We present the first continuous map of rain-snow air temperature thresholds over the Northern Hemisphere land surface, underlining the spatial variability of precipitation phase partitioning. Land surface models typically discriminate between rain and snow using a simple, spatially uniform air temperature threshold, but observations indicate the threshold is not static. Our analysis of a 29-year observational dataset (n = 17.8 million) shows the threshold varies significantly, averaging 1.0°C and ranging from -0.4°C to 2.4°C for 95% of Northern Hemisphere stations, with continental climates exhibiting the warmest thresholds and maritime the coolest. Relative humidity exerts the strongest control on phase partitioning, while surface pressure plays a secondary role. Simulations indicate the selection of a rain-snow threshold introduces significant uncertainty to snowfall frequency and that including relative humidity as a predictor variable provides the greatest improvement to precipitation phase prediction between 0.6°C and 3.4°C, the interval where phase partitioning is most uncertain.
Station locations and elevations
Latitude, longitude, and elevation for all meteorological and precipitation phase observation stations used in the Jennings et al. (2018) manuscript.
jennings_et_al_2018_file1_station_locs_elev.csv
Precipitation phase and meteorological observations
Observations of precipitation phase and meteorological data used in the Jennings et al. (2018) manuscript. Please note, this is a cleaned, formatted version of the http://rda.ucar.edu/datasets/ds464.0/ dataset. Be sure to cite the original dataset as well if you choose to use these data in your work.
jennings_et_al_2018_file2_ppt_phase_met_observations.csv
Observed 50% rain-snow air temperature thresholds
50% rain-snow air temperature thresholds for 6,883 Northern Hemisphere stations. Snow frequency was calculated per 1°C temperature bin using observations of precipitation phase. A hyperbolic tangent was then fit to the frequency data and used to calculate the 50% rain-snow air temperature threshold. All NA values indicate there were not enough data to fit the hyperbolic tangent.
jennings_et_al_2018_file3_temp50_observed_by_station.csv
Simulated 50% rain-snow air temperature thresholds (hyperbolic tangent)
Map of simulated 50% rain-snow air temperature thresholds based on a bivariate binary logistic regression phase prediction model applied to 27 years of MERRA-2 reanalysis data. The probability of snowfall was predicted for each precipitation event in the record as a function of air temperature and relative humidity. For each MERRA-2 grid cell, the probabilities were binned by 1°C air temperature bins to produce a snow frequency curve. A hyperbolic tangent was then fit to the data and used to calculate the 50% rain-snow air temperature threshold.
jennings_et_al_2018_file4_temp50_raster.tif
Simulated 50% rain-snow air temperature thresholds (linear regression)
Map of simulated 50% rain-snow air temperature thresholds based on a bivariate binary logistic regression phase prediction model applied to 27 years of MERRA-2 reanalysis data. The probability of snowfall was predicted for each precipitation event in the record as a function of air temperature and relative humidity. For each MERRA-2 grid cell, the probabilities were binned by 1°C air temperature bins to produce a snow frequency curve. The 50% rain-snow air temperature threshold was then calculated using a linear regression on snow frequency between 0.5°C and 6.5°C. This approach was used for grid cells without enough snow and rain events to compute the threshold using a hyperbolic tangent.
jennings_et_al_2018_file5_temp50_linregr_raster.tif
R code for calculating observed 50% rain-snow air temperature thresholds
R code for computing the 50% rain-snow air temperature threshold using the station data in jennings_et_al_2018_file2_ppt_phase_met_observations.csv
jennings_et_al_2018_file6_precipphase_station_observations_code.R
R code for evaluating precipitation phase methods
R code for evaluating the performance of different precipitation phase methods using the station data in jennings_et_al_2018_file2_ppt_phase_met_observations.csv
jennings_et_al_2018_file7_precipphase_phasemethods_code.R
R code for mapping simulated rain-snow thresholds
R code for creating the simulated 50% rain-snow air temperature threshold maps (jennings_et_al_2018_file4_temp50_raster.tif and jennings_et_al_2018_file5_temp50_linregr_raster.tif).
jennings_et_al_2018_file8_precipphase_merra_threshold_simulation_code.R
R code for snowfall frequency uncertainty
R code for analyzing the sensitivity of simulated snowfall frequency to 18 different precipitation phase methods.
jennings_et_al_2018_file9_precipphase_merra_snowfall_frequency_sensitivity_code.R