SNOTEL Water Level Analysis
In my Data Science Immersive course at General Assembly we recently worked on group projects, it was our first experience doing this kind of work within a group environment and it was a new challenge to not only look at dividing up the work amongst ourselves, but also make sure that we were able to keep our group members up on our work by maintaining a good git repository. Luckily my group was careful about this and we did not run into any issues and we were able to get all of our work together.
For our topic we looked at precipitation information in the western United States to see if we could use and ARIMA model to predict precipitation levels over the next few years. The Natural Resources Conservation Service (NRCS) uses an automated system to collect snowpack and climate data in the Western United States known as SNOTEL (SNOwpack TELemetry). Growing from a manual measurement system SNOTEL has been reliably collecting data to produce water supply forecasts and support resource management activities since 1980. SNOTEL uses meteor burst communications to collect and communicate data in near real time without the use of satellites. There are more than 730 SNOTEL sites in 11 states, all designed to operate without maintenance for a year as they are typically in remote locations and maintenance trips can involve long hikes or helicopter trips. The NRCS National Water and Climate Center in Portland, Oregon houses the central computer that controls operation of the sites and receives the data gathered.
The data we used spanned over the past 30 years, we focused on one date in particular, only February 10th. Using this information we were building a model to predict the level of precipitation on that certain date using previous years data for that date.
After cleaning and preparing the data we started running our models, we decided to break the large basin into each of the sub-basins to predict the levels in each area as that would be more useful than just figuring an average of the whole large basin. Most of our models turned out pretty well, for instance:
Not all of our models looked quite as good, though.
Overall, we found that our models in general worked fairly well for predicting precipitation levels with the data from SNOTEL.
More information on SNOTEL can be found at the following links
SNOTEL Data Collection Network Fact Sheet
Snow Telemetry and Snow Course Data and Products