Length of Growing Season

Identification

1. Indicator Description

This indicator measures the length of the growing season (or frost-free season) in the contiguous 48
states between 1895 and 2015. The growing season often determines which crops can be grown in an
area, as some crops require long growing seasons, while others mature rapidly. Growing season length is
limited by many different factors. Depending on the region and the climate, the growing season is
influenced by air temperatures, frost days, rainfall, or daylight hours. Air temperatures, frost days, and
rainfall are all associated with climate, so these drivers of the growing season could change as a result of
climate change.

This indicator focuses on the length of the growing season as defined by frost-free days. Components of
this indicator include:

•	Length of growing season in the contiguous 48 states, nationally (Figure 1), for the eastern and
western halves of the country (Figure 2), and by state (Figure 3).

•	Timing of the last spring frost and the first fall frost in the contiguous 48 states, both nationally
(Figure 4) and by state (Figures 5 and 6).

2. Revision History

April 2010:
December 2012:
August 2013:
May 2014:

June 2015:

August 2016:

Indicator published.

Updated indicator with data through 2011.

Updated indicator on EPA's website with data through 2012.

Updated indicator with data through 2013.

Updated indicator on EPA's website with data through 2014.

Added Figures 3, 5, and 6.

Updated indicator with data through 2015.

Data Sources

3.	Data Sources

Data were provided by Dr. Kenneth Kunkel of the National Oceanic and Atmospheric Administration's
(NOAA's) Cooperative Institute for Climate and Satellites (CICS), who analyzed minimum daily
temperature records from weather stations throughout the contiguous 48 states. Temperature
measurements come from weather stations in NOAA's Cooperative Observer Program (COOP).

4.	Data Availability

EPA obtained the data for this indicator from Dr. Kenneth Kunkel at NOAA CICS. Dr. Kunkel had
published an earlier version of this analysis in the peer-reviewed literature (Kunkel et al., 2004), and he

Technical Documentation: Length of Growing Season	1


-------
provided EPA with an updated file containing growing season data through 2015, aggregated by state,
two (east and west) regions, and nationwide.

All raw COOP data are maintained by NOAA's National Centers for Environmental Information (NCEI)
(formerly the National Climatic Data Center). Complete COOP data, embedded definitions, and data
descriptions can be downloaded from the web at: www.ncdc.noaa.gov/data-access/land-based-station-
data/land-based-datasets/cooperative-observer-network-coop. There are no confidentiality issues that
could limit accessibility, but some portions of the data set might need to be formally requested.
Complete metadata for the COOP data set can be found at: www.ncdc.noaa.gov/data-access/land-
based-station-data/station-metadata and www.nws.noaa.gov/om/coop.

Methodology

5.	Data Collection

This indicator focuses on the timing of frosts, specifically the last frost in spring and the first frost in fall.
It was developed by analyzing minimum daily temperature records from COOP weather stations
throughout the contiguous 48 states.

COOP stations generally measure temperature at least hourly, and they record the minimum
temperature for each 24-hour time span. Cooperative observers include state universities, state and
federal agencies, and private individuals whose stations are managed and maintained by NOAA's
National Weather Service (NWS). Observers are trained to collect data, and the NWS provides and
maintains standard equipment to gather these data. The COOP data set represents the core climate
network of the United States (Kunkel et al., 2005). Data collected by COOP sites are referred to as U.S.
Daily Surface Data or Summary of the Day data.

The study on which this indicator is based includes data from 750 stations in the contiguous 48 states.
These stations were selected because they met criteria for data availability; each station had to have
less than 10 percent of temperature data missing over the period from 1895 to 2015. For a map of these
station locations, see Kunkel et al. (2004). Pre-1948 COOP data were previously only available in hard
copy, but were recently digitized by NCEI, thus allowing analysis of more than 100 years of weather and
climate data. The data availability criteria resulted in there being no stations of record within the state
of Delaware.

