The availability and variability of snowmelt has become a serious concern because of increased water demand, and because of the high degree of uncertainty related to climate variability posing a threat to the magnitude and timing of this precious resource. Understanding the geophysical controls and interannual variability of the spatial patterns of seasonal montane snowpacks are critical for understanding the effects of a warmer climate on the snowpack water storage. To explicitly resolve snow hydrological controls in complex montane environments, it is necessary to provide high resolution spatially and temporally distributed estimates of snow water equivalent, while also taking into consideration the uncertainties in the system. Toward this end, this dissertation developed a retrospective data assimilation technique (SWE reanalysis) that aimed to optimally merge VIS-NIR remote sensing data into a snow prediction model, and at the same time, account for the limitations of measurements, forcings, and model errors. The SWE reanalysis was: first developed and implemented over a small region, in order to investigate the performance of the methods under their nominal scenarios; second implemented for the full Landsat-5 record (27 year) over a regional scale domain in order to test accuracy and gain insight on the spatial and interannual controls on the SWE patterns; third extended to the entire Sierra Nevada in order to benchmark the reanalysis for its application to the full Sierra Nevada and to preliminarly [i.e. preliminarily] understand what are the spatial controls on SWE patterns. The key findings of this dissertation can be summarized as follows: 1) The SWE reanalysis approach provided accurate spatially and continuous estimates of SWE and of its uncertainties due to measurement, forcings, and model errors. 2) The methods were found to be robust to input errors such as biases in solar radiation and precipitation, and robust to the number of available VIS-NIR observations. 3) The application of the methods over the Kern watershed for the full Landsat-5 record suggested that SWE accumulation patterns were in general not interannually consistent and that the interannual variability was dependent on whether a dry or wet year was analyzed. 4) The trend test analysis showed that peak-SWE and day-of-peak have not drastically changed over the analyzed 27 years for the Kern River watershed, but suggested that the lower elevations may be more susceptible to climate variability and change. 5) Elevation was found to be the primary control on spatial patterns of peak-SWE and day-of-peak for the entire Sierra Nevada range; however different patterns were found across the watersheds of the Sierra Nevada depending on their location. Ultimately, the methods can be applied to the full Sierra Nevada and other montane regions over the modern remote sensing record to generate a dataset that should be useful to scientists and practitioners not only in hydrology, but other fields where seasonal snow processes are a key driver such as biogeochemistry, mountain meteorology, and water resource management.

Improving the estimation of snow water equivalent (SWE) in the Sierra Nevada is critical for the water resources management in California. In this study, we carried out an experiment to estimate SWE in the Upper Kern Basin, Sierra Nevada, by assimilating AMSR-E observed brightness temperatures (Tb) into a coupled hydrology and radiative transfer model using an ensemble Kalman batch reanalysis. The data assimilation framework merges the complementary SWE information from modeling and observations to improve SWE estimates. The novelty of this assimilation study is that both the modeling and the radiance data processing were specifically improved to provide more information about SWE. With the enhanced SWE signals in both simulations and observations, the batch reanalysis stands a better chance of successfully improving the SWE estimates. The modeling was at a very high resolution (90m) and spanned a range of mountain environmental factors to better characterize the effects of the mountain environment on snow distribution and radiance emission. We have developed a dynamic snow grain size module to improve the radiance modeling during the intense snowfall events. The AMSR-E 37GHz V-pol observed Tb was processed at its native footprint resolution at ~100 square km. In the batch assimilation, the model predicted the prior SWE and Tb; the prior estimate of an entire year was then updated by the dry-season observations at one time. One advantage of this is that the prior SWE of a certain period is updated using the observations both before and after this period, which takes advantage of the temporally continuous signal of the seasonal snow accumulation in the observations. We found the posterior SWE estimates showed improved accuracy and robustness. During the study period of 2003 to 2008, at point-scale, the average bias of the six-year April 1st SWE was reduced from -0.17 m to -0.02m, the average temporal SWE RMSE of the snow accumulation season decreased by 51.2%. The basin-scale results showed that the April 1st SWE bias reduced from -0.17m to -0.11m, and the temporal SWE RMSE of the accumulation season decreased by 23.6%.

