1075 Development of Gridded Snow Depth Climatology in Eurasia

Wednesday, 25 January 2017
4E (Washington State Convention Center )
Lawrence M. Vulis, City College, New York, NY; and N. Devineni, S. R. Helfrich, and C. Kongoli

Near real time observations of snow depth are used as inputs to Numerical Weather Prediction (NWP) models at National Centers for Environmental Prediction (NCEP) and other climate and prediction agencies.  Recently, an operational blended snow depth analysis has been developed at NOAA and integrated into the Interactive Multisensor Snow and Ice Mapping System (IMS). The analysis blends snow depth from synoptic stations and from satellite passive microwave observations using a 2-Dimensional Optimal Interpolation methodology.  Here, spatial distribution of snow depth is modeled as a function of distance and elevation. The goal of this work is to develop a 1-km gridded snow depth climatology for use in the blended algorithm to improve estimates over poorly monitored high-elevation regions. .

A 1-km snow depth monthly climatology is developed over the Former Soviet Union using data from the All-Russia Research f Hydrometeorological Institute with 589 in-situ stations providing SD and temperature observations, and the WorldClim dataset providing gridded mean temperature observations. The grid was generated using two methodologies: k-Nearest Neighbors (kNN) regression and local polynomial regression. The kNN generated grid values by measuring the Mahalanobi’s distance of 4 predictor parameters (Latitude, Longitude, Elevation, and mean Air Temperature), of each grid point with all the stations and then selecting the k neighbors which are closest. The SD observations at the k neighbors are then given probability weights and 1000 weighted SD values are simulated for the station, and the median is selected as the simulated SD value. The local polynomial generated grid values by finding k neighbors and then fitting a polynomial to only those neighbors, using the same predictors.

Validation was conducted on the regression models using Leave One Out Cross-Validation on the stations. The initial results indicate that winter months tend to be more accurate, while summer and late spring months are not as accurate. There is still under-prediction at high-elevations. The introduction of predictors that reflect properties not currently being modelled, such as slope or aspect may also be useful. The results show promise for a global climatology.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner