USGS - science for a changing world

Northern Prairie Wildlife Research Center

  Home About NPWRC Our Science Staff Employment Contacts Common Questions About the Site

Forecasting Production of Arctic Nesting Geese by Monitoring Snow Cover with Advanced Very High Resolution Radiometer (AVHRR) Data

Laurence L. Strong and Robert E. Trost

Abstract: The remoteness and extensive distribution of habitats for Arctic nesting geese makes the acquisition of goose productivity data expensive and difficult. Although the reproductive success of Arctic nesting geese depends on a number of factors, perhaps none is more important than the timely disappearance of snow to allow for the initiation of nesting. We are attempting to develop regression estimators for predicting immature-to-adult ratios in the fall migration. We are using historical snow cover data derived from AVHRR imagery and production indices from surveys of geese during migration, on wintering areas, and from harvest surveys. The purpose of these models is to estimate expected production for input to the annual harvest regulation process in July. The image processing requirements for this objective are demanding and ideally include compositing and analysis of daily images to obtain cloud-free observations and near real-time analysis of snow cover. We are attempting to use a mixture model to estimate the proportion of an area that is free of snow and ice. This information can be used to determine more precisely when nesting habitat becomes available. We discuss the needs and requirements for an operational evaluation of nesting habitat conditions, the development and opportunities for enhancing a digital, image processing methodology, and obstacles to the development of the regression estimators using historical AVHRR data.

Table of Contents



The Canadian Arctic represents the major breeding area for most populations of snow (Chen caerulescens), Ross' (C. rossii), white-fronted (Anser albifrons), small Canada (Branta canadensis), and Atlantic brant geese (B. bernicla) (Bellrose 1976). Although the reproductive success of Arctic nesting geese depends on a number of factors, perhaps none is more important than the timely disappearance of snow, ice and melt water to allow for the initiation of nesting (Reeves et al. 1976). Arctic nesting geese have a relatively short period of time to build nests, lay and incubate eggs, and rear young to fledging. Delayed snow melt will retard the initiation of nesting, or if late enough, prevent it altogether (Cooch 1964). Early disappearance of snow usually results in a strong breeding effort and good gosling production.

The remoteness and extensive distribution of habitats for Arctic nesting geese makes the acquisition of goose productivity data expensive and difficult (Fig 1). Quantitative information on production of Arctic nesting geese is presently obtained from surveys conducted at various locations during migration and winter in those species in which young of the year are morphologically discernable from individuals born in previous years (Bellrose 1976). However, these data are not available until after the annual regulations process in July.

Figure 1: Map of Canada showing nesting areas of the Arctic Canada goose
Fig. 1.  Arctic Canada goose nesting areas.

Goose production forecasts in regulations processes are based on qualitative information from personal communication with researchers conducting nesting studies at specific locations. Indirect qualitative information is obtained from interpretation of Northern Hemisphere Snow and Ice Analysis Charts prepared weekly by the National Oceanic and Atmospheric Administration (NOAA) (Matson et al. 1986). Since 1987, the Canadian Wildlife Service (CWS) and the U.S. Fish and Wildlife Service (FWS) have conducted the Cooperative Arctic Goose Survey to provide a production forecast prior to the annual regulations process. The survey is an assessment from low-flying aircraft of goose social behavior and snow conditions at important nesting areas in Arctic Canada. These qualitative approaches have provided general indications of expected production for many Arctic nesting goose populations. However, the lack of quantitative data is seen as a serious impediment to long-term management efforts such as the development of predictive models and the ability to properly evaluate management actions. The relation between the timing of snow melt and goose productivity suggests the potential to develop regression models to estimate production prior to the fall flight from observations of snow cover and spring phenology on nesting areas.

