Climate Change Indicators in the United States Technical Documentation Seasonal Temperature

Seasonal Temperature

Identification

1. Indicator Description

This indicator describes changes in average seasonal air temperature for the United States from 1896 to
2020. Temperature data for each season (winter, spring, summer, fall) are presented as trends in
anomalies. Air temperature is an important indicator of climate, and changes in temperature can have
wide-ranging direct and indirect effects on the environment and society. Evidence of a warming climate
often begins with average surface temperature (USGCRP, 2017).

Components of this indicator include:

• Changes in average air temperature by season in the contiguous 48 states over time (Figure 1).

• Maps showing the total change in average air temperature by state and by season, across the
contiguous 48 states (Figure 2).

• A comparison of long-term trends in the contiguous 48 states for all four seasons (Figure 3).

2. Revision History

April 2021: Indicator published.

Data Sources

3. Data Sources

This indicator is based on temperature anomaly data provided by the National Oceanic and Atmospheric
Administration's (NOAA's) National Centers for Environmental Information (NCEI), formerly the National
Climatic Data Center (NCDC). Specifically, all three figures in this indicator use data from NCEI's nClimDiv
data set, which is based on data from the daily version of the Global Historical Climatology Network
(GHCN-Daily). GHCN-Daily data undergo more extensive processing by NCEI on a monthly basis for
inclusion in nClimDiv.

4. Data Availability

The nClimDiv data used in each figure can be accessed online, along with descriptions and metadata.
Surface temperature time series data for the contiguous 48 states (Figure 1 and Figure 3) and by state
(Figure 2) are based on nClimDiv data that were obtained from NCEI's "Climate at a Glance" web
interface (www.ncdc.noaa.gov/cag). For access to underlying nClimDiv data and documentation, see:
www.ncdc.noaa.gov/monitoring-references/maps/us-climate-divisions.php.

EPA downloaded data in early February 2021, at which point data for all four seasons were available
through 2020 (where winter 2020 is the season running from December 2019 through February 2020).

Technical Documentation: Seasonal Temperature

-------
Methodology

5. Data Collection

This indicator is based on surface temperature measurements that were compiled from thousands of
weather stations throughout the United States using standard meteorological instruments. All of the
stations are overseen by NOAA, and their methods of site selection and quality control have been
extensively peer reviewed. Data from these stations for the contiguous 48 states were compiled in the
nClimDiv data set, which represents the most complete long-term instrumental data set for analyzing
recent climate trends in the contiguous 48 states.

The nClimDiv divisional data set incorporates temperature data from GHCN-Daily stations in the
contiguous 48 states and Alaska. This data set includes stations that were previously part of the U.S.
Historical Climatology Network (USHCN), as well as additional stations that were able to be added to
nClimDiv as a result of quality-control adjustments and digitization of paper records. Altogether,
nClimDiv incorporates data from more than 10,000 stations. These stations are spread among 357
climate divisions in the contiguous 48 states and Alaska.

In addition to incorporating more stations, the nClimDiv data set differs from the USHCN because it
incorporates a grid-based computational approach known as climatologically aided interpolation
(Willmott and Robeson, 1995), which helps to address topographic variability. Data from individual
stations are combined in a grid that covers the entire contiguous 48 states and Alaska with 5-kilometer
resolution. These improvements have led to a new data set that maintains the strengths of its
predecessor data sets while providing more robust estimates of area averages and long-term trends.
The nClimDiv data set is NOAA's official temperature data set for the contiguous 48 states and Alaska,
replacing USHCN.

The nClimDiv data set starts in 1895. This indicator starts in 1896 for consistency, due to the way winter
is defined (see more details in Section 6). To learn more about nClimDiv, see:
www.ncdc.noaa.gov/news/ncdc-introduces-national-temperature-index-page and:
www.ncdc.noaa.gov/monitoring-references/maps/us-climate-divisions.php.

6. Indicator Derivation

General Data Processing Steps

NOAA calculated monthly temperature means for each station represented in the nClimDiv data set. In
populating the data set, NOAA adjusted the data to remove biases introduced by differences in the time
of observation. NOAA also employed a homogenization algorithm to identify and correct for substantial
shifts in local-scale data that might reflect changes in instrumentation, station moves, or urbanization
effects. These adjustments were performed according to published, peer-reviewed methods. For more
information on these quality assurance and error correction procedures, see Section 7.

