Improving Parameterizations of Snow in the National Water Model with Observations from the New York State Mesonet, to Better Simulate Snow and Streamflow in the Northeastern United States

Snow processes in the operational version of NOAA's National Water Model (NWM) are simulated at a 1km resolution by the Noah-MP land surface model. Parameterizations in Noah-MP include precipitation phase, snow density, surface snow cover fraction, and albedo. Shortcomings in these parameterizations can introduce error due to over-simplification at scales relevant to flood forecasting and water management. The extent of these errors is often unknown and variable according to region due to differences in climatological forcing and snowpack history. Such parameterizations have not been as extensively evaluated over the northeastern United States (NEUS) as compared to the Western US, in part due to limitations in the snow observation network in the NEUS.

This study works to improve the snow state predication within the NWM through the use of automated high quality meteorological, snow depth, and snow water equivalent (SWE) observations from the New York State Mesonet (NYSM). Snow course measurements are being taken at five NYSM sites to characterize the representativeness of the automated snow depth and SWE measurements for the surrounding area. We hypothesize that improving the snowpack simulations in the NWM will improve the quality of medium range streamflow forecasts.

We initially compared simulated snow depth output from NWM retrospective runs against snow depth data from NYSM as a baseline test of model performance in the NEUS. Results showed a negative bias across New York State in version 2.0 of the NWM. Version 2.1 of the NWM substantially reduced the bias, however, there still exists substantial errors at specific sites and times of the year. Next, we performed point simulations of Noah-MP, forced by meteorological data from the NYSM, testing the sensitivity of snow model parameterizations options including those controlling precipitation partitioning and snowpack albedo. Model output was compared against snow depth and SWE measurements from the NYSM. Early results have shown a strong sensitivity to precipitation phase partitioning and weak sensitivity to albedo scheme changes. Going forward, we plan to run sensitivity experiments on snow and non-snow parameterization in the NWM to determine snow and streamflow sensitivity in an operational forecast setting.