With my datasets cleaned and wrangled, I uploaded them into Tableau to begin visualizing the data for analysis. I began my analysis by looking at average temperatures over time to understand general patterns in Boston temperatures. I looked at these using a line chart, plotting average maximum and minimum temperatures from the NOAA dataset, and I found that, while seasonality was apparent, there has still been an increase in temperatures from a century ago. I added reference lines on the average temperatures in 1936 and in 2021 to better understand what these differences were, and I found that while 1936 average minimum and maximum temperatures were 42 °F and 58 °F respectively, and in 2021 these averages were 5 and 4 degrees hotter at 47 °F and 62 °F.

Having seen this increase in temperatures, I wanted to follow up by seeing how ENSO phenomena might have impacted the weather. Specifically, I was interested in seeing if there was any pattern in the frequency of ENSO events in recent years and the frequency and volume of snowfall in Boston. I started by showing the counts of 3-month intervals for El Niño and La Niña events side by side and from decade to decade. I placed the counter values for both phenomena from the 3-month ENSO database in my worksheet’s rows, and in its columns, I placed a calculated field categorizing decades by a date’s year. After overlapping these figures, I had a side-by-side bar chart that visualized both ENSO phenomena moving with time in an oscillating pattern, with some higher counts of El Niño months in the 1980s and 1990s.

I then made two separate bar charts to observe how these phenomena could have influenced snowfall. For both graphs, I placed the “Year” variable in the columns so that I could see movement from year to year. To look at influences on snowfall volume I set my rows to be the sum of snowfall, while to observe influences on snowfall frequency I changed the rows to be the sum of true values for my NOAA dataset’s snow Boolean column. Both charts were color coded according to whether or not the year itself was an El Niño, La Niña, or neutral year. This color coding was done using another calculated field that categorized years by analyzing if a phenomenon was labeled as True in the ENSO Boolean dataset. This was to intended to show how either snowfall or snow frequency might have been different in either ENSO category, but ultimately, I found that there seemed to be no pattern relating to them. There were plenty of years that had high, low, or normal snowfall in either category, and while snowfall frequency seemed to be higher and not lower during El Nino years, there were enough years where this wasn’t the case that it seemed to be a negligible relationship.   

I wanted to look more closely at how snowfall may be changing from season to season, so I plotted the sum of snowfall against winter seasons. I found here that snowfall volume rises and falls, and that the past few years haven’t looked especially different compared to others from previous decades. I decided to place a reference line at our annual average of 44 inches, but even when comparing snowfall with this value, there didn’t seem to be any significant variation away from this value over time. I added color variation in response with average temperatures during this time, and did find that warmer seasons had less snowfall than colder seasons, but overall, the numbers don’t suggest that snowing itself has changed in any way.

As I started looking into details about yearly snowfall, I shifted my focus towards “feels like” temperatures, as they would help account for both the low temperatures and humidity needed to snow. I also started to consider median values over averages, since I wanted a more consistent read on what winter temperatures were looking like, and I didn’t want these to be influenced by any potential outliers. I started by creating three line charts of the minimum, average, and maximum of median winter temperatures in my focus years, using the “Year” variable in my columns and these median temperature measurements in my rows to create three graphs in one. All three of these measurements showed steady increases over time. For added granularity, I took these same measurements by decade and used Tableau’s pages feature to scroll through values from month to month. Here, new patterns started to present themselves: from 2010 onwards, the monthly median feels-like temperatures appear higher than previous decades. For some months, these values weren’t too out of the ordinary and were even seen previously in the 70s. Other months, like March, were significantly warmer than before. Shockingly, the minimum temperatures rose the most, indicating that our low temperatures are not dipping to the levels they used to, and that months that might have had more variability are starting to show an abnormal shrinking of the gap between the highest and lowest temperatures. What’s more, March in particular had the most startling temperatures over recent years, as its temperatures stay closer to or above freezing, instead of varying over a wider range. This month typically gets its fair share of snow, but with warmer weather, that snow might be melting as it falls and not sticking to the ground

Next in my analysis was to consider these “feels like” temperatures as they related to snowfall. To watch how snowfall might have changed over time, I made scatter-plot of each day in the winter months using their minimum “feels like” temperature on the x-axis and the maximum on the y-axis. The plot points were colored using my snow Boolean from NOAA and were sized by the volume of snowfall, where the smallest point represented 2 inches of snow or less, and the largest represented just under 24 inches. I also added a reference line on each axis at the freezing point, so I could better track any possible movements in the data. I found that days where it did snow always had a minimum temperature at or to the left of the freezing reference line even when the maximum temperature might have varied more. Days that had the largest amount of snow showed both temperature readings below freezing, meaning that these days stayed very cold and allowed for bigger snowstorms. Over the years, more data points appeared to shift to warmer temperature values and clustered around each other more, implying less variability, just as the closing temperature gaps from the median temperature bar charts implied. Furthermore, there seems to be a decrease in the number of snow days with medium or larger volumes of snow, as well as in increase in snow day temperatures overall.

I duplicated this graph and filtered out all months but November to find any patterns in November snowfall to clarify my hypothesis on that month’s changes in snowfall. Based on this plot, November usually would see snowfall one to three times. Seasons without November snowfall would appear once in a while, but this skip would only last for a year. After the late 90s, these skips start becoming much more frequent and steadily increase to two and three years at a time. In this same time period, when it did snow in November, it decreased from a small handful of snows to just one. Between 2015 and 2021, it seems that there was only one November snow – in 2018.

Finally, to see if the start or end of the snowfall season has moved, I made a box plot for each, listing the months of a year in the graph rows and the date of first or last snowfalls in the columns. These days were extracted by creating a calculated field that returned the date of all dates where the first-snow or last-snow Boolean columns were listed as True. I then filtered out months that had no first or last snowfall dates to make them visually easier to interpret. The first snow plot shows almost all years since 1936 having their first snow fall into November, with the occasional start in October every few years. Only twice does it dip as far late as December until the late 90s, when the start of the season begins to vary. Recent years are starting to regularly see snowfall start in December, and from 2013 this has been the case every other year. On the other end of the season, it still seems that snowfall stops in April, though within the past decade there seem to be more frequent, earlier ends to the season in March. This would imply that snow-friendly temperatures are beginning to lift earlier in the year – contracting our winter season on both ends of the spectrum, though with a slower shift on the spring-half of the season.