Documenting wildlife–habitat relationships at multiple scales is essential for conservation. Remote sensing datasets and their derivatives (e.g., landcover data) enable efficient multi-scale assessment of ring-necked pheasant (Phasianus colchicus) habitat, albeit with trade-offs among their thematic, spatial, temporal, and/or spectral grains and extents. For example, the National Agriculture Imagery Program provides fine spatial but coarse spectral grain imagery, both important for identifying pheasant habitats. Spatial technologies and datasets relevant to pheasant research are advancing, yet the information on the data sources utilized in research to date is limited. Remote sensing and landcover datasets surveys in pheasant research could help fill information gaps in pheasant–habitat relationships. In this systematic review we filtered 1,110 peer-reviewed pheasant habitat studies to 65 from the Central U.S.A. Temporal trends were tested in the broad use of remote sensing and the selection of remote sensing platforms and data types. Of the selected studies, 26 used remote sensing or landcover data, which were classified by the thematic, spatial, temporal and spectral grains and extents. Remote sensing and landcover data products increased over time, particularly satellite-based landcover products with relatively coarse thematic resolutions (e.g., crops and grassland), moderate spatial grains (e.g., 30-meter), and spatial extents (e.g., smaller than the average U.S.A. county). Remote sensing photography/imagery with multispectral sensors and coarse spectral resolution (e.g., 3 bands with 100 nm width) was also prominent but remained constant over time. We found no evidence of research with remote sensing or landcover data at multiple temporal grains and extents. Several studies lacked scale reporting, potentially limiting our inference. Scale transparency is important due to species selecting their habitat at multiple scales, making findings scale-dependent. Effective conservation requires scale-appropriate strategies. As remote sensing advances, opportunities for ring-necked pheasant habitat multi-scale assessment that fill remaining pheasant–habitat relationships knowledge gaps and support management decisions will increase.

Research selection and classification criteria

Two authors of this paper independently collected and filtered articles following the same protocols. The search was performed in May 2024. The result from each author was then compared to check that the process was performed correctly. The data collection methods closely align with the ones employed by Barg et al (forthcoming 2024; doi: 10.5061/dryad.j3tx95xr4). The review followed the PRISMA guidelines for ecology and evolution (O’Dea et al., 2021) utilizing all databases and collections in the Web of Science. The final search query returned papers with titles or abstracts containing keywords related to pheasants and their habitats.

The search initially returned 1100 papers, which we filtered to 174 peer-reviewed articles based in the U.S.A. or Canada. These papers were further manually filtered to exclude (1) papers that were not about the ring-necked pheasant (Phasianus colchicus), (2) non-original primary peer-reviewed research articles, and (3) articles that addressed pheasant habitat outside the study area (Great Plains, Midwest, and Corn Belt core area). We used the Great Plains regional delineation of Lavin et al. (2011), the Midwest delineation of National Geographic (2024), and the Corn Belt delineation of Green et al. (2019). Studies that only covered a portion of the study area were included. If no location information was available, the study was considered outside the study area. Manual filtering also removed papers that (4) did not research pheasant habitat, resource use, or survival associated with habitat, (5) developed models with data from other studies, to avoid replication, (6) simulation studies, (7) studies describing the geographic range of pheasants, and (8) capture, release or translocation studies, due to their lack of information on wild pheasant habitat. Studies that used a mix of wild and captive/translocated pheasants were excluded if the majority were captive or translocated. If a study had both distinct captive or translocated and wild pheasants in a wild setting, only results from the wild pheasants were regarded. Finally, filtering disregarded (9) genetic and diet studies that did not provide information about pheasant habitat, and (10) multispecies studies that did not focus on pheasant habitat but instead focused on biodiversity or richness metrics.

Only papers that used remote sensing, landcover or vegetation indices were utilized in the scale classification. Papers using other variables such as land surface temperature or soil moisture were not considered, to maintain a focus on pheasant habitat and its characteristics. To address the time lag between the time the research was conducted and the publication year, which can inflate or deflate temporal findings, all papers were considered, regardless of when they were published. Remote sensing originates over a century ago with early technology using aerial photography from planes and even passenger pigeons. The inclusion of older studies illustrates the influence of fundamental methods in modern approaches to pheasant habitat research, especially studies from the mid-20th century onward. Such early studies provide historical context relevant to understanding long-term trends and methodology advancements (Gibson, 2013a and b). This process yielded a final set of 65 articles.