Accurately estimating basin‐wide snow water equivalent (SWE) is the most important unsolved problem in mountain hydrology. Models that rely on remotely sensed inputs are especially needed in ranges with few surface measurements. The NASA Airborne Snow Observatory (ASO) provides estimates of SWE at 50 m spatial resolution in several basins across the Western U.S. during the melt season. Primarily, water managers use this information to forecast snowmelt runoff into reservoirs; another impactful use of ASO measurements lies in validating and improving satellite‐based snow estimates or models that can scale to whole mountain ranges, even those without ground‐based measurements. We compare ASO measurements from 2013 to 2015 to four methods that estimate spatially distributed SWE: two versions of a SWE reconstruction method, spatial interpolation from snow pillows and courses, and NOAA's Snow Data Assimilation System (SNODAS). SWE reconstruction downscales energy forcings to compute potential melt, then multiplies those values by satellite‐derived estimates of fractional snow‐covered area to calculate snowmelt. The snowpack is then built in reverse from the date the snow is observed to disappear. The two SWE reconstruction models tested include one that employs an energy balance calculation of snowmelt, and one that combines net radiation and degree‐day approaches to estimate melt. Our full energy balance model, without ground observations, performed slightly better than spatial interpolation from snow pillows, having no systematic bias and 26% mean absolute error when compared to SWE from ASO. Both reconstruction models and interpolation were more accurate than SNODAS.

The Sierra Nevada winter snowpack is the major water resource for the state of California. To better quantify the input of the water system, we deployed wireless-sensor networks across several basins in the Sierra Nevada. Together with operational and scientific research agencies, we also collected numerous scans of snow-on and snow-off lidar data over several basins in the high Sierra. We mined the lidar data and found how spatial patterns of snow depth are affected by topography and vegetation while elevation is the primary variable, other lidar-derived attributes slope, aspect, northness, canopy penetration fraction explained much of the remaining variance. By segmenting the vegetation into individual trees using lidar point clouds, we were able to extract tree wells from the high resolution snow-depth maps and we found the spatial snow distribution to be affected by the interactions of terrain and canopies. The snowpack is deeper at the downslope direction from the tree bole, however the snowpack at upslope direction being deeper. On sub-meter to meter scales, non-parametric machine-learning models, such as the extra-gradient boosting and the random-forest model, were found to be effective in predicting snow depth in both open and under-canopy areas. At spatial scales that are larger than 100 × 100 m2, we developed a novel approach of using the k-NN algorithm to combine the real-time wireless-sensor-network data with historical spatial products to estimate snow water equivalent spatially. The results suggest only a few historical snow-water-equivalent maps are needed if the historical maps can accurately represent the spatial distribution of snow water equivalent. The residual from the k-NN estimates can be distributed spatially using a Gaussian-process regression model. The entire estimation process can explain 90% of the variability of the spatial SWE.

Since 1982, under an agreement between the California Department of Water Resources and the USDA Forest Service, snow sensors have been installed and operated in Forest Service-administered wilderness areas in the Sierra Nevada of California. The sensors are to be removed by 2005 because of the premise that sufficient data will have been collected to allow "correlation" and, by implication, prediction of wilderness snow data by nonwilderness sensors that are typically at a lower elevation. Because analysis of snow water equivalent (SWE) data from these wilderness sensors would not be possible until just before they are due to be removed, "surrogate pairs" of high- and low-elevation snow sensors were selected to determine whether correlation and prediction might be achieved. Surrogate pairs of sensors with between 5 and 15 years of concurrent data were selected, and correlation and regression were used to examine the statistical feasibility of SWE prediction after "removal" of the wilderness sensors. Of the 10 pairs analyzed, two pairs achieved a correlation coefficient of 0.95 or greater. Four more had a correlation of 0.94 for the accumulation period after the snow season was split into accumulation and melt periods. Standard errors of estimate for the better fits ranged from 15 to 25 percent of the mean April 1 snow water equivalent at the high-elevation sensor. With the best sensor pairs, standard errors of 10 percent were achieved. If this prediction error is acceptable to water supply forecasters, sensor operation through 2005 in the wilderness may produce predictive relationships that are useful after the wilderness sensors are removed