Temperature monitoring procedures are described in the full metadata for the COOP data set available
at: www.nws.noaa.gov/om/coop. General information on COOP weather data can be found at:
www.nws.noaa.gov/os/coop/what-is-coop.html.

6.	Indicator Derivation

For this indicator, the length of the growing season is defined as the period of time between the last
frost of spring and the first frost of fall, when the air temperature drops below the freezing point of
32°F. Minimum daily temperature data from the COOP data set were used to determine the dates of last
spring frost and first fall frost using an inclusive threshold of 32°F. Methods for producing state, regional,
and national values by year were designed to weight all areas evenly regardless of station density.

Technical Documentation: Length of Growing Season

2


-------
Kunkel et al. (2004) provide a complete description of the analytical procedures used to determine
changes in length of growing season. No attempt has been made to represent data outside the
contiguous 48 states or to estimate trends before or after the 1895-2015 time period.

Figures 1, 2, and 4. Length of Growing Season, Timing of Last Spring Frost, and Timing of First Fall Frost
in the Contiguous 48 States, 1895-2015

Figure 1 provides a time series showing the length of the growing season, which is the number of days
between the last spring frost and the first fall frost. Figure 2 shows a time series of the length of growing
season for the eastern United States versus the western United States, using 100°W longitude as the
dividing line between the two halves of the country. Figure 4 shows the timing of the last spring frost
and the first fall frost, also using units of days.

Figures 1, 2, and 4 show the deviation from the 1895-2015 long-term average, which is set at zero for a
reference baseline. Thus, if spring frost timing in year n is shown as -4, it means the last spring frost
arrived four days earlier than usual. Note that the choice of baseline period will not affect the shape or
the statistical significance of the overall trend; it merely moves the trend up or down on the graph in
relation to the point defined as "zero."

To smooth out some of the year-to-year variability and make the results easier to understand visually, all
three graphs plot 11-year moving averages rather than annual data. EPA chose this averaging period to
be consistent with the recommended averaging method used by Kunkel et al. (2004) in an earlier
version of this analysis. Each average is plotted at the center of the corresponding 11-year window. For
example, the average from 2005 to 2015 is plotted at year 2010. EPA used endpoint padding to extend
the 11-year smoothed lines all the way to the ends of the period of record. Per the data provider's
recommendation, EPA calculated smoothed values centered at 2011, 2012, 2013, 2014, and 2015 by
inserting the 2010-2015 average into the equation in place of the as-yet unreported annual data points
for 2016 and beyond. EPA used an equivalent approach at the beginning of the time series.

Figures 3, 5, and 6. Change in Length of Growing Season, Timing of Last Spring Frost, and Timing of First
Fall Frost by State, 1895-2015

Figure 3 shows a map of long-term trends in the length of growing season for each of the contiguous 48
states. Figures 5 and 6 show state-level trends in the timing of the last spring frost date and first fall
frost date for the contiguous 48 states, respectively.

Each state-level trend was calculated using the Sen slope method, which is a widely accepted method of
estimating linear trends by finding the median of the slopes between all pairs of years and values. The
median slope for each state was then multiplied by the length of the period of record to estimate the
total change overtime.

Due to a lack of stations meeting the data availability criteria throughout the period of record, no state
trend could be calculated for Delaware. Several other states—Connecticut, Maryland, Nevada, Rhode
Island, West Virginia, and Wyoming—had occasional years in which data were not available.

Technical Documentation: Length of Growing Season

3


-------
7. Quality Assurance and Quality Control

NOAA follows extensive quality assurance and quality control (QA/QC) procedures for collecting and
compiling COOP weather station data. For documentation of COOP methods, including training manuals
and maintenance of equipment, see: www.nws.noaa.gov/os/coop/training.htm. These training
materials also discuss QC of the underlying data set. Pre-1948 COOP data were recently digitized from
hard copy. Kunkel et al. (2005) discuss QC steps associated with digitization and other factors that might
introduce error into the growing season analysis.

