The Sierra Nevada system plays an integral role in the hydrologic cycle, energy cycle, ecological systems, and in water resources supply in Western North America. Trends toward earlier spring snowmelt runoff and decreasing springtime observations of snow water equivalent (SWE) from in situ networks have been observed in the past fifty years. These trends will result in both increased likelihood of winter floods and decreased water availability. It is therefore urgent to develop detailed, process-based understanding of the observed changes. Moreover, it is vital to develop a spatially-continuous characterization of these trends, spanning the physiographic gradients present in the Sierra Nevada. Not all available information has been used in the diagnosis of the trends, viz. spaceborne remote sensing measurements. This means that trends have only been evaluated where particular stations are located (not over the entire Sierra Nevada). Moreover, formal estimates of observation uncertainty have not been taken into account in the assessment of the trends. Detailed understanding of the physical processes and complex sensitivity to climatic change across physiographic gradients has not been achieved. We thus propose a reanalysis utilizing all available datasets, including both in situ and remote sensing measurements. Our motivation is the unique and complementary characteristics of each of four primary datastreams: in situ snow measurements, streamflow, snow-covered area (SCA) derived from visible and near-infrared data, and passive microwave measurements. We will use ensemble model simulations of snowpack physical processes to provide a priori estimates of snowpack variables. Measurement models will be used to relate snowpack states to all four primary datastreams. The data assimilation analysis step calculates an a posteriori estimate of snowpack variables that takes into account all four datastreams, as well as meteorological data and scientific knowledge of snow physics processes. Estimates of the spatiotemporal uncertainty in each snowpack variable can be calculated directly from the ensemble. Trends will be assessed using these posterior estimates of snowpack states. During our two-year project, we will perform the reanalysis described above for the King River and Kaweah River basins in the Southern Sierra Nevada. Each of these river basins overlaps spatially with study areas from the Southern Sierra Critical Zone Observatory (https://snri.ucmerced.edu/CZO). The Providence CZO in the King River basin and the Wolverton CZO in the Kaweah River basin represent locations where ongoing research is leading to a deeper process-level understanding of the snow cover across elevation gradients, which may provide insight into the long-term trends that have been observed. We will focus this two-year project on developing solid methodology that could be applied to the entire Sierra Nevada in follow-on work. We will focus on development of the statistical and physical models of spatial variability and uncertainty that form the core of the reanalysis assimilation scheme. In particular, we will focus on development of models for relating observing network point measurements of snow water equivalent to grid-based estimates of snow properties, and on parameterization of the uncertainty models for the hydrometeorological inputs to the snow physics models. We will integrate the models, methods, and results from the work into existing undergraduate and graduate courses at UCLA, and into ongoing outreach work of the Byrd Polar Research Center. Additionally, the high-resolution modeling framework developed here could provide a unique testbed for future climate change studies, where atmospheric model output from regional or global climate models could be used as forcing to the offline model to add to the existing literature on how the Sierra snowpack and spring streamflow is expected to change.
Snow plays a critical role in hydrology and water resources, but snow properties in mountainous areas vary dramatically in space, making characterization difficult: in situ measurements represent a single point of spatially-variable snow properties. Remote sensing via spaceborne measurements is thus an attractive resource to utilize in estimating snow properties in mountainous regions. In this project, two kinds of remote sensing measurements were explored. Spaceborne passive microwave (PM) remote sensing has coarse spatial resolution, but is directly sensitive to the amount of snow on the ground. The minimum Tb of each water year (WY), which starts from October 1 of a calendar year and ends at September 30 of the next, showed a strong inverse relationship to SWE measured in situ for the Kern River basin; the correlation coefficient between Tb and an in situ basin average SWE was −0.94. A major advance was processing raw Level 2A data at 88 km2 resolution, rather than the commonly-used EASE-Grid data at 625 km2 resolution. It was also found that the Tb data were highly sensitive to melt timing, which varies significantly from year-to-year. Visible and near-infrared (VisNIR) data are much higher resolution (90 m – 500 m) but are sensitive only to the presence or absence of snow, not its depth or water equivalent (SWE). Reconstruction methods use VisNIR data to reconstruct SWE. We compared two reconstruction techniques (a commonly used deterministic reconstruction vs a probabilistic data assimilation framework). The methods retrospectively estimate SWE from a time series of remotely sensed maps of fractional snow-covered area (FSCA). In testing both methods over the Tokopah watershed in the Sierra Nevada (California), the probabilistic reconstruction approach is shown to be a more robust generalization of the deterministic reconstruction. Under idealized conditions, both probabilistic and deterministic approaches perform reasonably well and yield similar results when compared with in situ verification data, whereas the probabilistic reconstruction was found to be in slightly better agreement with snow-pit observations. More importantly, the probabilistic approach was found to be more robust: unaccounted for biases in solar radiation impacted the probabilistic SWE estimates less than the deterministic case; the probabilistic reconstruction was found to be less sensitive to the number of available observations, as well as to the prior precipitation accuracy. The additional robustness of the probabilistic SWE reconstruction technique should prove useful in future applications over larger basins and longer periods in mountainous terrain.