In 1975, the FWS and the CWS began a cooperative effort to examine the potential of remotely sensed data for estimating snow cover and habitat conditions (Reeves et al. 1976, Reeves 1978). This application used manual interpretation of small-scale black and white glossy prints of the AVHRR red (0.58-0.68 µm) spectral band to estimate the date of snow disappearance on specific nesting areas. Discrimination between snow, ice and clouds on a single date was often difficult. However the application was successful because the discrimination could be made using images from multiple dates and from image texture. Qualitative predictions of goose production were evaluated by comparison with age ratios of immature and adult geese in the fall harvest. Reeves et al. (1976) concluded that AVHRR imagery is an economical method for identifying probable areas and times of catastrophic or outstanding goose production. The FWS Migratory Bird Management Office (MBMO) has continued using AVHRR imagery as a source of information for preparing the annual Status of Waterfowl and Fall Flight Forecast administrative report. Their experience has generally been successful, although estimation of snow cover was quite subjective and the scale of the photos made the estimation of conditions for a specific nesting area difficult.

Reeves et al. (1976) suggested that future efforts should focus on two enhancements to the methodology: (1) estimating the proportion of the nesting area that is free of snow and ice cover, since geese may commence nesting when their nest sites are exposed even when considerable snow cover may be present nearby, and 2. improving the precision of the estimate of the date at which nesting habitat becomes available, since each day of delay may result in a reduction in average clutch size and the proportion of successful nests.

With the rapid development of computers, digital image processing of AVHRR images has replaced manual interpretation in many applications (Hastings and Emery 1992). One of the major obstacles to the use of satellite data in Arctic environments, the discrimination of clouds from snow cover, has been overcome using AVHRR channel 3. In this region of the electromagnetic spectrum, significant quantities of radiation originate from both solar and terrestrial sources. Several investigators have developed methods for separating the reflected and emitted components of the signal in channel 3, and used the channel 3 reflectance as an additional variable for discriminating snow and ice from clouds (Gesell 1989; Stowe et al. 1991; Gutman 1992).

The spatial resolution of the AVHRR presents difficulties for estimating snow cover at nesting areas. The image measurement of reflectance is the integration of the reflectance from all the different land covers within the instantaneous field-of-view (IFOV) which at nadir is 1.2 km² for AVHRR Local Area Coverage (LAC) data. Thus, the satellite does not resolve individual nest sites, but rather the landscape within which the nest sites occur. The need to quantify the composition of a scene in which the objects of interest are smaller than the IFOV of the sensing device has led to the development of image processing models referred to as mixture models (Shimabukuro and Smith 1991; Strong et al. 1991). A primary objective of mixture models is the estimation of the target proportions that are present within an IFOV.

Our objectives are to (1) identify the needs and requirements for an evaluation of habitat for Arctic nesting geese, (2) describe the development and opportunities for enhancing an AVHRR image processing methodology, and (3) describe obstacles to the development of regression models using snow cover estimates from historical AVHRR data to predict goose production.


We began a pilot study in August 1991 to investigate the potential to estimate snow cover from digital image processing of AVHRR data and to develop models for forecasting goose production. AVHRR data are available in three digital data types as described by Hastings and Emery (1992). We performed a computer search for AVHRR data of the Arctic using the NOAH Online Search Catalog and Retrieval (OSCAR) and ILAB (Image Library and Browse) automated data search programs. We decided to use AVHRR Global Area Coverage (GAC) data, despite problems with the sampling scheme used to construct the images and their low spatial resolution, because the data are consistently available for the Canadian Arctic since 1979 (Belward 1992).

We selected a single GAC image for each of three one-week periods for each year from 1979 to 1991. The three weekly periods were 6-12 May, 3-9 June, and 18-24 June. These periods were chosen to span the range of snow melt over the Canadian Arctic. In most years, snow melt occurs earlier in the western Arctic than in the eastern Arctic. The image for each period was chosen to contain minimal clouds as determined by inspecting paper copies of the visible red band for the Canadian Arctic as a whole. The search for images began with the image NOAA used to construct the Northern Hemisphere Weekly Snow and Ice Cover Chart.