The data used in this indicator were carefully analyzed in order to identify and eliminate outlying
observations. A value was identified as an outlier if a climatologist judged the value to be physically
impossible based on the surrounding values, or if the value of a data point was more than five standard
deviations from the station's monthly mean. Readers can find more details on QC analysis for this
indicator in Kunkel et al. (2004) and Kunkel et al. (2005).

Analysis

8.	Comparability Over Time and Space

Data from individual weather stations were averaged to determine national, regional, and state-level
annual values in the length of growing season and the timing of spring and fall frosts. To provide spatial
representativeness, national, regional, and state-level values were computed using a spatially weighted
average, and as a result, stations in low-station-density areas make a larger contribution to the national,
regional, or state average than stations in high-density areas.

9.	Data Limitations

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

1.	Changes in measurement techniques and instruments over time can affect trends. These data
were carefully reviewed for quality, however, and values that appeared invalid were not
included in the indicator. This indicator includes only data from weather stations that did not
have many missing data points.

2.	The urban heat island effect can influence growing season data; however, these data were
carefully quality-controlled and outlying data points were not included in the calculation of
trends.

10.	Sources of Uncertainty

Kunkel et al. (2004) present uncertainty measurements for an earlier (but mostly similar) version of the
national and regional analyses. To test worst-case conditions, Kunkel et al. (2004) computed growing
season trends for a thinned-out subset of stations across the country, attempting to simulate the
density of the portions of the country with the lowest overall station density. The 95 percent confidence
intervals for the resulting trend in length of growing season were ±2 days. Thus, there is very high
likelihood that observed changes in growing season are real and not an artifact of sampling error.

Technical Documentation: Length of Growing Season

4


-------
11.	Sources of Variability

At any given location, the timing of spring and fall frosts naturally varies from year to year as a result of
normal variation in weather patterns, multi-year climate cycles such as the El Nino-Southern Oscillation
and Pacific Decadal Oscillation, and other factors. This indicator accounts for these factors by applying
an 11-year smoothing filter for Figures 1, 2, and 4 and by presenting a long-term record (more than a
century of data). In addition, the Sen slope analysis used in calculating trends in Figures 3, 5, and 6
minimizes the impact of outliers and statistical noise by using the median slope over the period of
record. Overall, variability should not impact the conclusions that can be inferred from the trends shown
in this indicator.

12.	Statistical/Trend Analysis

EPA calculated long-term trends in Figures 1, 2, and 4 by ordinary least-squares regression to support
statements in the "Key Points" text, as shown in Table TD-1. All of these trends were statistically
significant at a 95 percent confidence level (p < 0.05). These results corroborate earlier findings by
Kunkel et al. (2004), who determined that the overall increase in growing season was statistically
significant at a 95 percent confidence level in both the East and the West. For confirmation, EPA ran
several of the same analyses using Sen slope linear regression, which yielded nearly identical results.

Table TD-1. Linear Regression Results, 1895-2015

Parameter

Slope

(days/year)

P-value

National average

0.118

<0.0001

West

0.218

<0.0001

East

0.083

<0.0001

Spring

-0.0622

<0.0001

Fall

0.0554

<0.0001

The maps in Figures 3, 5, and 6 all show trends over time that have been calculated using a Sen slope
analysis based on a weighted annual average of stations in each state. All three sets of state-level trends
were evaluated for statistical significance at a 95 percent confidence interval. Overall, 37, 27, and 32
states had statistically significant trends for Figures 3, 5, and 6, respectively.

References

Kunkel, K.E., D.R. Easterling, K. Hubbard, and K. Redmond. 2004. Temporal variations in frost-free season
in the United States: 1895-2000. Geophys. Res. Lett. 31:L03201.

Kunkel, K.E., D.R. Easterling, K. Hubbard, K. Redmond, K. Andsager, M.C. Kruk, and M.L. Spinar. 2005.
Quality control of pre-1948 Cooperative Observer Network data. J. Atmos. Ocean. Tech. 22:1691-1705.

Technical Documentation: Length of Growing Season

5


-------