We classified the 65 studies according to information found in their Introduction, Study Area, and Methods sections. Within the Methods sections of the 65 studies, we searched for keywords such as aerial, imagery, photography, remote sensing, landcover, and land use to determine whether the study used remote sensing and/or landcover data. Mixtures of remote sensing and landcover data with additional data, such as field observations and management records, were classified as utilizing these data types. For the studies that did use remote sensing and/or landcover, we noted whether or not ground-truthing was employed. Studies that used remote sensing or landcover were categorized based on the remote sensing data collection platform (aircraft, satellite or unknown) and the data type (photographs, landcover, indices, or unknown). The unknown categories for each platform, data type and scales were added to deal with missing information, report study quality and report their occurrence.

From the 65 studies, we calculated the percentage of articles researching pheasant habitat using remotely sensed or landcover data products as a data source. If the study area or remote sensing use was not clearly stated, we regarded those studies as outside the specific study area and as not using remote sensing. For example, if a paper stated that their study area was in a region that overlaps our study area but doesn’t specify where specifically in the region the study was conducted, it was counted as outside our study area. This ensured that reviewed papers were conducted in our study area and used remote sensing. We also determined the proportion of different remote sensing platforms and data types used in these articles. Percentages were calculated by dividing the usage of each platform or data type by the total usage of remote sensing products across all studies. We also determined the proportional use of different remote sensing datasets. This was calculated by dividing the times each remote sensing dataset was used by the total number of times all the remote sensing dataset were used. To understand the effect of the missing information described by the unknown categories, percentages were also calculated excluding such categories. All percentages were rounded to ensure a collective sum of 100%. The presence of multiple remote sensing products with different platforms and data types in the same studies was accounted for when calculating percentages.

Temporal trends

We conducted the Mann-Kendall tau trend test in R (version 4.4.0) with the ‘Kendall’ package accounting for ties to examine changes of pheasant habitat publication types in time. Ties occur when numerous values are identical in a dataset (McLeod, 2022; R core Team, 2024). The test was employed to detect whether remote sensing use in the pheasant habitat literature increased or decreased over time. This analysis was also performed to test long-term temporal patterns in pheasant habitat publications that utilize remote sensing, providing insights on the evolution of research methodologies used in time. This is a non-parametric test that identifies positive or negative trends in a time series by comparing each point to the subsequent one. This test is well suited for non-normal data, such as the one we had, and can handle outliers well, being an optimal choice for assessing temporal trends in pheasant studies. The outputs of the test are Kendall’s tau coefficient indicating the strength and direction of the relationship and the p-value indicating the significance of the relationship (Kendall, 1945; Sen, 1968, McLeod, 2022). The tidyr package was used to structure and organize the data while the ggplot2 and patchwork packages were used for plot visualizations (Wickham, 2016; Pedersen, 2024; Wickham et al., 2024).

Thematic, spatial, and temporal scale

Papers that used remote sensing or landcover data (N =26) were classified based on the thematic, temporal, spatial, and spectral grain, and temporal and spatial extent with methodologies developed by the authors. Spectral grain (ability of sensors to distinguish between wavelengths) excludes landcover and indices usage in remote sensing. We calculated the percentages of papers using different thematic and spectral resolutions and spatial and temporal grains and extents. The thematic categories are from level 1 (most general) to 5 (most specific in categorization). The National Land Cover Dataset (NLCD) classification system was the base for levels 3 and 4 of the classification, which is a modified version of the Anderson Land Cover Classification System (Anderson et al., 1976; U.S. Geological Survey, 2019). The finer levels of classification are integrated from other standard thematic schemes. The Conservation Reserve Program (CRP) land classification was integrated into level 4 from the Hierarchical All Bird System (HABS) habitat classification that identifies bird habitats at multiple scales (McLachlan et al., 2007). Only shrubs, CRP practices, cultivated crops, herbaceous wetlands, and open water were described by pheasant studies in more detail for level 5 categorization. These levels were categorized according to the USDA CRP practice library, the Cropland Data Layer (CDL) classification, and the Stewart & Kantrud Classification System. The latter is a widely used method for wetland and water systems classification (Stewart & Kantrud, 1971; Farm Service Agency, 2024; National Agricultural Statistics Service, 2024). No standardized classification was used for shrubs in level 5 due to only one paper characterizing shrubs at the species level. Level 2 group classes include categories that are closely related (share similar landcover characteristics) in level 3 in the NLCD classification, except for wetland and water, as they have in common the presence of an aquatic environment. Level 1 provides a general classification of all the classes previously described.

