Snow accumulation variability limits InSAR SWE retrieval

Abstract. Repeat-pass interferometric Synthetic Aperture Radar (InSAR) can retrieve changes in snow water equivalent (ΔSWE) from the phase delay of radar waves through new snow, but typical interferogram processing averages the phase over windows of tens of meters and assumes the ΔSWE signal is constant within them. We tested this assumption with four pairs of repeat airborne-lidar snow surveys over the Tuolumne basin and two NASA-ISRO SAR (NISAR) InSAR pairs. The lidar-derived ΔSWE power spectrum breaks at a consistent 25 m (IQR 21–30 m) length scale, so the field varies strongly within an 80 m multi-look window, and the spatially uniform assumption fails. We capture the resulting effects in a complex-valued sub-window factor M. The variance within the window shortens M and decreases coherence, and the ΔSWE skew within the window rotates M and biases the recovered phase from the mean ΔSWE phase. At L-band, this ΔSWE variability within the window alone lowers the basin-median coherence to 0.25–0.47, before any temporal or thermal decorrelation, while the retrieval bias causes at most millimeters of SWE error in windows that pass a coherence mask. The two NISAR pairs, one spanning a storm and one during a no-accumulation period, show coherence at 20 and 80 m pixel sizes consistent in space and time with these predictions. The effect of sub-window variability is frequency-dependent, lowering the higher frequency C-band coherence factor to a median of 0.10 while leaving the lower frequency P-band nearly unaffected (0.88). Because the decorrelation effect grows with the multi-look footprint size, processing interferograms finer than NISAR's standard 80 m posting recovers much of the L-band coherence loss, and a 20 m footprint nearly doubles the median coherence factor. Sub-window ΔSWE variability is, therefore, a sampling limit that can be predicted from high-resolution snow measurements and used to guide the choice of multi-look footprint for SWE retrievals.