A new snow reanalysis method is presented that is designed to jointly assimilate Landsat-and MODIS-derived (MODSCAG) fractional snow covered area (fSCA) to characterize seasonal snow in remote regions like High Mountain Asia (HMA) where in situ data is severely lacking. The method leverages existing readily available global datasets for forcing a snow model and uses the fSCA retrievals along with the ensemble prior model estimates to derive posterior estimates using a Bayesian framework. The method addresses MODIS viewing-geometry effects on the fSCA retrievals by accounting for viewing angle dependent measurement errors and using a CDF-matching technique to improve the joint fSCA measurement consistency before assimilation. The method was verified through comparison with the Airborne Snow Observatory (ASO) snow water equivalent (SWE) estimates over the Tuolumne River watershed in California. The posterior SWE estimates were shown to be much more consistent with the independent ASO estimates across the three WYs examined. Tests over Tuolumne showed that in cases where sufficient Landsat observations are available (i.e., with multiple sensors and in areas of overlapping Landsat tiles), assimilation of only Landsat data may be optimal, which is attributable primarily to the higher spatial resolution of the raw Landsat data, but that in cases with fewer Landsat measurements (i.e., with single Landsat tiles and/or significant reduction due to clouds), the additional screened and CDF-matched MODIS-based measurements can have a positive (albeit marginal) impact. Illustrative results are presented for nine HMA test tiles to illustrate how the method can provide posterior estimates of the space-time climatology of SWE storage in areas where in situ data does not generally exist. Ongoing work is being conducted to use the method outlined herein to generate an HMA-wide reanalysis dataset that will provide an opportunity for a more thorough characterization of HMA seasonal snow storage and dynamics over the joint Landsat-MODIS era. The method is generalizable to any midlatitude montane region where seasonal snow is important.

No abstract

Using a uniform model resolution over a domain is not necessarily the optimal approach for simulating hydrologic processes when considering both model error and computational cost. Fine‐resolution simulations at 100 m or less can provide fine‐scale process representation, but can be costly to apply over large domains. On the other hand, coarser spatial resolutions are more computationally inexpensive, but at the expense of fine‐scale model accuracy. Defining a multiresolution (MR) grid spanning from fine resolutions over complex mountainous areas to coarser resolutions over less complex regions can conceivably reduce computational costs, while preserving the accuracy of fine‐resolution simulations on a uniform grid. A MR scheme was developed using a physiographic complexity metric (CM) that combines surface heterogeneity in forested fraction, elevation, slope, and aspect. A data reduction term was defined as a metric (relative to a uniform fine‐resolution grid) related to the available computational resources for a simulation. The focus of the effort was on the snowmelt season where physiographic complexity is known to have a significant signature. MR simulations were run for different data reduction factors to generate melt rate estimates for three representative water years over a test headwater catchment in the Colorado River Basin. The MR approach with data reductions up to 47% led to negligible cumulative snowmelt differences compared to the fine‐resolution baseline case, while tests with data reductions up to 60% showed differences lower than 2%. Large snow‐dominated domains could therefore benefit from a MR approach to be more efficiently simulated while mitigating error.

Abstract. A multiresolution (MR) approach was successfully implemented in the context of a data assimilation (DA) framework to efficiently estimate snow water equivalent (SWE) over a large head water catchment in the Colorado River basin (CRB), while decreasing computational constraints by 60 %. A total of 31 years of fractional snow cover area (fSCA) images derived from Landsat TM, ETM+, and OLI sensor measurements were assimilated to generate two SWE reanalysis datasets, a baseline case at a uniform 90 m spatial resolution and another using the MR approach. A comparison of the two showed negligible differences in terms of snow accumulation, melt, and timing for the posterior estimates (in terms of both ensemble median and coefficient of variation). The MR approach underestimated the baseline peak SWE by less than 2 % and underestimated day of peak and duration of the accumulation season by a day on average. The largest differences were, by construct, limited primarily to areas of low complexity, where shallow snowpacks tend to exist. The MR approach should allow for more computationally efficient implementations of snow data assimilation applications over large-scale mountain ranges, with accuracies similar to those that would be obtained using ∼ 100 m simulations. Such uniform resolution applications are generally infeasible due to the computationally expensive nature of ensemble-based DA frameworks.

No abstract