Calculating snowmelt backwards - using the date of snowpack disappearance to determine how much snow fell over a season

The amount of water available in a snowpack, snow water equivalent (SWE), is an important parameter for many communities, such as water resource managers, climate scientists, and the agricultural sector. However, spatial patterns of SWE between observation stations are often unknown and precipitation gauge errors present challenges in acquiring reliable snowfall data. One method of reconstructing the seasonal accumulation of SWE is by summing estimated snowmelt from the start of the ablation season to the date when snowpack disappears (determined by remote sensing or in-situ surface observations). This reconstruction method has been applied in several previous studies, but no study has tested the method at control sites to identify its accuracy in estimating seasonal snowfall accumulation. In the present study, the reconstruction method was evaluated at 175 mountain weather stations in the maritime mountain ranges of Washington, Oregon, and California across 13 water years. Snowmelt was calculated using the Snow-17 operational snow model, the temperature-index Snowmelt Runoff Model with a net radiation component, and the full energy balance Utah Energy Balance model. The reconstruction method was applied by iteratively running each snowmelt model and adjusting the precipitation data with a scaling factor until the simulated snowpack disappeared on or near the observed disappearance date. A baseline collection of simulations was also produced by forcing each model without adjusting the precipitation data. These results were compared to the reconstruction method results to evaluate whether the reconstruction method improved simulations of SWE. Simulations of snowmelt were forced with observed daily, simulated hourly, and observed hourly data sets to determine what type of data was necessary to reconstruct SWE accurately. The reconstruction approach only produced more accurate results than baseline simulations when Snow-17 was driven by observed hourly data. The median results of the Snowmelt Runoff Model and Utah Energy Balance simulations demonstrated that errors from melt overestimation with these models were generally much greater than errors associated with precipitation gauge undercatch. Because the reconstruction method estimates SWE based on snowmelt energy exchanges, it will become more reliable as snow models improve.