Arctic Coastal Fog over Greenland Glaciers using an Improved MODIS Fog Detection Method and Ground Observations

Abstract:

Annual breakup of sea ice causes fog in Arctic coastal regions, which can both reduce and enhance glacier melt. With progressive sea ice loss and increasing temperatures and atmospheric moisture in the Arctic, it is essential to determine the frequency and spatial extent of fog in order to understand its present and future effects on glacier mass balance. Previously, we determined Greenland coastal fog to peak with 15-25% of days in July. Here, we present the spatial and vertical extent of significant melt-season fog events over Greenland coastal glaciers and the ice sheet. To this end, we modified a MODIS fog/low stratus detection method by Bendix et al. (2005), with verification by weather and radiosonde data, timelapse and Landsat imagery, and independent fog classifications. Our fog-detection method uses MODIS Levels 1b and 2, processed in an ENVI-ArcGIS environment as follows: 1) visual examination and application of vegetation and snow indices; 2) initial fog/low stratus discrimination with novel band thresholds and cloud products; 3) verification using cloud phase/temperature products; 4) cleaning misclassified pixels; 5) calculating fog/low stratus optical and geometrical thickness; 6) final differentiation of fog from low stratus using edge-pixel detection, trend-surface fitting, and DEM filling. The end product consists of 500 m fog-mask pixel maps over Greenland, with minimum and maximum possible extents based on classification of fog versus low stratus. Our results show that fog can cover extensive areas of the Greenland ice masses. Persistent fog events in early, mid, and late melt-season were extracted for East Greenland using fog rim detection overlain on the GIMP DEM, the Randolph Glacier Inventory, and a coast shapefile. E.g., a 4 July 2002 fog event covers 4300-5000 km² of ice, with a maximum inland extent of 85 km to an elevation of 1250 m asl. Fog thickness over ice is 20-800 m, but can be underestimated by >50 m compared to radiosonde data.