Climate Change and Snowpack

SNOTEL Stations

SNOTEL: “Snow Telemetry”
Telemetry: Automated Communications Process
114 Stations
Elevations: 8397 to 11618
Provided by the National Resources Conservation Service (NRCS)

Climate Change

Colloborated with Denver Channel 2 Meteorologist, Matt Makens
Suggested I analyze how warming climate impacts SNOTEL snowpack at different elevations
Suggested I used data provided from NRCS

Key Terms

Snow Water Equivalent: How much precipitation in snow
Snow Ratio: Linear coefficient between Snow Depth and SWE
Snow Depth: Measured by “snow pillows” via hydrostatic pressure
Snow Ratios were important: Since we always have SWE, but not Snow Depth, the Snow Ratio was the key in predicting Snow Depth

Data Collection

Queried weather data from NRCS website : 75480 rows
Measurements are semi-monthly, or roughly every 15 days
SNOTEL station metadata obtained by snotelr R package
1978 - 2018
Data Cleaning: format Date, selected useful columns
Useful columns: Date, Year, Station Name, SWE, Snow Depth, Max, Min, Elevation, Lat, Long, County

Exploring Temperature

Many missing or impossible temperatures
Some stations defaulted to 32 for both daily min and daily max
Daily max: decided to impute top and bottom .8% values, intuitively were outliers
Daily min: top .8%, bottom 1.6%

Imputing Temperature

Used middle 98% as training data
Tried KNN regressor in Python: good (.86 r^2, 585737 SSE) but only month/elevation combinations
Decided to do custom KNN: mean of rows returned with within 90 miles, 1000 feet elevation, same date (or month)
Validated: most residuals within 3 degrees (used high temp measurements), SSE = 75190

Snow Depth Data Problems

Most stations didn’t report snow depth until 2003
NRCS at data entry inputted 10:1 or 5:1 snow ratios when problematic values are observed
Snow ratio trends: variability within ratio with month, temperature, but not elevation
Snow ratio used to help backtrack snow depth values through SWE.

Exploring Snow Depth

Many snow ratios were improbable
Early/Late season snow ratio is higher

Finding Regression

Method: Use average temperature, SWE, and Month to explain snow ratio patterns
Average temperature: (High + Low)/2
Explored 3 models, used BoxCox method to pick Y transformation
Decided to apply square root transformation to both SWE and Snow Depth(y)
Model (70% data) = .94, Validation (30% data) = .94, Full Data = .94
StDev around 5 inches, 95% of data within 10 inches of actual depth

Imputing Snow Depth

Passed month, SWE, and average temp through predict() for prediction
90.8% of June-October observations had no snow
41.2% of observations had no snow

Climate Change and Snow Depth

Increase temperature means smaller snow season
Snow season has decreased over last 40 years
At a glance, warming climate has slightly more negative affect on lower elevation snowpack
Nightly low temperature steadily increasing
Daily temperature disparity is shrinking
Snowpack decrease over last 40 years
Decrease seems mostly constant across elevations