The Earth is warming. Period. Humans are responsible for most of the warming in the last 200 years. Period. Let's get that out of the way. These are uncontroversial statements in the climate science community. The people who doubt these facts are appropriately know as climate deniers.
What I am interested in knowing is how daily temperature record have changed over time. It's common to hear the phrase, "a new record high temperature was set today." Less frequently you will hear, "a new record low temperature was set today." What I am interested in knowing is how often this occurs and are they occurring more frequently or less frequently with time.
The first hurdle to get past is something called the period-of-record (PoR). This refers to how long records have been kept at a location. The longer the PoR, the harder it is to set records. Imagine a station with only 4 years of daily temperature data. In year 5, there is a 20% chance of setting a record on any given day based strictly on probabilities. If a station had 100 years of data, that chance drops to 1% in any given year. How do we reconcile this? The strategy I employed was to fix the period of record. In this case, I used the 1943-2022 period (80 years) and assume that Jan 1, 1943, is the first day that records were kept. In this manner, every station has exactly 80 years of records - no more, no less.
Methodology
For each day of the year, Jan 1, Jan 2, and so on, I identified the year that the record high/low temp was set. That year gets a tally of 1 for each daily record that occurred during that year. For example, if the record high for Mar 1 was set in 1980, the record high for Apr 1 was set in 1980, and the record high for May 1 was set in 1980, and no other days in 1980 are identified as having any record highs, the year 1980 gets a tally of 3. If in 1981, the record set on May 1 in 1980 was tied, the May 1 record is assigned 50% to 1980 and 50% to 1981. This means the tally of 3 for 1980 is now reduced to 2.5. In this way, every day of the year counts toward a record value of 1. If the record is only in one year, that year gets the entire value. If the record is shared by two , three , or more years, each of those years gets a fractional share of that record. For an 80-year period, each year should have 4.56 records on average. Of course some will have more and some will have less. This study looks at the trend in records.
The Global Historical Climatology Database - Daily (GHCN-D) is used to assess daily temperature values. For inclusion, a station must be 95% complete for the 1943-2022 period and each year must have at least 250 valid temperature observations. At this point, readers are probably asking why 1943 was chosen as the start year. There are two primary reasons. First, the longer time period you look at, the fewer stations there are with sufficient data. Going back 90, 100, or more years greatly reduces the number of station that would qualify. Second, I very intentionally wanted to avoid the Dust Bowl years since there is a large contribution from human activities to the temperatures in large parts of the Lower 48.
WBAN and Cooperative Stations
The analysis includes both WBAN (Major or 1st Order) stations and Cooperative (Coop) stations. WBAN stations are typically found in big cities and always restrict the time period of high/low temps to midnight-to-midnight. Coop stations record the daily min/max temps in the morning (frequently 7 am) or sometimes in the evening (frequently 7 pm). An individual station is exclusively a morning station or an evening station. Major cities have a well-know propensity to be warmer than their surrounding Coop counterparts. This effect, the urban heat island effect, is real and because of it, we separate the analysis into two sets - WBAN and Coop.
Trends
The time series in Fig. 1 shows the number of daily high maximum temperature records assigned to each year for Nome, Alaska. A linear regression trend line is shown on the chart as a dotted line. As you can clearly see, record high temperatures are much more frequent in recent years and much less frequent in earlier years. The trend in daily records for each station is assessed using a Mann-Kendall (MK) test for significance. This test is well suited to data sets with large internal variability. The Mann-Kendall test for this time series indicates that it is increasing (duh!) at greater than 99th percent confidence interval. That means there's less than a 1% chance the distribution has the shape it does purely by chance.
Types of Records
Four types of temperature records are evaluated in this analysis: high maximum, high minimum, low maximum, and low minimum. The first two are warm records and the latter two are cold records. The maps for the records show the statistical significance. The categories are:
- Increasing/Decreasing at a rate that is statistically significant at the 90th confidence interval.
- Increasing/Decreasing at a rate that is not statistically significant (No Trend).
The Maps
The maps are organized by a) all stations, b) major stations, and c) Coop stations. Within each of those groupings, there are maps for high 1) high maximum, 2) high minimum, 3) low maximum, and 4) low minimum.
Maps for All Stations (WBAN & Coop)
High Maximum - All Stations
High Minimum - All Stations
Low Maximum - All Stations
Low Minimum- All Stations
Maps for Major (WBAN) Stations
High Maximum - Major Stations
Maps for Cooperative (Coop) Stations
High Maximum - Coop Stations
Supplemental Plots