To achieve uniform spatial coverage (i.e., not biased toward areas with a higher concentration of
measuring stations), NOAA calculated area-weighted averages of grid-point estimates interpolated from
station data. The surface time series for the contiguous 48 states (Figure 1) is based on the nClimDiv
gridded data set, which reflects a high-resolution (5-kilometer) interpolated grid that accounts for
station density and topography. Results were initially averaged within each climate division, then

Technical Documentation: Seasonal Temperature

-------
aggregated by state and for the contiguous 48 states as a whole based on an area-weighted average of
climate divisions. See: wwwl.ncdc.noaa.gov/pub/data/cirs/climdiv/divisional-readme.txt for more
information.

In this indicator, the four seasons correspond with the meteorological seasons. Winter corresponds to
the months December-February, spring is March-May, summer is June-August, and fall is September-
November. NOAA averaged monthly means across each three-month period to derive seasonal
temperature data points for each year. Winter data are nominally assigned to the year in which the
winter ended; for example, "1896" refers to the consecutive three-month period of December 1895,
January 1896, and February 1896. Because the nClimDiv data set does not include data for December
1894, the seasonal temperature time series in this indicator start in the year 1896 to allow for
consistency across all four seasons.

Figure 1. Average Seasonal Temperatures in the Contiguous 48 States, 1896-2020

Figure 1 presents temperature data across the contiguous 48 states in terms of anomalies for each
season. An anomaly represents the difference between an observed value and the corresponding value
from a baseline period. This indicator uses a baseline period of 1901 to 2000. The choice of a sufficiently
long-term average or baseline period (e.g., at least 30 years) will not substantially affect the shape or the
statistical significance of the overall trend in anomalies. For absolute anomalies in degrees, it only moves
the trend up or down on the graph in relation to the point defined as "zero."

Figure 2. Change in Seasonal Temperatures by State, 1896-2020

The maps in Figure 2 show the long-term change in temperature by state for each season for the period
from 1896 to 2020. The slope of each temperature trend was calculated from the three-month climate
division anomalies by ordinary least-squares regression and then multiplied by the total number of years
(124) to estimate the total change over the period of record.

Figure 3. Temperature Change by Season in the Contiguous 48 States, 1896-2020

Figure 3 presents long-term temperature changes in all four seasons. The graph shows the total change
for each season, derived from the ordinary least-squares regression slope (°F per year) multiplied by 124
years.

Indicator Development

NOAA continually refines historical data points in the GHCN and nClimDiv, often as a result of improved
methods to reduce bias and exclude erroneous measurements. As EPA updates this indicator to reflect
these upgrades, slight changes to some historical data points may become apparent. No attempt has
been made to portray data beyond the time and space in which measurements were made.

7. Quality Assurance and Quality Control

NCEI's databases have undergone extensive quality assurance procedures to identify errors and biases in
the data and either remove these stations from the time series or apply correction factors. The nClimDiv
data set follows the USHCN's methods to detect and correct station biases brought on by changes to the
station network over time. The transition to a grid-based calculation did not significantly change

Technical Documentation: Seasonal Temperature

-------
national averages and totals, but it has led to improved historical temperature values in certain regions,
particularly regions with extensive topography above the average station elevation—topography that is
now being more thoroughly accounted for. An assessment of the major impacts of the transition to
nClimDiv can be found at: www.ncdc.noaa.gov/monitoring-references/docs/GrDD-Transition.pdf.

Analysis

8. Comparability Over Time and Space

Both nClimDiv and the GHCN have undergone extensive testing to identify errors and biases in the data
and either remove these stations from the time series or apply scientifically appropriate correction
factors to improve the utility of the data. In particular, these corrections address changes in the time-of-
day of observation, advances in instrumentation, and station location changes. See Section 7 for
documentation.

All GHCN-Daily stations are routinely processed through a suite of logical, serial, and spatial quality
assurance reviews to identify erroneous observations. For nClimDiv, all such observations were set to
"missing" before computing monthly values, which in turn were subjected to additional serial and
spatial checks to eliminate residual outliers. Stations having at least 10 years of valid monthly data since
1950 were used in nClimDiv.

For temperature, bias adjustments were computed to account for historical changes in observation
time, station location, temperature instrumentation, and siting conditions. As with USHCN, the method
of Karl et al. (1986) was applied to remove the observation time bias from the COOP network, and the
pairwise method of Menne and Williams (2009) was used to address changes in station location and
instrumentation in all networks. Because the pairwise method also largely accounts for local,
unrepresentative trends that arise from changes in siting conditions, nClimDiv contains no separate
adjustment in that regard.

For more documentation about nClimDiv, see: wwwl.ncdc.noaa.gov/pub/data/cirs/climdiv/divisional-
readme.txt.