The images were resampled to a Lambert Conic Conformal map projection with 4000-m square pixels using an affine piecewise plane projective transformation algorithm. Latitude and longitude data in the tape records were used for control points (Kidwell 1986). The data were calibrated to reflectance factors and brightness temperatures using the methods of Planet (1988). The solar zenith angle at the center of the image was used in the calculation of reflectance factors. No atmospheric or view angle corrections were made. Brightness temperatures were calculated from average coefficients for the entire image rather than on a line-by-line basis. We estimated the channel 3 reflectance factor by the methods of Stowe et al. (1991) for data from NOAA 11 and 9, and those of Baglio and Holroyd (1989) for data from NOAA 7 and earlier satellites. For each image, we examined each nesting area for the presence of clouds using the reflectance factors for channels 3, 2, and 1 displayed in red, green, and blue with a normal contrast stretch. Because cloud cover was frequent, we examined the commonly used Normalized Difference Vegetation Index (NDVI) image-compositing procedure for its ability to produce minimum cloud cover composites from images of the Arctic.

We have just begun the digital classification of the images into four land cover types: (1) snow and ice, (2) not snow, land surface without snow or ice cover, (3) water and (4) clouds. The classification process uses the reflectance factors and brightness temperatures as explanatory variables. We began with inspection of univariate and bivariate histograms to identify thresholds and decision surfaces for discriminating among the land cover types. We also used supervised training procedures to compute descriptive statistics for each land cover type. Training areas were selected from contiguous pixels of what appeared to be a single land cover type. Using the descriptive statistics, we interactively and iteratively applied a range of thresholds to individual spectral channel images and subjectively evaluated the classification results on a display monitor. Based upon this process of successive approximation, rules were developed for a multispectral parallelepiped classification of the land cover types (Carroll 1990).

We used the descriptive statistics from the supervised training procedures to estimate the mean vector for each land cover type for use in a mixture model analysis. A pooled covariance matrix was estimated using a large sample of image data. The mixture model utilized a generalized least squares solution and is described in Hlavka and Spanner (in press). The percent snow cover at a nesting area was estimated as the average of the predicted proportion of snow cover for all pixels comprising the area.

The FWS and CWS Cooperative Arctic Goose Survey provides an opportunity to evaluate the snow cover estimates from digital image processing of late June AVHRR data. Subjective estimates of percent snow cover are made from a twin engine aircraft flying at approximately 75 m above ground during the survey. In 1992 and 1993, we acquired oblique video data of nesting areas. The video system consisted of a color camera, a wide-angle 5.9-mm lens, a video cassette recorder, a color monitor, and a global positioning system (GPS). A sample of video frames for each nesting area was captured as digital images and percent snow cover estimated using image thresholding (Schowengerdt 1983:68).

Age ratio data from surveys of geese on migration and wintering areas, and from the individual parts-collection data from harvest surveys are being obtained from Flyway Technical Reports and FWS administrative reports by biologists with the MBMO. Band recovery data distributions are being used to define harvest areas by nesting location and to summarize age ratio data.

Multiple regression estimators for predicting age ratios in the fall flight are being developed using estimates of snow cover as explanatory variables and the age ratio as the response variable. Estimates of the percent snow cover at a nesting area for three periods for each year and a corresponding annual age ratio will be a single sample unit in the regression model. A separate regression model is being developed for each nesting location or combination of locations for which age ratio data can be obtained. Verification of these models will be attempted in future years by comparing predictions with survey data and modifications will be made as additional data become available.


Figure 2: Color AVHRR GAC image
Fig. 2.  AVHRR GAC image of the Great Plain of Koukdjuak, Foxe Peninsula, and Southampton Island on 19 June 1991. Image prepared from reflectance factors for channels 3, 2, and 1 displayed in red, green, and blue, respectively.

