Seasonal Water Supply Forecasting in the Western U.S. Under Declining Snowpack
AltmetricsView Usage Statistics
Seasonal water supply forecasts (i.e. April-July streamflow volume predictions) are a key decision-support tool for water resource managers overseeing reservoir operations, water rights allocation, and ecosystem and wildlife protection, particularly in semi-arid regions such as the western U.S. However, extremely low snowpack and non-stationarity due to climate change challenge the empirical relationships and computer simulations used by operational hydrologic forecasters. In this thesis, we present findings from two studies that use empirical and simulated data to recreate these operational models across two separate study domains: 51 minimally-impacted, snow-dominated basins in the western U.S. and 26 headwater basins in the Sierra Nevada of California and Nevada. In the first study, we use a model benchmarking approach to evaluate and compare the retrospective performance of two of the most widely used water supply forecasting models in the western U.S., the USDA Natural Resource Conservation Service’s (NRCS) Principal Component Regression models and the NOAA National Weather Service’s (NWS) SNOW17/Sacramento Soil Moisture Accounting models. Results from this retrospective analysis across water years 1981-2014 show that statistical models (NRCS PCR) are generally more skillful than conceptual models (NWS SNOW17/SAC-SMA). However, at longer forecast lead times, are results suggest that statistical models are more disadvantage by low snowpack, making them comparatively less skillful than the conceptual models. In our second study, we extend this analysis by using hydrologic simulations from the Variable Infiltration Capacity model to mimic California Department of Water Resources water supply forecasting procedures in the Sierra Nevada from 1950-2099 under low and high emissions scenarios. Results show widespread but uneven loss of skill by the second half of the 21st century (average loss 10-20%). Basins at mid-elevations (mean elevation 1000-1700 meters) were simulated to be most vulnerable to loss of snowpack and the forecast information it provides. Within the model environment, we then evaluate mitigation strategies to buffer loss of forecast skill through the introduction of supplemental synthetic observations. These two simulated datasets include 1) basin-wide snowpack measurements (representing remote sensing products) and 2) soil moisture station observations. Our research suggests that remotely-sensed snowpack data may buffer loss of skill by an average of 40% in the most vulnerable basins before 2050. As the century passes and the role of snow in Sierra Nevada hydrology declines, there will eventually be dwindling returns from remotely-sensed snowpack data, but this loss of forecast information may be somewhat ameliorated by the inclusion of soil moisture observations in the regression equations.