Everyone needs confidence! But confidence is especially needed when it comes to data analysis and climate research. Weather is complex and is best summed up by the famous quote: “Climate is what you expect, weather is what you get.”
Participants at last week’s Think Tank meeting also came to this conclusion. While it is more effort, a degree of confidence must be assigned to all the extreme climate trends we are analyzing. In other words, if we see a trend (or lack of a trend) in our regional data, how representative is that trend to a specific location in Maryland or Virginia?
In most of my previous posts, I have been using the gridded data set, HadEX2, to display the regional climate trends for the near-shore Chesapeake. However, we also have 18 individual weather stations (Figure 1) scattered throughout Maryland and Virginia AND another gridded data product called GHCNdex. GHNCdex has a slightly smaller grid size (2.5° X 2.5°) but incorporates less data than the HadEX2 (2.5° by 3.75°). It is also a shorter regional time series with a date range from 1951-2014. Using R, we can calculate the same extreme climate indices for each of these individual weather stations, allowing us to have 20 different trends per climate index.
Confused? Let me demonstrate how we will be assessing confidence in the near-shore Chesapeake Bay region using the extreme climate index: Summer Days. Summer Days is the annual count of days when the daily maximum temperature is greater than 25°C, or 77°F. While this temperature may not seem particularly “hot,” any changes in the quantity of days which reach that temperature could have definitive disturbances to the ecosystem.
Using the Hadex2 gridded product (Figure 2), we can say that the trend follows a cyclical pattern, thus there is no linear increase or decrease in the number of Summer Days observed. But do we observe any trends in the GHCNdex gridded product? Looking at Figure 3, we can say no. The GHCNdex gridded product also shows a cyclical trend, but has the suggestion of a sharper increase in the annual count of Summer Days towards the end of the time series.
Now how well do the individual weather stations agree with the HadEX2? Are Summer Days increasing at any locations? Decreasing? Or do they all agree with the regional-scale trend?
How about all of the above! Table 1 is a perfect example at how variable and spatially complex climate trends can be. Factors such as land-type, latitude, and location can influence local weather. For example, the ocean and large structures like mountains or even a dense city can alter certain temperature and precipitation events.
What we observe is that quite a few local weather stations were analyzed to reveal a general increase in the number of Summer Days. More specifically, 78% of the Maryland weather stations and 56% of the Virginia stations suggested an increase in amount of days above 77°F. This is, of course, controlled by the number of weather stations, but these 18 best fit our quality control selection.
Also not to go unnoticed, one station in Virginia, located in Lincoln, saw a decrease in the amount of Summer Days. In total, of the 20 data sets investigated, 60% saw an increase in the Summer Day index. By IPCC confidence terminology, this trend would be classified as “about as likely as not.”
So, for the Summer Days index, we can suggest here that increases in days > 77°F is a localized trend observed in some places and not in others. We also must be mindful of the length of each time series. GHCNdex had a cyclical trend with a period of increase starting in the 1970’s. This is why we looked at weather stations that began data collection no later than 1956. One could image if we analyzed the GHCNdex trend from 1970-2000, we would falsely see a sharp increase!
In conclusion, looking at only the gridded products would never have uncovered intriguing trends such as the decrease in Lincoln, VA, the lack of trend at Hopewell, VA, or the increase in Laurel, MD.