A spatiotemporal analysis of errors in InSAR SWE measurements caused by non-snow phase changes

Abstract

Spatially distributed measurements of snow water equivalent (SWE) in mountainous terrain are not currently feasible from existing satellite platforms. The NISAR satellite has the potential to provide high resolution (80 m) SWE measurements on a 12 d orbit cycle over many of Earth’s snowy regions, which would represent a new era of spaceborne snow monitoring. The most promising approach for NISAR SWE measurements uses interferometric synthetic aperture radar (InSAR) techniques to derive the 12 d change in SWE (ΔSWE) from the change in phase between two SAR acquisitions. However, many non-snow factors can also change in this 12 d period which subsequently modulate the SAR phase. These non-snow factors can vary differently in both space and time, and in turn introduce spatially and temporally variable errors into InSAR-derived ΔSWE measurements. Here we explore the effects of six non-snow factors that can affect InSAR phase: electron content of the ionosphere, atmospheric water vapor, atmospheric pressure, soil permittivity, vegetation permittivity, and surface deformation. We show how these factors affect phase-based SWE measurements at 13 SNOTEL stations across the western US, as well as regionally across North America. We consider errors resulting from individual 12 d periods, as well as the cumulative effects of the errors when a timeseries of ΔSWE measurements is integrated to derive peak seasonal SWE. Ionosphere effects result in the largest cumulative error at all SNOTEL stations in our analysis, with changes in the total electron content resulting in phase changes equivalent to 0.271–0.414 m of SWE, or more than 500 % larger than the median 1 April SWE at some shallow snow stations. When ionosphere effects are removed, the remaining cumulative error ranges from -0.074–0.022 m of SWE, equivalent to 0 %–89 % of 1 April SWE. Relative error results are affected primarily by differences in peak SWE rather than differences in absolute error values. For a randomly selected 12 d period, exceedance probability analysis shows that there is a 50 % chance the ionospheric component introduces an error larger than 0.211 m into the overall ΔSWE measurement, while the remaining five components have a 50 % exceedance probability of 0.031 m. We also find that individual error components can show offsetting effects, where positive and negative biases partially cancel out to lower the total cumulative error. Accurate ΔSWE measurements using NISAR data will not be possible unless ionospheric effects can be appropriately addressed. Removal of other error sources requires careful consideration of the SWE monitoring application: for tracking seasonal SWE accumulation in areas with deeper snowpacks, correcting some errors but not others may actually decrease accuracy by removing offsetting cumulative effects. For individual 12 d periods, wet and dry tropospheric effects (due to changes in water vapor and pressure) should be removed for accurate interpretation of spatial patterns of snow accumulation at basin to range scales, and site-specific factors should be considered to assess the relative influence of vegetation, soils, and surface deformation.