Arbogast Analytics

US Drought Monitor

https://droughtmonitor.unl.edu/DmData/DataDownload/ComprehensiveStatistics.aspx

The U.S. Drought Monitor is produced through a partnership between the National Drought Mitigation Center at the University of Nebraska-Lincoln, the United States Department of Agriculture and the National Oceanic and Atmospheric Administration.

This data has either the total number of acres affected by the drought category per state on that date, or the percentage of the states area that is affected by that drought category.

Drought Severity and Coverage Index – DSCI

Not Dry/Drought: None

Dryness Category: D0-Abnormally Dry—used for areas showing dryness but not yet in drought, or for areas recovering from drought.

Drought Category Intensity Level: D1-Moderate Drought, D2-Severe Drought, D3-Extreme Drought, D4-Exceptional Drought

Here we can see a strong correlation between ‘Acres Burned’ and ‘Acres Burned per Fire’ (somewhat expected). This relationship is to be expected as an acre burned by a single fire is a subset of the total acres burned. This is observation that will be necessary to consider when developing a model in the future. Another quick visual observation that this pair plot provides is the general shape of the different distributions. It may be important to note that these are aggregate over all the months and years, so they may be misleading. Further distribution plots below will give a deeper understanding into the distributions of these three quantitative values. That said, there is a clear positive skew for the variables ‘Acres Burned’ and ‘Acres Burned per Fire’. This may be caused by a few years having outliers of extremely high area of burned land. ‘Number of Fires’ is slightly less skewed, meaning that there is more consistency in the number of fires spread. A quick glance at the boxplots reveals that there exists months that act differently in their values of the three variables.

Continuing the exploratory data analysis, now it is time to look a little closer into the distributions of the variables on a deeper level.

As we guessed from the aggregated density graphs the ‘Acres Burned’ and ‘Acres Burned per Fire’ are distributed with a positive skew (this raises a red flag about using mean as the only measure of center). Also, the ‘Number of Fires’ densities have a larger spread. The “bumpiness” of the ‘Number of Fires’ graphs might be the result of certain months having higher trends. For example in the density of ‘Number of Fires’ for the year 2000 we see two distinct peaks. This might be the result of two common levels of fires that year (lots and small). What is also worth noting in the three plots above is that there a significant peaks in the far right values. This hints to there being some months that have a distinctly higher amount of fires than the other months. It may be interesting to look into these months are try and see the cause of their extreme high values.

Another look into the densities over time is for the two acres burned plots. In more recent years of spread of the densities have increased. In the years 2000 to 2005 there was a consistent trend of a distinct peak in the low end of acres burned. However, in the year 2011, 2016, 2017, 2018, 2020, and 2021 it can be seen to have more spread in these densities. This is going to show that these higher values of aces burned are becoming less “unique”. Where in 2004 the far right peak stands alone (outliers still to be determined), in 2018 this similar value feels closer to that average acres burned.

Discussing the changes over time in the graphs above we can see that for ‘Number of Fires’ the spread of the densities shrink and shift to a lower value. This demonstrates that the number of fires are more consistent across the different months in later years to that of the early 2000s.

The above boxplot shows what we were assuming that the spread of the number of fires does indeed “shrink” over time. It would be interesting to perform further research into the reason why, the spread has decreased over time, and if it is human intervention or environmental reasons.

To look closer at the trends of ‘Acres Burned’ and ‘Number of Fires’ refer to the scatter plots with regression lines below.

The above plots demonstrate more into the trend of acres burned and number of fires in there total values as apposed to the spread as demonstrate in the density ridge plots. How these scatter plots were made was that each year the variables values were summed up over the months to show what the total number of the variable (acres burned or number of fires respectively) for that year. As one can see, there is a noticeable increase in ‘Acres Burned’ from 2000 to 2022 but a decrease in the ‘Number of Fires’. This leads to an observation that even though the number of fires have decreased over the past two decades, the severity of each of these fires has increased. So much so that the smaller number of fires is producing more burned area than the larger number of fires in total. This can be quite alarming as it reflects on the “out of control” nature that fires can have. Further understanding might help with protecting people and the environment in the US from these wildfires that are burning more and more than previously.