We determined the percentage of studies that used data in each hierarchical level of the classification system. The percentage was calculated by dividing the number of times a category was researched by the total number of reviewed studies (26). These were the categories selected by the paper in their method section before they were analyzed, as sometimes the classification changed to better-fit models during analysis. In cases where a landuse or landcover class was classified differently than in our system, we used our system. For instance, if a paper classified oats as herbaceous (level 2), we classified them as cultivated crops (level 5). When the paper thematic classification did not fit any of our system or was missing, it was classified as ‘other’ in our system. This ensured the standardization of the classification across papers. For example, urban environments were classified as other (level 2).

The spatial grain was the pixel dimensions or stated photograph scale (e.g. 1:3000). If the grain was not stated on the paper, it was classified as unknown to report the occurrence of the missing information. Spatial extent was determined by both explicit reporting (study area) and by estimating it if not clearly stated. This estimation was conducted by assessing study maps and scale bars, considering indicators such as town size, study area, or study block areas, and by summing land cover buffer areas around survey points. If many count routes were present in a state or county (e.g., Breeding Bird Survey routes), the whole state/county was the study area. The percentages of various spatial grains and extents were calculated by dividing the frequency of each grain or extent by the total frequency of remote sensing or land cover products used in the studies. Remote sensing and land cover products were utilized a total of 37 times across the 26 studies. This helped account for the fact that each remote sensing product may have had different grains (depending on the spectral band used and year of remote sensing dataset publication) and extent in each study.

The temporal grain refers to the frequency of remote sensing photography/imagery or land cover mapping use in a study area. The temporal extent denotes the duration of the utilization of the remote-sensing product. Such information was linked to the sensor's temporal resolution and the frequency of imagery and land cover updates, tailored to its use in the specific study. For example, until 2024 the NLCD data had a temporal resolution of 3 to 5 years (now it is yearly) and a temporal extent going from 1992 to 2023 (now it is 1985/86 to present; U.S. Geological Survey, 2024a and b). If the study used such data in 2001, 2006, and 2011, the temporal grain would have been 5 years and the extent 10 years. If the temporal grain or extent was unclear it was classified as unknown to report the occurrence of missing temporal information. The percentages of different temporal grains and extents were calculated similarly to the spatial grain and extent.

The spectral grain determined the habitat type and quality extractable from remote sensing data. Different spectral bands can extrapolate diverse vegetation information which can be relevant to pheasants. For example, natural color photographs (visible spectrum) can only identify broad habitats such as grasslands. Multispectral images using the red and infrared bands can identify vegetation characteristics such as greenness which may be better suited for pheasant habitat identification. The spectral information informs the kind of habitat that can be extrapolated from remote sensing (Gibson et al., 2013a and b; Amirkhiz et al., 2023). The spectral grain was calculated for photographs and images, not landcover or indices (e.g., NDVI), and referred to the number of spectral bands and their width difference to create the photograph or image. If the spectral grain was not clearly stated in the study, it was estimated from the information given, as indicators of it were often present. For instance, if the study reported that natural color aerial photographs were used, the study most likely used the blue, green and red bands that respectively range from 400 to 500 nm, 500 nm to 600 nm and 600 to 700 nm. In that case, three bands and spectral resolution width of 100 nm (difference between wavelength ranges) was reported as the most likely scenario. If the spectral grain was not possible to estimate, it was classified as unknown to report the occurrence of such missing information. The percentage of band number and bandwidth difference used was calculated similarly to the spatial grain.

References

Amirkhiz, R. G., John, R., & Swanson, D. L. (2023). A Bayesian approach for multiscale modeling of the influence of seasonal and annual habitat variation on relative abundance of ring-necked pheasant roosters. Ecological Informatics, 75(1), 102003. https://doi.org/10.1016/j.ecoinf.2023.102003.

