Fine Beam RADARSAT-1 for Monitoring Pond Dynamics in the Alaskan Boreal Forest

By Dave Verbyla, Department of Forest Sciences, University of Alaska Fairbanks

Several recent scientific studies have used historic remotely-sensed imagery to document a substantial reduction in the number and size of ponds in many boreal regions of Alaska, including interior Alaska, the Kenai Peninsula, and the Seward Peninsula. This reduction may be associated with climate warming effects such as increased transpiration and evaporation of water during longer above-freezing seasons. In regions of discontinuous permafrost, shrinking ponds may also be due to thinning permafrost and development of unfrozen zones beneath ponds, called taliks, allowing for drainage.

Shallow lakes and ponds are of particular concern because they are important habitat for a variety of mammals and waterfowl, and they are the most susceptible to drying in a warming climate. By monitoring the dynamics of open water from spring break-up to fall freeze-up, it is possible to classify boreal ponds as deep and stable ponds versus shallow and ephemeral ponds likely to succeed to terrestrial habitat under a warmer climate.

One potential source of imagery for monitoring lakes and ponds across boreal Alaska is Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper (ETM+) imagery, available on an 8-day orbit. There are two problems with this source of imagery. First, since TM and ETM+ are optical sensors, cloud cover limits the ability to delineate ponds, especially at high latitudes where cloud shadows cover a greater area than the clouds. For example a scene with 25 percent cloud cover could be over 60 percent unusable due to clouds and cloud shadow. Shadows are especially problematic when trying to spectrally classify ponds using optical data, as open water and shadow both have low spectral reflectance in most reflective bands. Second, the Scan Line Corrector (SLC) failed in Landsat-7 ETM+, creating gaps in images, especially away from the center of an image (Figure 1).

One potential alternative to optical imagery for monitoring interseasonal pond dynamics is Synthetic Aperture Radar (SAR) imagery. Since many of the shallow ponds in boreal Alaska are relatively small, I requested RADARSAT-1 fine-beam mode acquired on at least a 2-week interval during the unfrozen period of April to October 2005 as a test case for the Michumina Basin in Denali National Park. RADARSAT-1 data were downloaded via ftp and processed to geotiff images using the ASF Convert tool. The Convert tool was used to resample the imagery to an Alaska Albers projection with a 6-m pixel size. The resulting images were then spatially co-registered using a linear rectification model based on locations that were visible on at least two images. Once a time series of RADARSAT-1 images were spatially co-registered (Figure 2), ponds that changed could be visualized by displaying a SAR image from two time periods as a red-green-blue color composite, where areas that changed from water to land would appear green on the color composite.

It is possible to create polygons representing ponds and lakes from any RADARSAT-1 image by using some Geographic Information Systems (GIS) techniques. I used an unsupervised classification technique to group pixels with similar values into classes. One problem inherent with this approach is that some isolated pixels are typically misclassified due to the speckle nature of SAR data. These scattered pixels were eliminated by using a majority filter and then groups of connected water pixels were extracted using a GIS. These groups were then converted into polygons using a GIS.

One of the advantages of polygons is that metrics such as pond area and shoreline length can easily be computed for each water body using a GIS (Figure 3). Polygons can then easily be selected based on area criteria, for example, select and display all ponds greater than 1 hectare in area. Since this methodology can be applied to a time series of SAR images, ponds can be selected based on changes in area such as selection of all ponds that have decreased by at least 10 percent between two time periods or shorelines that have declined by at least 50 m.

For the 2005 season, the shallow lake surfaces peaked around May 20, following the spring snowmelt period. There was a slight decrease in surface area in all lakes throughout the summer until late September. Less than 5 percent of the lakes had a decrease in surface area, but the lakes that changed did so substantially (Figure 4). This seasonal pattern of lake surface water dynamics may differ considerably from year to year. For example, in 2003, the month of July had record precipitation, while in 2004, May was relatively wet, followed by a record drought period.

SAR is the only cost-effective way to monitor seasonal surface water dynamics across the remote Alaskan boreal landscape.

Click here to download a copy of the newsletter featuring this article