The influence of snow cover variability and tundra lakes on passive microwave remote sensing of late winter snow water equivalent in the Hudson Bay lowlands

Four years of regional snow surveys in the Northwest Territories and northern Manitoba, Canada, have shown that current North American operational passive microwave snow water equivalent (SWE) retrieval algorithms consistently underestimate in tundra environments. Almost all contemporary SWE algorithms are based on the brightness temperature difference between the 37GHz and 19GHz frequencies found onboard both past and present spaceborne sensors. The passive microwave underestimation of SWE is likely a result of the distribution and deposition of the tundra snow, coupled with the influence of tundra lakes on brightness temperatures at 19GHz. The tundra environment is dominated by thousands of small lakes and ponds which can cover up to 30-40% of the landscape and current SWE algorithms do not consider the impact of sub-grid scale lake-cover fraction. To better our understanding concerning the underestimation of passive microwave SWE retrievals on the tundra, Environment Canada conducted an intensive field campaign in March 2006, just south of Churchill, Manitoba. In situ measurements of snow depth, SWE, snow density, snow grain size and snow stratigraphy were recorded at 87 sites within a 25km by 25km grid located over the Marantz Lake region of the Hudson Bay Coastal Plains. Coincident multi-scale passive microwave airborne (70m resolution) and spaceborne (regridded to 25km resolution) data were measured at 6.9GHz, 19GHz, 37GHz and ∼89 GHz frequencies during the same time period.;Analysis of the high-resolution airborne data revealed a sensitivity of the 19GHz frequency to tundra lake features. The magnitude of 19GHz emission measured over some lakes was far less than the surrounding land surfaces, effectively minimizing the difference between the 37GHz and 19GHz brightness temperatures, thereby resulting in lower SWE retrievals. Interestingly, the 19GHz frequency was not sensitive to all lakes. A unique characteristic of tundra lakes and ponds are their shallow nature; lakes and ponds can completely freeze to their beds, and when this occurs, it was hypothesized that the 19GHz frequency would no longer be as sensitive to the lakes. To provide a possible explanation for why the 19GHz brightness temperatures were influenced in different ways by different tundra lakes, the production of a regional tundra lake ice map using a time series of RADARSAT-1 ScanSAR Wide-A imagery was conducted.;The production of the regional tundra lake ice map was based on the established technique of using a synthetic aperture radar winter time series for identifying a probable set of lakes that were most likely frozen to bottom or had free-floating ice based on the observed change in backscatter intensity values throughout the winter season. There was no way of confirming whether the different lake ice types identified were accurate because no field observations of lake ice information were collected during the March 2006 field campaign. Instead, a comparison of the pattern and magnitude of the changing backscatter values for each different ice type was conducted with previous SAR lake ice studies. There were some differences in magnitude and ranges of backscatter values, but the shape of the backscatter intensity patterns compared nicely with previous studies for the floating and grounded lake ice types indicating that these types of lake ice can likely be identified with some confidence.;The results of this study highlight the difficulty in using coarse resolution passive microwave spaceborne sensors to estimate SWE in an environment with heterogeneous sub-grid lake cover and snow distribution. There is great potential in using a time series of RADARSAT ScanSAR Wide images for the purpose of mapping regional lake ice conditions to assist in the interpretation of tundra passive microwave brightness temperatures. An accurate regional lake ice map and information on snow cover distribution based on land cover information would be very useful as inputs into future tundra specific SWE algorithms so that appropriate correction coefficients could be applied.;An analysis of the snow survey data highlighted small-scale localized patterns of snow distribution and deposition on the tundra that likely influenced the SWE underestimation from large-scale passive microwave spaceborne sensors. The distribution of snow in this environment was controlled largely by wind and the presence of taller vegetation (approximately greater than 1m in height). Snow from the open tundra plains was re-distributed into smaller scale features with taller vegetation such as narrow creekbeds, lake edge willows and small tundra tree-islands. The very large amount of snow deposited in these features has a reduced influence on the microwave emission measured by large-scale passive microwave spaceborne sensors because of the small proportion of the land cover these features encompass, and is therefore most likely unaccounted for in current methods of satellite SWE estimation.