2012 AGU Chapman Conference on Remote Sensing of the ...

Vanderjagt, Benjamin J.How sub-pixel snow depth, vegetation, and grainsize variability affect the ability to estimate largescaleSWE from microwave measurements in alpineareasVanderjagt, Benjamin J. 1 ; Durand, Michael 1 ; Margulis, Steve 2 ;Kim, Ed 3 ; Molotch, Noah 41. Ohio State, Columbus, OH, USA2. University of California at Los Angeles, Los Angeles, CA,USA3. NASA Goddard Space Flight Center, Greenbelt, MD, USA4. University of Colorado at Boulder, Boulder, CO, USACurrent methods to retrieve snow water equivalent(SWE) using passive microwave (PM) remote sensingmeasurements are often characterized by large uncertaintieswhich result from the natural heterogeneity of snowpackand vegetation within the microwave footprint. Snowpackheterogeneities include snow grain size, snow depth, andlayering of snowpack. Vegetation height, needle or leafdensity, and species also vary within microwave footprints. Itis not currently understood the extent to which the passivemicrowave measurement is sensitive (or insensitive) to SWE(as well as to the different variables listed above), as afunction of the scale of the microwave measurement. In ourinvestigation, we utilized the multi-scale Cold LandProcesses Experiment dataset for in situ and microwavedatasets. In order to better characterize the effect thatvariability in the snowpack and vegetative states has on themicrowave observation as function of scale, we conductanalysis in two different ways. 1) First, we estimate what weexpect the observed radiances (brightness temperature) to beat specific locations, based on in situ and LiDAR derivedsnow data gathered from those same locations, using aradiative transfer model (RTM). We then compare ourestimated radiances to the actual observed radiances, andcharacterize the errors, as a function of the variability of thedifferent snowpack and vegetative properties. 2) Based onthe “real” variability of snow and vegetative states, we willcreate synthetic observations of microwave brightnesstemperatures, and perturb the different snowpack andvegetative properties in order to examine the extent to whichthe synthetic microwave signal is affected by each individualproperty, at differing scales. Based on the synthetic analysisdescribed above, there still exists sensitivity to the meansnow depth within microwave footprints, even with highlyheterogeneous alpine snowpack and snow grain exponentialcorrelation lengths ranging from 0.05-0.3 millimeters.Furthermore, we find that the modeled microwave signal issensitive to the mean snow depth even as we aggregate tolarger scales. We also characterize the effect to whichvegetation attenuates the microwave observation, as afunction of the forest cover contained within the PMfootprint. We find that even relatively little forest coverwithin the subpixel (> 15%) can increase the microwavemeasurement by up to 10 K, as well as reduce the signal-tonoiseratio between the microwave measurements and theSWE. While the increase in brightness temperature can inprinciple be accounted for in SWE retrieval algorithms, thedecrease in signal-to-noise ratio results in a loss ofinformation about SWE. The primary research advance thatwe expect from our work is a fundamental understanding ofthe sensitivity (or lack thereof) of the passive microwaveobservation to spatial average SWE as a function of the scaleof the microwave measurement. Limiting cases, such as thethreshold vegetation cover fraction at which SWE is nolonger retrievable are discussed.Varhola, AndrésCombining Coordinate-Transformed AirborneLiDAR and LANDSAT Indices to Obtain ForestStructure Metrics Relevant to DistributedHydrologic ModelingVarhola, Andrés 1 ; Coops, Nicholas C. 11. University of British Columbia, Vancouver, BC, CanadaCurrent remote sensing technologies are not yet capableof directly and reliably measuring snow water equivalent atspatial and temporal resolutions adequate for the estimationof streamflow and flood risk in snow-dominatedcatchments. Hydrologists therefore depend on detailedprocess-based hydrologic models to simulate snowaccumulation and depletion patterns, which are in turnhighly affected by the presence and structure of vegetation.However, the lack of spatially-explicit and relevant foreststructure metrics needed to parameterize these models overlarge basins has led to the use of oversimplifiedrepresentations of heterogeneous forested landscapes,introducing additional uncertainties to predictions. One keyremote sensing technology that has demonstratedsignificant potential for measuring forest structure is LightDetection and Ranging (LiDAR). As airborne LiDARbecomes more available across larger areas, its use as a driverof vegetation structure for hydrologic models has alsobecome increasingly important. Traditional hydrologicmodels require four vegetation structure metrics —leaf areaindex (LAI), sky-view factor (SVF), canopy cover and height—to simulate processes related to radiation fluxes, windattenuation, snow or rain interception andevapotranspiration. In this study we developed a novelmethodology with two objectives: a) derive these fourstructural metrics at the tree and plot levels using a LiDARdataset, and b) explore their relationships with Landsatderivedvegetation and foliar moisture indices forextrapolation purposes. To achieve this we projected LiDARreturns into a polar coordinate system to generate synthetichemispherical images that directly provided LAI and SVFwhen calibrated with their field-based optical counterparts,whereas canopy cover and height were accurately obtainedfrom LiDAR without coordinate transformation. Themethod allowed the retrieval of these metrics at any locationwithin the 8,000 ha LiDAR transect we acquired in centralBritish Columbia, resulting in a sampling design thatincluded 376 different Landsat pixels and nearly 16,000synthetic images. Results indicated that the EnhancedVegetation Index, and spectral brightness and greenness are145