So… How Much Glacier Are We Losing?
As a final semester project, looking at glacier extent over the decades of 1979 to 2020 was something my group and I thought would be an exceptionally fun and insightful way to learn about time series forecasting as well as the impact of climate change. Over the decades, glacier extent has dramatically changed over the years, and it seems almost always a topic of discussion in climate change.
“The ice-bergs are melting!”, “Polar bears are losing their homes!”, “Water is rising from glaciers melting!”.
Within the first 10 observations (years in our data) and the last 11 years (2008–2019) in only the month of September, we see quite a big change. Variations of glacier extent ranged largely in the high 7 millions per square kilometer whereas in the past 11 years, glacier extent has reached as low as 3 millions per square kilometer and ranged consistently within 4 millions per square kilometer. Indonesia is approximately 1.9 millions per square miles in land and glacier extent, since the late 1970s, have decreased to about TWICE as big as that. That’s a huge difference! (For the remainder of our project, we’ve used all months over the decade, but this is to show how dramatically glacier extent has changed even in the same month every year).
It was in our interest to investigate and analyze glacier extent using different models and factors to better understand how and why glacier extent moves the way that it does. To clarify, glacier extent is the amount of glacier covering the area of the earth. This means that lower glacier levels is when the glacier covers the least amount of earth’s surface and the higher the levels are means more glacier present.
When figuring out which data to use for our time series forecasting project, finding data that consisted of the year and month was almost as hard as finding a needle in a haystack. Many of the data available on Kaggle had “date” measured in only years. We wanted data that was grouped by their year while also providing the month the information was measured in instead of duplicates of multiple data for the same year. This would provide us a better analysis of any trends that could be identified as yearly, monthly, or maybe even seasonality, thus, giving us time-series data to work with. Luckily, we had found the perfect data provided by the National Snow & Data Center which encompassed everything we needed to understand glacier extent.
The blue cyclical patterns indicate all of our glacier data on its extent from January 1979 to January 2020.
By itself, the data showcases a steady decline in extent with obvious dips and rises in between the decades as seen in the trendline (in red). Unlike the blue patterns that are to show the overall patterns of glacier extent, the trend line helps us visualize variations in how the glacier extent has been rising and decreasing but still having an overall decreasing pattern over the years. To further emphasize and understand these patterns, a Seasonal and Trend Decomposition method was used to divide the patterns into smaller chunks.
The three components that we looked for first in our time series was the…
- trend: the gradual change in the data. This is the one that is most simplest and provides us a clear picture of the increases and decreases in the time series.
- seasonality: repeating patterns that are seen within the same time period (ex. every 3 months, yearly, monthly, etc).
- noise: randomness in our data.There are multiple types of decomposition methods; two others being SEATS and X11. However, STL was used in this project to work with any type of seasonality trends outside of just monthly and quarterly data.
In the case of our time series, our seasonal trend could possibly be more yearly than it is, for instance, quarterly (ex. every fall no matter the year, the glacier extent decreases at a consistent pace over the years).
In this STL decomposition method, “Extent” produces our original data which is made out of the trend, seasonality, and the noise patterns. When taking away, for example, “trend” from our “Extent”, we are left with our 3rd patterns which reveals the seasonality trend of our data. This can be also seen in producing the “remainder” which is our “Extent” graph without the “trend” and “seasonality” patterns. Looking more closely at our seasonality pattern, we can see that it almost looks identical to our original graph, “Extent”; however, the cycles seem taller vertically while also having a subtle increase in vertical size as the years go by whereas “Extent” shows more consistent cyclical patterns with an overall decreasing trend. What the “seasonality” pattern reveals to us is that glacier extent has a consistent trend throughout the seasons. That means, glacier extent moves in a predictable manner of being the highest in the winter and melting gradually and the most as warmer weather approaches (spring and summer), thus, being at its lowest point the beginning of fall. Below is a clearer view of this trend:
Glacier extent within all of the years shows similar patterns over time while maintaining less and less overall glacier extent. When comparing each monthly extent with the average extent in January, a quick analysis showed that from January to March, an average of 0.99 millions per square miles grew in glacier extent. However, from January to September, glacier extent had dropped to an average of 7.949 millions per square miles. That’s almost 4X as big as Indonesia!