Seasonal snowpack is a vital water resource that impacts downstream water availability. However, accurately estimating snow storage and characterizing its spatiotemporal distribution remain challenging, in particular for data-scarce regions such as High Mountain Asia (HMA). In this dissertation, a newly developed snow reanalysis method is used to estimate snow water equivalent (SWE) over the HMA region, assessing its spatiotemporal distribution and quantifying the regional snow storage. The method assimilates fractional snow-covered area (fSCA) from the Landsat and MODIS platforms, over the joint Landsat-MODIS record (Water Year (WY) 2000 - 2017). A fine resolution (16 arc-second, ~480 m) and daily High Mountain Asia snow reanalysis (HMASR) dataset is derived and analyzed in the dissertation. The key conclusions are summarized as follows: 1) Snowfall precipitation is found underestimated in most precipitation products with sizeable uncertainty when evaluated in sub-domains of HMA. The research shows the potential of using satellite snow observations as a constraint, to infer biases and uncertainties in snowfall precipitation in remote regions and complex terrain where in-situ stations are very scarce. 2) Through examining the HMASR dataset, the domain-wide peak seasonal snow storage is quantified as 163 km3 when aggregated across the full HMA domain and averaged across WYs 2000-2017, with notable interannual variations between 114 km3 and 227 km3. 3) Existing global snow products over HMA on average underestimate the peak snow storage by 33% 52% over the entire HMA, and the uncertainty in peak snow storage estimates is primarily explained by accumulation season snowfall (88%) over HMA, partly due to a wide range (uncertainty) in precipitation (snowfall). Ultimately, the snow storage and its spatiotemporal variations characterized in this work can be used to understand the role of seasonal snowpack in the regional climate and water cycle over this region.

In-situ measurements and numerical models were used to quantify and improve understanding of the processes governing snowpack dynamics in mountainous terrain. Three studies were conducted in Sequoia National Park in the southern Sierra Nevada, California. The first two studies evaluated and simulated the variability of observed melt rates at the point-scale in a mixed conifer forest. The third study evaluated the accuracy of a distributed snow model run over 1800 km2; a 3600 m elevation gradient that includes ecosystems ranging from semi-arid grasslands to massive sequoia stands to alpine tundra. In the first study, a network of 24 automated snow depth sensors and repeated monthly snow density surveys in a conifer forest were used to measure snow ablation rates for three years. A model was developed to estimate the direct beam solar radiation beneath the forest canopy from upward-looking hemispherical photos and above-canopy measurements. Sub-canopy solar beam irradiance and the bulk canopy metric sky view factor explained the most (58% and 87%, respectively) of the observed ablation rates in years with the least and most cloud cover, respectively; no single metric could explain> 41% of the melt rate variability for all years. In the second study, the time-varying photo-derived direct beam canopy transmissivity and the sky view factor canopy parameter were incorporated into a one-dimensional physically based snowmelt model. Compared to a bulk parameterization of canopy radiative transfer, when the model was modified to accept the time-varying canopy transmissivity, errors in the simulated snow disappearance date were reduced by one week and errors in the timing of soil water fluxes were reduced by 11 days, on average. In the third study, a distributed land surface model was used to simulate snow depth and SWE dynamics for three years. The model was evaluated against data from regional automated SWE measurement stations, repeated catchment-scale depth and density surveys, and airborne LiDAR snow depth data. In general, the model accurately simulated the seasonal maximum snow depth and SWE at lower and middle elevation forested areas. The model tended to overestimate SWE at upper elevations where no precipitation measurements were available. The SWE errors could largely be explained (R2/super” 0.80, p0.01) by distance of the SWE measurement from the nearest precipitation gauge. The results suggest that precipitation uncertainty is a critical limitation on snow model accuracy. Finally, an analysis of seasonal and inter-annual snowmelt patterns highlighted distinct melt differences between lower, middle, and upper elevations. Snowmelt was generally most frequent (70% - 95% of the snow-covered season) at the lower elevations where snow cover was ephemeral and seasonal mean melt rates computed on days when melt was simulated were generally low (3 mm daysuper-1). At upper elevations, melt occurred during less than 65% of the snow-covered period, it occurred later in the season, and mean melt rates were the highest of the region ( 6 mm daysuper-1/super). Middle elevations remained continuously snow covered throughout the winter and early spring, were prone to frequent but intermittent melt, and provided the most sustained period of seasonal mean snowmelt (~ 5 mm day