Soil emissivity of Arctic regions is a key parameter for assessing surface properties from microwave brightness temperature (Tb) measurements. Particularly in winter, frozen soil permittivity and roughness are two poorly characterized unknowns that must be considered. Here, we show that after removing snow, the 3D soil roughness can be accurately inferred from in-situ photogrammetry using Structure from Motion (SfM). We focus on using SfM techniques to provide accurate roughness measurements and improve emissivity models parametrization of frozen arctic soil for microwave applications. Validation was performed from ground-based radiometric measurements at 19 and 37 GHz using three different soil emission models: the Wegmüller and Mätzler (1999) model (Weg99), the Wang and Choudhury ( 1981) model (QNH), and a geometrical optics model (Geo Optics). Measured and simulated brightness temperatures over different tundra and rock sites in the Canadian High Arctic show that Weg99, parametrized with SfM-based roughness and optimized permittivity ( , yielded a RMSE of 3.1 K ( ) 𝜀)𝑅 2 = 0.71 for all frequencies and polarizations. Our SfM based approach allowed us to measure roughness with 0.1 mm accuracy at 55 locations of different land cover type using a digital camera and metal plates of know dimensions.

Increased surface temperatures (0.7°C per decade) in the Arctic affects polar ecosystems by reducing the extent and duration of annual snow cover. Monitoring of these important ecosystems needs detailed information on snow cover properties at resolutions (<100 m) that influence ecological habitats and permafrost thaw. A machine learning method using topographic parameters with the Random Forest (RF) algorithm previously developed in alpine environments was applied over an arctic landscape for the first time. The topographic parameters used in the RF algorithm were Topographic Position Index (TPI) and up‐wind slope index (Sx), which were estimated from the freely available Arctic DEM at 2 m resolution. Addition of an ecotype parameter (proxy for vegetation height) showed minimal predictive improvement. Using RF, snow depth distributions were predicted from topographic parameters with a root mean square error = 8 cm (23%) (R2 = 0.79) at 10 m resolution for an arctic watershed (1500 km2) in western Nunavut, Canada.

Abstract. Topography and vegetation play a major role in sub-pixel variability of Arctic snowpack properties but are not considered in current passive microwave (PMW) satellite SWE retrievals. Simulation of sub-pixel variability of snow properties is also problematic when downscaling snow and climate models. In this study, we simplified observed variability of snowpack properties (depth, density, microstructure) in a two-layer model with mean values and distributions of two multi-year tundra dataset so they could be incorporated in SWE retrieval schemes. Spatial variation of snow depth was parameterized by a log-normal distribution with mean (μsd) values and coefficients of variation (CVsd). Snow depth variability (CVsd) was found to increase as a function of the area measured by a remotely piloted aircraft system (RPAS). Distributions of snow specific surface area (SSA) and density were found for the wind slab (WS) and depth hoar (DH) layers. The mean depth hoar fraction (DHF) was found to be higher in Trail Valley Creek (TVC) than in Cambridge Bay (CB), where TVC is at a lower latitude with a subarctic shrub tundra compared to CB, which is a graminoid tundra. DHFs were fitted with a Gaussian process and predicted from snow depth. Simulations of brightness temperatures using the Snow Microwave Radiative Transfer (SMRT) model incorporating snow depth and DHF variation were evaluated with measurements from the Special Sensor Microwave/Imager and Sounder (SSMIS) sensor. Variation in snow depth (CVsd) is proposed as an effective parameter to account for sub-pixel variability in PMW emission, improving simulation by 8 K. SMRT simulations using a CVsd of 0.9 best matched CVsd observations from spatial datasets for areas > 3 km2, which is comparable to the 3.125 km pixel size of the Equal-Area Scalable Earth (EASE)-Grid 2.0 enhanced resolution at 37 GHz.