The search for historical AVHRR data of the Arctic was difficult. The OSCAR data catalog does not provide a complete listing of data acquired at all receiving stations. For this study, the omission of AVHRR High Resolution Picture Transmission (HRPT) data collected at the Canadian Centre for Remote Sensing (CCRS) Prince Albert Satellite Station in Saskatchewan was unfortunate. The acquisition of AVHRR HRPT data for much of the Canadian Arctic is not possible from NOAA receiving stations in Wallops Island, Virginia or Gilmore Creek, Alaska. The acquisition of AVHRR LAC data is limited by the on-board recording capacity and high demand for the data exceeds the tape recorder capacity. Although we requested that NOAA collect LAC data of the Canadian Arctic during May and June in 1992 and 1993, fewer than five images were collected because of higher priorities for tape recorder space. The OSCAR system did not provide a quick look capability or cloud cover estimates for images. This required that we examine file copies of the imagery to determine cloud cover. Paper copies of LAC were often not available. The collection of prints for GAC data was complete.

The georegistration of the images provided a first order correction which allowed nesting areas to be easily located. However, georegistration of AVHRR data using the latitude and longitude locations on the tape did not provide sufficient accuracy to permit the overlay of images from different dates. Calibration of the images to reflectance factors and brightness temperature enhanced the visual discrimination and interpretation of the land covers (Fig. 2).

A principal obstacle to the estimation of snow cover has been the extensive and frequent cloud cover on the AVHRR images. For some nesting areas, all three observations within a nesting season have extensive cloud cover. This suggested the need to investigate image compositing procedures. The compositing procedure based on the maximum NDVI was not effective in removing clouds for the Arctic environment. Snow, ice, and water often had an NDVI that was less than clouds.

Inspection of univariate and bivariate histograms revealed few obvious thresholds or decision surfaces for discriminating among land cover types. This was not surprising given the large pixel size of AVHRR data and the common occurrence of mixed pixels. Supervised training provided a first approximation of the decision surfaces for discriminating among the land cover types (Fig. 3). The channel 3 reflectance factor was useful for discriminating snow and ice from clouds. The channel 4 or 5 brightness temperatures were useful for separating snow free terrestrial surfaces from clouds which had overlapping channel 3 reflectance. Channel 1 and 2 reflectance factors were useful for separating snow from snow free terrestrial land cover. The channel 2 reflectance factor was useful for identifying water. The decision surfaces for discriminating among land cover types were scene dependent and probably reflected temporal changes in bidirectional reflectance factors and imperfect calibration of the data.

Figure 3: Bar graph showing values for Clouds, Thin Clouds, Snow, Ocean Ice, Not Snow, and Water
Fig. 3.  Reflectance factor and brightness temperature for four land cover types and two cloud types.

Subjective evaluation of the mixture model results was encouraging. A more objective, though imperfect, evaluation of mixture model estimates of the proportion of snow cover at nesting areas will be possible using the estimates of snow cover from the video imagery acquired during the Cooperative Arctic Goose Survey.


The AVHRR, with its daily synoptic coverage, is useful for estimating snow cover and spring phenology for the remote and extensive habitats used by Arctic nesting geese. From the perspective of building regression models forecasting production of Arctic nesting geese, the selection of AVHRR GAC data is justified because the data are available for a greater number of years than either AVHRR LAC or HRPT. However, for operational use, HRPT or LAC data should be used since these data have 16 times the spatial resolution of GAC and the sampling scheme used in the construction of GAC data may result in estimates of snow conditions for areas adjacent to, but not at the nesting area (Belward 1992).

Cloud cover for some portion of an image is inevitable given the synoptic coverage of AVHRR. Our experience has been that it is difficult to select cloud-free images for all nesting areas. This problem has led to the development of image-compositing procedures. However, the digital image processing requirements for this procedure are computer intensive. Furthermore, the software for precise georegistration and calibration of AVHRR data including solar, view angle and atmospheric effects is often absent from commercial image processing systems, although software availability is improving.

Present development of programs that provide near real-time geocoded and radiometrically corrected AVHRR data products and time series composites, where cloud cover has been minimized, it's very important to users such as the FWS. The CCRS is now producing 10-day composites of AVHRR data from AVHRR HRPT data acquired at the CCRS receiving station in Prince Albert, Saskatchewan, for regions within Canada, one of which is the Northwest territories (Erickson et al. 1991). However, these images are prepared using the maximum NDVI-compositing procedure which, unfortunately, does not always minimize cloud cover in Arctic environments where snow, ice and water are major components of the landscape. The CCRS is considering the incorporation of other compositing criteria into their NOAA AVHRR Geocoding and Compositing System.

