Sussman A.L., Kenow K.P., Luukkonen D.R., Monfils M.J., Mueller W.P., Williams K.A., and Zipkin E.F. In prep. Combining models to identify waterbird hotspots in the Great Lakes. Journal of Fish and Wildlife Management.
Waterbird species play an important role in ecosystems and are known indicators of ecosystem health. Most waterbird species exhibit patchy distributions with high aggregations that vary throughout the year, making it difficult to identify spatial patterns; however, waterbirds rapidly react to changes in the environment providing beneficial information about aquatic ecosystems and species. Detecting hotspots is an effective way to identify species-specific patterns and simultaneously inform decisions regarding habitat and ecosystem protection. The Great Lakes, and surrounding areas, offer many resources for both humans and wildlife. The region provides unparalleled habitat for many wildlife species, including many waterbirds throughout the year during migration, wintering, and breeding seasons. There are several methods to identify hotspots, many of which are arbitrary in selecting metrics or thresholds and may thus provide incongruent results. One solution is to combine methods, which may result in more accurate hotspot estimates than with any single analysis framework. We selected and combined two hotspot models commonly used for waterbird hotspot analyses to identify species-specific hotspots in the Great Lakes: one spatial method, Getis-Ord Gi*, and one non-spatial parametric method, hotspots conditional on presence. Our objective was to delineate a single hotspot value per location (i.e., 5 x 5 km grid cell), using a post-hoc integrated hotspot modeling approach, for each of the selected species and species groups. Our combined model showed Lake St. Clair and western Lake Erie had more hotspots than expected for half the species analyzed, which is likely due to the shallow depths of these two lakes. Lakes Michigan and Huron exhibited a higher proportion of hotspots than expected for long-tailed duck. The difference in pattern between the diving/sea duck group and long-tailed duck suggests that diving and sea ducks exhibit very different distributional patterns and should (given sufficient data) be split out and analyzed separately. Uneven sampling across the Great Lakes region affects confidence that locations are or are not true hotspots, but using a combined hotspot approach can help alleviate some concerns associated with limited data availability. Our integrated modeling approach increases the consistency of hotspot detection, increasing accuracy in assessments of waterbird spatial patterns.
- alldata_attributed_v04122017.csv - raw aerial visual observations
- gl_effort-corrected-counts.csv - standardized effort-corrected counts
- ALLSP_conditional.csv - model results for the hotspots conditional on presence approach
- ALLSP_gstat.csv - model results for the Getis-Ord Gi* approach
The raw observations and transect shapefiles are publicly available through the Midwest Avian Data Center (MWADC), a regional node of the Avian Knowledge Network hosted by Point Blue Conservation Science: https://data.pointblue.org/partners/mwadc/. The observation data were imported into the opensource relational database management system PostgreSQL v9.5.0, with the PostGIS extension v2.2.1, using GDAL ogr2ogr (http://www.gdal.org/ogr2ogr.html). Shapefiles of survey transects were downloaded from the MWADC and were also imported into PostgreSQL. Data were QA/QC'd and standardized to account for differences in survey methods in PostgreSQL. Using the RPostgreSQL library (Conway et al. 2017), the data were called directly from PostgreSQL within R (RStudio v1.0.136). The individual models and the combined model were conducted in R.
combined_hotspot_analysis.R - R code for the integrated hotspot modeling approach, which uses the hotspots conditional on presence and Getis-Ord Gi* models.