By Michon Scott and Walt Meier
The satellite record on sea ice dates back to late October 1978. The National Snow and Ice Data Center (NSIDC) tracks and reports sea ice extent—area of ocean with at least 15 percent sea ice concentration—from the start of the satellite record through the present day.
A record-high extent, such as the Antarctic maximum on September 21, 2014, or a record-low extent, such as the Arctic minimum on September 17, 2012, is noteworthy. By itself, however, an isolated value, even an extreme one, says less about what is happening with sea ice than a decades-long time series. By assessing daily values in both the Arctic and Antarctic, NSIDC sees the big picture: the highs, lows, and whether sea ice exhibits a clear trend over time.
How the Arctic and Antarctic differ
Despite their similarities—distance from the equator and cold climates—the two polar regions have important differences. They experience opposite seasons and have roughly opposite land-ocean arrangements. These differences affect sea ice in multiple ways, and ignoring the differences can lead to misinterpretations of sea ice behavior.
Arctic sea ice behavior has historically been tied to global warming, driven in part by Arctic amplification. From November 1978 to March 2025, Arctic sea ice extent showed a declining trend in all months. Meanwhile, Antarctic sea ice has been linked more closely to natural variability. Antarctic sea ice began exhibiting a string of low extents in 2016, which might or might not signal a long-term change; more time and observations are needed to be sure. Prior to the string of low-extent years, Antarctic sea ice showed a slightly positive trend.
NSIDC scientists have observed another difference between Arctic and Antarctic sea ice extent trends. The Arctic trend is statistically significant, while the Antarctic trend is not.
Identifying statistical significance
In simple terms, a trend is statistically significant if what it shows is unlikely to arise from chance. Statistical significance in a time series results from the number of data points and how widely individual data points vary from one another. Read on to understand the details.
The process of determining statistical significance starts with a null hypothesis, something to be disproven. In the case of a time series, the null hypothesis would hold that the data show no trend, so identifying a trend means disproving the null hypothesis. The next step involves deciding how much confidence is desired to reject the null hypothesis and consider the trend statistically significant. Typically, the desired level of confidence is 95 percent, meaning there is less than a 5 percent chance that the identified trend is due to chance. In statistical terms, this would be designated as a probability value (P) of less than 5 percent, or P < 0.05. Desired confidence levels can be higher or lower, e.g., a confidence level of 99 percent (P < 0.01) or a level of just 50 percent (P < 0.5).
A scientist hoping to meet a 95 percent confidence level can use statistical techniques to derive that confidence level from the data. A simpler approach is to calculate the range that corresponds to twice the standard deviation. Standard deviation identifies how spread out the individual data points are from the mean, which is itself calculated by adding all the values and dividing the sum by the number of values.
Researchers rarely calculate averages, trends, or standard deviations manually. Software, including widely available spreadsheet packages, can do the calculations quickly and easily. Here, NSIDC provides a glimpse into the underlying approaches.
To calculate standard deviation, calculate the mean of all the data points, identify how far each individual data point falls above or below the mean, square that difference for each data point, then average the squared differences. This identifies the variance. Calculate the square root of the variance to arrive at the standard deviation.
Here is an example of standard deviation calculation, using sea ice extent measurements. To keep the example simple, the sample size and measurements are all small: 15, 24, 27, 12, and 17 square kilometers.
First, calculate the mean:
(15 + 24 + 27 + 12 + 17 = 95 square kilometers) / 5 measurements = 19 square kilometer mean
Next, identify the difference between each measurement and the mean, then square the difference (and remember, the square of a negative number is positive):
15 - 19 = -42 = 16
24 - 19 = 52 = 25
27 - 19 = 82 = 64
12 - 19 = -72 = 49
17 - 19 = -22 = 4
Then sum the squares and divide by the number of measurements:
(16 + 25 + 64 + 49 + 4 = 158 square kilometers) / 5 measurements = 31.6 square kilometer variance
Finally, take the square root of the variance:
√31.6 square kilometers = 5.62 square kilometer standard deviation
These calculations can be cumbersome, but they are useful, especially when working with large amounts of data. One standard deviation captures roughly 68 percent of the data points, and two standard deviations capture roughly 95 percent.
The standard deviation for a time series trend helps determine statistical significance. Calculating the standard deviation of a trend is somewhat more involved than calculating the standard deviation for a mean, but it follows a similar process. Instead of calculating individual data point distances from a mean, this process calculates data point distances from the expected value along the trend line. If adding or subtracting two standard deviations from the trend changes the trend from positive to negative or vice versa, the trend is not statistically significant. An example illustrates this test.
The February monthly trend for Arctic sea ice from 1979 to 2025 showed a decline of 2.7 percent per decade. The monthly trend for Antarctic sea ice over the same years was nearly identical, a decline of 2.6 percent per decade. While the trends were similar, the standard deviations were very different. They showed that only one of the trends was statistically significant.
Arctic: 1979 to 2025 February trend ± two standard deviations is -2.7 percent ± 0.4 percent, giving a range of -3.1 to -2.3. The negative sign remains unchanged when adding two standard deviations. This trend can be considered statistically significant.
Antarctic: 1979 to 2025 February trend ± two standard deviations is -2.6 percent ± 3.3 percent, giving a range of -5.9 to +0.7. Adding two standard deviations changes the sign from negative to positive. This trend cannot be considered statistically significant.
Visual clues to significance
Mathematical calculation is the surest way to identify statistical significance, but looking at anomalies over a time series also gives quick visual clues to significance.
The same kind of visual clues come from graphs of the linear trend lines derived from individual data points, incorporating R2 values. R2 (R-squared) is a measure of how well—or badly—the trend line fits the data points. It is also known as the “goodness of the fit” of the trend line. The closer the R2 value is to 1, the more confidence can be placed in the trend.
Context adds credibility
More data points are better than fewer data points, provided they are measuring the same thing at a regular interval. For example, blood pressure readings taken every morning reveal more information if they span the last year than if they only span the last week. Observations of monthly sea ice extent, recorded every January, yield more information if they include the last sixty years than if they only include the last six years.
The continuous satellite record now spans decades, with daily polar observations dating back to October 1978. If the satellite record were a person, it would be middle-aged by now, old enough to have grown children and taken on a mortgage (if it could afford one).
Satellites have provided the most regular, expansive surveys of polar ice, but scientists are not limited to satellite observations. Pre-satellite historical records also yield data. The value of these data is especially pronounced in the Arctic. People have lived around the Arctic Ocean for thousands of years. Explorers have made documented attempts to cross the Arctic Ocean for centuries, often looking for shorter trade routes between eastern and western Eurasia, and North America. During the twentieth-century Cold War, submarines took detailed readings of sea ice from below. In contrast, Antarctica is not on the way to or from any populous place. Even though the frozen continent now enjoys a continuous human presence devoted primarily to research, recorded observations of Antarctic sea ice are more recent.
These historical observations were spottier than satellite observations, but the records still provide clues to sea ice behavior before the late 1970s.
So, as much as individual events such as record highs or lows matter, they are better understood in context. Studying sea ice data over time—whether a time series shows a trend, whether any trend is statistically significant—helps scientists, decision makers, and everybody else better understand the past, and prepare for the future.