UCLA

Accurately representing the spatial variability of montane snowpack is challenging due to the high degree of complexity in the terrain's topography, and the lack of good quality in-situ data. To explicitly resolve snow processes in montane environment while taking into account the uncertainties in the system's high spatial and temporal resolution snow reanalyses are required. Ensemble-based approaches assimilating Landsat VIS-NIR remote sensing data at spatial resolutions of 100 m or less are, however, prohibitively expensive to run at large scales, and sub-optimal given that only the most complex parts of a montane range require such fine resolution. In addition, the assimilation of remote sensing data from a single source can also be unsatisfactory due to a lack of global coverage, hardware malfunction etc. In order to optimize computational needs while preserving the accuracy of ~ 100 m reanalyses, a raster-based multi-resolution approach was first developed and successfully implemented for a headwater catchment in the Colorado River Basin over the full length of Landsat record (30+ years). The potential use of MODIS-derived snow cover information in addition to Landsat in snow reanalyses was then investigated over three different regions in the Western U.S. and High Mountain Asia in order to make up for Landsat's shortcomings over midlatitude snowpacks. The key findings of this dissertation can be summarized as follows: 1) The physiographic complexity of a terrain can be characterized by its standard deviations of elevation, northness index and forested fraction. Using such a complexity metric to discretize the terrain into different spatial resolutions via a multi-resolution approach can significantly reduce computational needs, while mitigating errors in snow processes representation. 2) The multi-resolution approach did not significantly impact the remote sensing observations assimilated, and the posterior snow water equivalent (SWE) ensemble median and standard deviation matched the 90 m reanalysis, thus leading to a robust implementation in the context of a data assimilation framework. 3) A MODIS-derived snow cover product was found to be a useful complementary source of remote sensing data to be simultaneously assimilated with Landsat. Off-nadir-looking observations first had to be screened out of the reanalysis due to the distorting and snow obscuring effects of the sensor viewing geometry at high zenith angles. Ultimately, the methods developed in this work can be applied to all midlatitude montane ranges over the full lengths of Landsat and MODIS records to generate a fine spatial resolution SWE reanalysis dataset that will be useful to snow hydrologists to solve many unanswered science questions over such challenging regions.