9. Data Limitations

Factors that may impact the confidence, application, or conclusions drawn from this indicator are as
follows:

1. Biases in surface measurements may have occurred as a result of changes over time in
instrumentation, measuring procedures (e.g., time of day), and the exposure and location of the
instruments. Where possible, data have been adjusted to account for changes in these variables.
For more information on these corrections, see Section 8. Some scientists believe that the
empirical debiasing models used to adjust the data might themselves introduce non-climatic
biases (e.g., Pielke et al., 2007).

2. Uncertainties in surface temperature data increase as one goes back in time, as there are fewer
stations early in the record. These uncertainties are not sufficient, however, to mislead the user
about fundamental trends in the data (Vose and Menne, 2004).

Technical Documentation: Seasonal Temperature

-------
3. Alaska, Hawaii, and U.S. territories are not included in this indicator. Alaska data are included in
the nClimDiv data set, but sufficient data are not available before 1925. Hawaii and U.S.
territories also have limited data, and nClimDiv spatial interpolation methods have not been
developed and validated for these regions.

10. Sources of Uncertainty

Error estimates are not readily available for U.S. temperature. Uncertainties in temperature data
increase as one goes back in time, as there are fewer stations early in the record. These uncertainties
are not sufficient, however, to undermine the fundamental trends in the data. Vose and Menne (2004)
suggest that the station density in the U.S. climate network is sufficient to produce a robust spatial
average.

11. Sources of Variability

Seasonal temperature anomalies naturally vary from location to location and from year to year as a
result of normal variations in weather patterns, multi-year climate cycles such as the El Nino-Southern
Oscillation and Pacific Decadal Oscillation, and other factors. This indicator minimizes the effect of these
factors by presenting a long-term record (more than a century of data) and averaging consistently over
time and space.

12. Statistical/Trend Analysis

This indicator uses ordinary least-squares regression to calculate the slope of the observed trends in
temperature for each season. This slope is the basis for presenting total long-term change in Figures 2
and 3. Table TD-1 shows the resulting regression coefficients and the corresponding p-values (via simple
t-test) for the contiguous 48 states, which are significant at the 95 percent confidence level (p < 0.05) in
all four seasons. Table TD-1 also summarizes the results of state-level significance testing. All significant
state trends showed increasing temperatures.

Table TD-1. Linear Regression of Seasonal Average Temperature Data, 1896-2020

Season

Contiguous 48 states

State-specific trends

Slope (°F/year)

P-value

Number of states
with significant
change (p < 0.05)

Number of states
with no significant
change

Winter

0.023

<0.0001

Spring

0.015

<0.0001

Summer

0.011

<0.0001

Fall

0.011

<0.0001

Use of linear regression and a simple t-test inevitably requires some assumptions about the shape and
underlying characteristics of the data. These tests are used to provide initial approximations about the
statistical attributes of this indicator.

Technical Documentation: Seasonal Temperature

-------
References

Karl, T.R., C.N. Williams Jr., P.J. Young, and W.M. Wendland. 1986. A model to estimate the time of
observation bias associated with monthly mean maximum, minimum, and mean temperature for the
United States. J. Climate Appl. Meteor. 25:145-160.

Menne, M.J., and C.N. Williams, Jr. 2009. Homogenization of temperature series via pairwise
comparisons. J. Climate 22(7):1700-1717.

Pielke, R., J. Nielsen-Gammon, C. Davey, J. Angel, O. Bliss, N. Doesken, M. Cai, S. Fall, D. Niyogi, K. Gallo,
R. Hale, K.G. Hubbard, X. Lin, H. Li, and S. Raman. 2007. Documentation of uncertainties and biases
associated with surface temperature measurement sites for climate change assessment. B. Am.
Meteorol. Soc. 88:913-928.

USGCRP (U.S. Global Change Research Program). 2017. Climate science special report: Fourth National
Climate Assessment, volume I. Wuebbles, D.J., D.W. Fahey, K.A. Hibbard, D.J. Dokken, B.C. Stewart, and
T.K. Maycock (eds.). https://science2017.globalchange.gov. doi:10.7930/J0J964J6

Vose, R.S., and M.J. Menne. 2004. A method to determine station density requirements for climate
observing networks. J. Climate 17(15):2961-2971.

Willmott, C.J., and S.M. Robeson. 1995. Climatologically aided interpolation (CAI) of terrestrial air
temperature. Int. J. Climatol. 15(2):221-229.

Technical Documentation: Seasonal Temperature

6

-------