The spatial resolution of AVHRR data presents another obstacle to estimation of habitat conditions for Arctic nesting geese. Mixture model estimates of the proportion of a nesting area free of snow and ice cover should improve the precision of the estimate of the date at which nesting sites become available. Evaluation of the mixture model estimates of land cover proportions could be improved by comparison with land cover estimates from satellite remote sensing data with higher spatial resolution. Landsat MSS data would be useful for specific nesting areas, however, data availability, the 16-day repeat time, and frequent cloud cover will limit this application.

The capabilities of computerized searches for AVHRR data are improving. The NOAA ILAB system provides a quick look of the visible and a thermal spectral band image. The U.S. Geological Survey Global Land Information System (GLIS) includes AVHRR HRPT data acquired at the Canadian Prince Albert Satellite Station. The GLIS system provides quick look capability using the red spectral band image. The CCRS has upgraded its ability to acquire, process, archive and inspect NOAA AVHRR data including microfiche for quick look and calculation of cloud cover (Erickson et al. 1991).

Our study is part of a long-term effort by the FWS and CWS to develop improved methods for estimating the status and trends of Arctic goose populations. These tools will allow Federal and State regulatory personnel to predict production prior to the annual regulations process and improve population management decisions. Near real-time maps of the distribution of snow cover could also be used to develop dynamic, optimal sampling plans for the Cooperative Arctic Goose Survey.


We thank Dale Humburg, Daniel Nieman, Bruce Turner, Ron Renolyds, Harvey Nelson, Fred Johnson, Richard Kerbes, and Austin Reed for sharing their experiences with Arctic nesting geese and their habitat. We thank Larry Stowe, Garik Gutman, Chris Hlavka, Tammy Rockvam, and Mike Unverferth for valuable discussions on processing of AVHRR data. We thank Gerry Bergin, Roy Irwin, John Thoma, Kay Metcalf, Dave Douglas, Jeff Eidenshink, Arvon Erickson, Gene Fosnight, Bonnie Harris and Pat Hulbert for assistance in determining the availability of and obtaining AVHRR data. We thank Betty Euliss for assistance in processing the AVHRR data. We thank Betty Euliss, Ray Thielman and Toni Hanson for assistance in preparing material for the manuscript and the talk. We thank Dave Gilmer, Jeff Price, John Ramsey, and Terry Shaffer for reviewing the manuscript.

Literature Cited

Baglio, J.V., Jr. and E.W. Holroyd III. 1989. Methods for operational snow cover area mapping using the Advanced Very High Resolution Radiometer - San Juan Mountains Test Study. Research Technical Report, U.S. Geol. Surv., EROS Data Center, Sioux Falls, SD.

Bellrose, F.C. 1976. Ducks, geese and swans of North America. Stackpole Books, Harrisburg, PA.

Belward, A.S. 1992. Spatial attributes of AVHRR imagery for environmental monitoring. Int. J. Remote Sensing 13(2):193-208.

Carroll, T.R. 1990. Operational airborne and satellite snow cover products of the National Operational Hydrologic Remote Sensing Center. In: Proceedings of the 47th Annual Eastern Snow Conference. U.S. Army, Cold Regions Research and Engineering Laboratory, Hanover, NH, CRREL Report 90-44. pp. 87-98.

Cooch, G. 1964. Snows and Blues. In: J.P Linduska, ed. Waterfowl tomorrow. U.S. Government Printing Office, Washington, D.C. pp. 125-133.

Erickson, A.D., P.S. Leung, T.A. Fisher, P.M. Teillet, B. Guindon, and R.J. Brown. 1991. AVHRR data archiving and processing. In: Proceedings 14th Canadian Symp. on Remote Sensing, Calgary, Alberta, Canada. pp. 205-210.