In the above bar chart, one can see that not only have the number of acres burned increased throughout the years, but there has also been an increase in acres burned from specific causes. Fires are started in many ways, lightning is one of the more famous “natural” ways. Since lightning is an unavoidable weather pattern, it is important to be aware of these kinds of trends. Further analysis should be performed to see if the increase in acres burned has been the result of more lightning or other factors such as poor land management and drought.

To continue the exploration into the wildfire dataset, summary statistics showed the maximum and minimum years/months for the three variables of interest. Looking a little more closely at these extremes may help us gain a better understanding of the pattern around wildfires.

Maximums and Minimums – Looking for Trends in Months

The above density plots represent that distributions of the three variables (‘Acres Burned’, ‘Acres Burned per Fire’, ‘Number of Fires’) for each of the months. As one can see in the acres burned ridge plots there are distinct months in the year that have more acres burned than others. These have been highlighted with shades of red (darkest meaning the highest average acres burned month for the time span recorded). As one might note, in the northern hemisphere where the United States of America is geographically located has Summer between June 21st and September 23rd. Since there is such a noticeable change in the distribution of acres burned during these months, an analysis into weather and water content may be extremely useful in building a stronger knowledge on the situation at hand.

Another noteworthy observation is the number of fires is a little more consistent between months, but still has a significant increase in its center and spread in the months highlighted in red. For the number of fires, the darkest blue and the darkest red demonstrate the extreme ends of the number of fires. So January has a history of having less total fires throughout the past two decades and July has the most. It is interesting to note that that January and February have the least amount of fires and the it jumps up in March and the starts to slow down in October. Also within the distributions of the number of fires one can see that there is a lot of variation in December. This is interesting because for the acres burned variables the colder months have fairly normal distributions and gain more and more of a skew as the months change to summer. Yet, for fall into winter months, one can still see lots of variation in the number of fires. Further research can give a better understanding of why this may be. This same observation can be seen in the boxplot below.

Here the boxplot develops a clearer story of how the number of fires quickly increases in spring, settles at a higher level, and then drops off as it moves towards winter. Then shocking a lot more fires in December. One curiosity to look into is Christmas fires. Instead of it being a climate drought making trees easier to burn, Christmas trees that have dried out could be adding to the increase in December fires.

Maximums and Minimums – Years

Darkest Red to Darkest Blue – Highest 5 Years for Acres Burned in red and Lowest 5 Years for Acres Burned in blue. The purpose of this plot is showing which how these two extremes differ in there distributions over the span of the year. 2020 holds the highest amount of acres burned between the year 2000 and 2022 in the US, one can see that a significant amount of this space happened in the months July through September. It is also important to note that September 2020 was significantly larger than the next highest burn year’s of 2015, but that the July 2015 had significantly more acres burned than July 2020. It is even worth mentioning that three of the bottom 5 have higher acres burned in July compared to the year 2020’s July.

It is also important to note that for months with smaller acres burned there is less of a difference between the top 5 and bottom 5, but in months with more burn the differences start to be more noticeable, especially in March, July, and September.

Here one can see that the 5 highest number of fires and 5 lowest are not the same years as for the acres burned. As noted above 2020 had the most acres burned in the last two decades, but here we see that 2020 is one of the lowest – the fourth lowest. Which means that the fires in 2020 may have be low in the number of fires, but as seen in the first plot months such as August and September are severe per fire. Another point to observe is that for the number of fires, it stays fairly consistent in the 5 lowest being noticeably lower than the the five highest (except for January and November). This plot stays consistent with the trends seen in the boxplot “Number of Fires Each Month for the year 2000 to 2022” as described above, where the number of fires increase in March and then stays somewhat steady in summer then decreases in the fall and winter.