And could there also be changes to glacier extent that aren’t consistent throughout the years? Such as, has a disruption in the Artic-environment led to more drastic changes in one year…? To understand more about unique changes, our noise patterns can provide us some insight into individual changes and their importance in the overall data. According to Figure 2, there are visible spikes, one being between 1995 to 2000 as well as 2005 to 2015 where a sharp increase sometimes after 1995 and sharp decreases sometimes around 2007 and 2013 can be seen. These trends aren’t visible within our seasonality and “Extent” graphs but can be interpreted using the grey box beside each one. The grey box beside our noise graph shows that it is 1.5 units long. When comparing this to our original data, “Extent”, the noise patterns are almost negligible, meaning that the noise isn’t doing much in influencing overall glacier extent.
But what could be influencing glacier extent, you might ask? Especially when looking at Figure 3, the black lines seem to indicate that, overtime, glaciers have been growing more in the colder months but also melting more in the warmer months. Is the Artic getting more colder during the winter time but hotter during the summer time?
A bit of google research led to an article written by Trevor Nace called “Artic Sea Ice Is Growing Faster Than Before, But There’s a Catch”. He explained that sea ice has, in fact, grown faster and higher over the past decades but warmer temperatures have increased as well. Although sea ice has been larger than they were before, the warmer temperatures have off-set this growth and led to an overall decrease in sea ice in the Artic. As Trevor states,
“Temperatures in the Arctic have warmed much faster than temperatures in tropical locations. The doubled rate of warming has led to increasingly smaller sea ice extents during Arctic summer months and an overall reduction in sea ice.”
This largely explains our data’s black lines (Figure 3) that shows dramatic increases in recent years in glacier extent (compared to the 1970s) in the winter months as well as the drop in extent in warmer weathers while averaging less and less glacier overall. Global warming is REAL, y'all!
But how REAL is it? An analysis utilizing time series forecasting with CO2 as a predictor was produced by one of my group member’s on the extent that increases of CO2 can have on the future of our glaciers.
According to Figure 4, when CO2 is predicted to reach historical levels consistent to past years, glacier extent would fall, on average, within the red lines. Similarly, if CO2 is forecasted to increase to about 400 ppm (parts per million volume) on average, glacier extent will fall at the forecasted blue line with 80% chance of falling into the darker shaded blue and 95% within the lighter shaded blue. In hopes that CO2 doesn’t reach extreme cases, glacier extent would fall within the green line with 80% predicted to be anywhere within the darker green and 95% within the lighter green area. THAT’S SCARY! As we all know, global carbon emissions have significantly increased in the past hundred years and thus it is inevitable that our glaciers will become dramatically thinner (or maybe even disappear) sooner than later!! AHH!!! FORTUNATELY, while CO2 isn’t getting better by the day, glacier extent DID NOT go below 0 in 2010.
Takeaways
Though it wasn’t hard to come up with a topic my group and I were all curious and passionate about, it was finding the right data due to specific “restrictions” that were.
Not many data were available that had month, day, and year while also satisfying our topic (as I mentioned earlier) online. Kaggle had a large selection of time series data but none of them seemed to fit the category of “glacier” while also having month and day by year. Though we were lucky to find one for our topic of interest, sometimes original time series project ideas may have to wrangle many data from different sources into one or even start from scratch. But this is common in working with data and isn’t specific to time series in any way. You might like the color, but the shoe may not fit!
Other technical problems include understanding how to turn our data into a time series data which took a lot of google searches and textbook readings. However, other than that, this project was smooth sailings and I’ve become more sensitive into my own contributions to the inevitable demise of our world…
On a lighter note though, this project has been one of the most fun I’ve had at Smith College. Never in a million years had I thought about Data Science the way that I do now. I’ve learned to love working with others and be okay in asking for help. I’ve also learned how personal this major can be to those who are natural healers and helpers…but also math nerds too! There is more to healing people and the world than extending a hand. Sometimes it’s discovering what’s wrong in seemingly obvious places and bringing awareness for change. In the case of Data Science, sometimes it’s largely making it easier to see trends and creating predictions! Woo-Hoo!!