Anderson, J. R., Hardy, E. E., Roach, J. T., & Witmer, R. E. (1976). A land use and land cover classification system for use with remote sensor data. U.S. Geological Survey, Professional Paper, 964.

Campbell, J. B., & Wynne, R. H. (2011). Image resolution. In J. B. Campbell & R. H. Wynne (Eds.), Introduction to remote sensing (pp. 285–298). New York, NY: Guilford Press.

Farm Service Agency. (2024). CRP practices library. U.S. Department of Agriculture. https://www.fsa.usda.gov/programs-and-services/conservation-programs/crp-practices-library/index.

Gibson, P. J. & Power. C. H., (2013a). Principles of Remote Sensing. Introductory remote sensing principles and concepts. In P. J. Gibson & C. H., Powers (Eds.), Introductory Remote Sensing Principles and Concepts (pp. 11–66). New York, NY: Routledge.

Gibson, P. J. & Power. C. H., (2013b). Remote sensing Systems. In P. J. Gibson & C. H., Powers (Eds.), Introductory Remote Sensing Principles and Concepts (pp. 66–127). New York, NY: Routledge.

Green, T. R., Kipka, H., David, O., & McMaster, G. S. (2018). Where is the USA Corn Belt, and how is it changing?. Science of the Total Environment, 618, 1613-1618. https://doi.org/10.1016/j.scitotenv.2017.09.325.

Kendall, M. G. (1945). The treatment of ties in ranking problems. Biometrika, 33(3), 239-251. https://doi.org/10.2307/2332303.

Lavin, S. J., Shelley, F. M., & Archer, J. C. (2011). Atlas of the Great Plains (pp 18-21). China: University of Nebraska Press.

McLeod A (2022). _Kendall: Kendall Rank Correlation and Mann-Kendall Trend Test_. R package version 2.2.1. https://CRAN.R-project.org/package=Kendall.

McLachlan, M., Carter, M., Rustay, C., & Venture, P. L. J. (2007). Effects of the Conservation Reserve Program on priority mixed-grass prairie birds. USDA Natural Resources Conservation Service, WA.

National Agricultural Statistics Service. (2024). CropScape - Cropland Data Layer. Retrieved June 20, 2024, from https://nassgeodata.gmu.edu/CropScape/.

National Geographic. (2024). United States regions, Education. Retrieved May 03, 2024, from https://education.nationalgeographic.org/resource/united-states-regions/.

O’Dea, R. E., Lagisz, M., Jennions, M. D., Koricheva, J., Noble, D. W. A., Parker, T. H., Gurevitch, J., Page, M. J., Stewart, G., Moher, D., & Nakagawa, S. (2021). Preferred reporting items for systematic reviews and meta-analyses in ecology and evolutionary biology: A PRISMA extension. Biological Reviews, 96(5), 1695–1722. https://doi.org/10.1111/brv.12721 

Pedersen T (2024). patchwork: The Composer of Plots. R package version 1.3.0.9000, https://github.com/thomasp85/patchwork, https://patchwork.data-imaginist.com.

R Core Team (2024). _R: A Language and Environment for Statistical Computing_. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org.

Sen, P. K. (1968). Estimates of the regression coefficient based on Kendall's tau. Journal of the American statistical association, 63(324), 1379-1389. https://doi.org/10.2307/2285891.

Stewart, R. E., & Kantrud, H. A. (1971). Classification of natural ponds and lakes in the glaciated prairie region (Vol. 92, pp 27-57). Washington: US Bureau of Sport Fisheries and Wildlife.

Turner, M. G., Gardner, R. H., O'neill, R. V., & O'Neill, R. V. (2001). Introduction to Landscape Ecology and Scale. In: Turner, M. G., and Gardner, R. H., Landscape ecology in theory and practice (pp. 1-32). New York: Springer.

U.S. Geological Survey. (2019). NLCD classes. U.S. Geological Survey. Retrieved November 7, 2024, from https://www.mrlc.gov/.

Wickham (2016), H. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York.

Wickham H, Vaughan D, Girlich M (2024). tidyr: Tidy Messy Data. R package version 1.3.1, https://github.com/tidyverse/tidyr, https://tidyr.tidyverse.org.