Gesell, G. 1989. An algorithm for snow and ice detection using AVHRR data: An extension to the APOLLO software package. Int. J. Remote Sensing. 10(4): 897-905.

Gutman, G.G. 1992. Satellite daytime image classification for global studies of earth's surface parameters from polar orbiters. Int. J. Remote Sensing 13(2):209-234.

Hastings, D.A. and W.J. Emery. 1992. The Advanced Very High Resolution Radiometer (AVHRR): A brief reference guide. Photogrammetric Engineering and Remote Sensing 58(8):1183-1188.

Hlavka, C.A. and M.A. Spanner. In press. Unmixing AVHRR imagery to assess clearcuts and forest regrowth in Oregon. IEEE Trans. Geosci. and Remote Sensing.

Kidwell, K.B., ed. 1986. NOAA polar orbiter data users guide. NOAA/NES-DIS/ NCDC/SDSD, Washington, D.C.

Matson, M., C.F. Ropelewski, and M.S. Varnadore. 1986. An atlas of satellite- derived Northern Hemispheric snow cover frequency. NOAA atlas, Washington, DC.

Planet, W.G., ed. 1988. Data extraction and calibration of TIROS-N/NOAA radiometers. NOAA Tech. Memorandum NES-107.

Reeves, H.M. 1978. Applications of satellite imagery to management for Arctic nesting geese. In: Proceedings of Pecora IV symposium. Natl. Wildl. Fed. Sci. and Tech. Ser. 3. pp. 219-226.

Reeves, H.M, F.G. Cooch, and R.E. Munro. 1976. Monitoring Arctic habitat and goose production by satellite imagery. J. Wildl. Manage. 40(3):532-541.

Schowengerdt, R.A. 1983. Techniques for image processing and classification in remote sensing. Academic Press, New York, N.Y.

Shimabukuro, Y.E. and J.A. Smith. 1991. The least-squares mixing models to generate fraction images derived from remote sensing multispectral data. IEEE Trans. Geosci. and Remote Sensing 29(1):16-20.

Stowe, L.L., E.P. McClain, R. Carey, P. Pellegrino, G.G. Gutman, P. Davis, D. Long, and S. Hart. 1991. Global distribution of cloud cover derived from NOAA/AVHRR operational satellite data. Adv. Space Res. 11(3):51-54.

Strong, L. L., D.S. Gilmer, and J.A. Brass. 1991. Inventory of wintering geese with a multispectral scanner. J. Wildl. Manage. 55(2):250-259.

This resource is based on the following source (Northern Prairie Publication 0926):

Strong, Laurence L., and Robert E. Trost. 1994. Forecasting production of arctic nesting geese by monitoring snow cover with advanced very high resolution radiometer (AVHRR) data. Proceedings of the Pecora Symposium 12:425-430.

This resource should be cited as:

Strong, Laurence L., and Robert E. Trost. 1994. Forecasting production of arctic nesting geese by monitoring snow cover with advanced very high resolution radiometer (AVHRR) data. Proceedings of the Pecora Symposium 12:425-430. Jamestown, ND: Northern Prairie Wildlife Research Center Online. (Version 26APR2002).

Laurence L. Strong, U.S. Fish and Wildlife Service, Northern Prairie Wildlife, Research Center, Jamestown, ND, 58401

Robert E. Trost, U.S. Fish and Wildlife Service, Office of Migratory Bird, Management, Washington, DC, 20240

Downloading Instructions -- Instructions on downloading and extracting files from this site.
(Download) ( 82K ) -- Forecasting Production of Arctic Nesting Geese by Monitoring Snow Cover with Advanced Very High Resolution Radiometer (AVHRR) Data
Installation: Extract all files and open index.htm in a web browser.

Accessibility FOIA Privacy Policies and Notices

Take Pride in America logo logo U.S. Department of the Interior | U.S. Geological Survey
Page Contact Information: Webmaster
Page Last Modified: Friday, 01-Feb-2013 19:20:12 EST
Sioux Falls, SD [sdww55]