallowly flooded. A total of 451 field-seasons were enrolled in the BirdReturns program during the study period, where a field-season refers to a particular field enrolled in a given season and some fields were enrolled in multiple seasons. We surveyed fields an average of 5 to 10 times per year (Table 1).

Whenever possible, we identified shorebirds to species. Exceptions in species-level identifications include dowitchers and yellowlegs; we did not distinguish between short- and long-billed dowitchers and greater (Tringa melanoleuca) and lesser (T. flavipeswere) yellowlegs. We recorded 24 unique taxa (Figure 2). Following the work of many others studying waterbirds in rice (e.g., Elphick 2008, Sesser et al. 2018), we did not calculate detection probabilities. We assume detection probabilities were high owing to minimal vegetation and open sightlines; although, some species were likely undercounted because of their secretive behavior (e.g., Wilson’s snipe [Gallinago delicata]) or small size (e.g., least sandpiper [Calidris minutilla]).

Site-level and landscape characteristics and covariates

At each site, we measured 2 types of characteristics: site-level (local) and extended temporal and landscape variables (landscape; Table 2). The surveyor visually estimated local characteristics during the same time windows that they conducted bird counts. Local characteristics included the fraction of the survey area that was either flooded, moist, or dry (summing to ≤100%) and the fraction of the survey area that was bare earth, non-rice vegetation, standing dead rice stubble, or horizontal (cut down) rice straw (which could sum to greater than 100% because of overlapping categories). Following methods established in previous studies that characterized water depth in rice fields (Sesser et al. 2018), we determined depth (to the nearest 2.5 cm) by viewing 2 stakes in the survey area that were painted with 5-cm-wide color bands. The first stake was positioned 50 m from the observer point location and the second was 200 m from the observer location, at the distant boundary of the survey area. We used the mean of these 2 depth measurements to represent the depth in the enrollment field (as described in Golet et al. 2018). We placed the stakes in the survey area after exploring the region for locations that were representative of a particular field’s elevation. A pair of properly placed stakes are sufficient to characterize depth in Sacramento Valley rice where about 66% of the fields are set up to grow only rice and have little to no slope. Fields that rotate between row and field crops typically have slopes of only 0.05% to 0.1% (University of California 2023).

Within each survey area, we measured the average percent clay. Using a 100-m resolution, gridded map of the fraction of clay in soil (Ramcharan et al. 2018; derived from SSURGO data), we averaged the amount of soil within a survey area after averaging percent clay across soil depth layers.

We calculated landscape spatial variables from satellite images taken by the Landsat 8 Operational Land Imager and Thermal Infrared Sensor using a classification model to estimate the amount of flooded habitat that was available in the Sacramento Valley (Reiter et al. 2015, 2018a, 2018b). Because cloud cover can mask imagery of the Earth’s surface, we generated estimates of flooding for days with cloud-filled images using models relying on the average probability of water in that pixel in that month based on the previous 10 years, the normalized difference water index value from Moderate Resolution Imaging Spectroradiometer (MODIS), and the proportion of open water in the rest of the Central Valley Joint Venture planning basin (Reiter et al. 2018a).

Within the Landsat time window in which we surveyed birds, we considered the amount of flooded rice and wetlands in the larger landscape within a 2-km, 5-km, and 10-km radius around the survey area. These covariates allowed us to consider whether the amount of flooded rice or wetland influenced shorebird use of the survey areas. To explore whether fields or landscapes that were predictably flooded across years would attract more birds, we also considered flooded rice and wetlands in the larger landscape in the preceding year and the preceding 3 years. Given that the amount of flooding within the broader landscape in a given year is highly correlated (r > 0.8) to landscape flooding in the preceding year and preceding 3 years, we considered only landscape-level variables from the year in which we made observations. The proportion of the landscape that was flooded within the survey area in the months or years preceding enrollment was not highly correlated across years and thus could be included in the models together with contemporaneous flooding. For spring enrollment only, we considered flooding in October through January leading up to the spring enrollment. To calculate this variable, we averaged the flooded area within the survey area across available 16-day images from October to January leading up to the bird observation.

In addition to landscape characteristics, we tested whether the total number of shorebirds in a survey area was influenced by the start date of enrollment, duration of enrollment, or size of enrollment fields. Across a season, all survey areas within an enrollment field shared the same values for these 3 variables. Finally, we included the month of the bird observation to account for seasonality in shorebird use of enrollment fields. In the fall, there were few observations prior to September so we pooled observations from August and September (as if they were the same month: September). We also included an interaction between observation month and landscape-level rice and wetland flooding because we expected that seasonality would affect how shorebirds use the larger landscape.

None of the variables in the spring or fall had pair-wise correlation coefficients greater than r = 0.6. In fall and spring, respectively, only 8 and 5 pairs of coefficients had positive or negative correlations greater than r = 0.27 (Figure S1, available in Supporting Information). In models that did not include quadratics or interaction terms, variance inflation factors for all covariates in spring and fall models were <1.75.

Data management

Data were curated by The Nature Conservancy staff until transfer to E. Conlisk for derivation of remotely sensed landscape cahracteristics and analysis.