Cryospheric Climate Indicators

Introduction

Sea Ice Index

Monthly Sea Ice Images

Daily Sea Ice Images

Browse & Compare Images

Archived Data & Images

Image Derivation

Interpretation Resources

Soil Temperature

Alaska

Russia

Snow Cover

Snow Off Day

Greenness

SINDVI

Nascence

Peak

Greenness: Overview

Introduction
• Measuring greenness with AVHRR data
• Continuing the time series with MODIS data
Indicator Processing
Discussion
Statistical Analysis
Authors
Acknowledgments
References

Introduction

"Climate warming is projected to cause vegetation shifts because rising temperatures favor taller, denser vegetation, and will thus promote the expansion of forests into the Arctic tundra, and tundra into the polar deserts…Half the current tundra area is projected to disappear in this century."
- Dr. Robert W. Corell, testifying before the U.S. Senate Committee on Commerce, Science and Transportation, March 3, 2004.

Greenness, or the density and vigor of green vegetation growth, can be estimated with satellite data. Changes in greenness are related to photosynthetic activity, net primary production, and growing season length, all of which are qualities of vegetative cover that respond to changes in climate such as increases or decreases in precipitation, temperature, and insolation (solar radiation received by a plant, moderated via changes in cloud cover). Changes in greenness tell a story, then, that is somewhat more complicated than that of changes in other Arctic climate indicators such as soil temperature or sea ice.

From a time series of land surface greenness, vegetation phenology and seasonally integrated or peak greenness can be monitored. Phenology refers to periodic biological phenomena that are correlated with climatic conditions, such as time of green up (nascence) in spring and the growth phase in a plant or leaf from full maturity to death (senescence) in fall. Greenness is tied to vegetation type; therefore, indices that quantify greenness can also be used for land cover classification. For example, Wang and Overland (2004) use a commonly derived measure of greenness, the normalized difference vegetation index (NDVI), as one means to estimate changes in the area covered by tundra. Results can be viewed on the NOAA Arctic Change Indicator Web site.

Plants respond quickly to changes in climate so changes in vegetation phenology and production are evident in time series of greenness indicators. Climate induced changes in land cover may take decades to become manifest (Stow et al., 2004). Our satellite data record is long enough to register both types of changes. In general, the northern hemisphere is exhibiting a greening trend (Zhou et al., 2001) with larger greenness values, roughly 10 percent higher than about 20 years ago, (Shabanov et al., 2002; Myneni et al., 1997) an increase in the growing season of about 12 days (Shabanov et al., 2002), and a decrease in shrub free tundra area of about 18 percent over the past 20 years (Wang and Overland, 2004). These changes are consistent with warmer soil and air temperatures, earlier snow melt, and the expansion of shrubs and tree line to the north. While changes are uniformly evident over wide areas of Eurasia, changes in North America, especially Alaska, show more spatial variability and provoke a closer look.

We track more than one indicator, because they may be indicative of different processes. For example, changes in seasonally integrated NDVI may indicate a shift in vegetative production, while phenological indicators such as nascence date may provide clues to causes of increased production (Stow et al., 2004).

Measuring greenness with AVHRR data

Green vegetation absorbs more solar radiation in the visible light part of the electromagnetic spectrum than in the near infrared (IR). By differencing visible band reflectance with near IR band reflectance data from the Advanced Very High Resolution Radiometer (AVHRR), an instrument series carried on NOAA polar orbiting satellites beginning in 1978, an estimate of greenness can be made. A standard measure of greenness is Normalized Difference Vegetation Index, or NDVI.

NDVI = (near IR - vis)/(near IR + vis)

Using a ratio reduces noise in the greenness signal due to factors such as illumination differences from place to place and cloud shadows.

Globally, NDVI ranges from about 0.1 to 0.8. NDVI is lower for plants under stress or dying. NOAA and other agencies have used NDVI to monitor vegetation health operationally for many years, but arriving at a climate data record-quality data set has been challenging, because NDVI is sensitive to the calibration and processing methods used (for a discussion of the limitations of NDVI from AVHRR, visit the "caveat" pages from the NOAA National Climatic Data Center). In the Arctic, the bright background of snow and the patchy nature of some vegetation types, small leaf area, a short growing season, the low sun angle, and persistent cloudiness make it difficult to estimate NDVI with accuracy and precision. When snow cover is present as a background to vegetation, it can reduce the NDVI signal by as much as 50% (A. Huete, personal communication, July 2004). Because of this, the greening observed in spring may be due to the removal of the snow background, by new vegetation growth, or both. To a large extent, we avoid this problem by taking data between the equinoxes only (March 21 and September 23), however, some areas may still have snow. A climatologic snow cover frequency shows where lingering snow may be a problem. (See Snow Cover Frequency (1966-2003) from Armstrong et al., 2005).

Fortunately, the NOAA-NASA Pathfinder program has produced a data set that has been carefully processed to reduce uncertainty in the instrument record. The Pathfinder AVHRR NDVI data set covers July 1981 through September 2001. For most biomes (a biome is a major ecological community type), the Pathfinder NDVI data set is not contaminated by changes due to orbital drift and changes in satellite instrument, although biomes with low leaf area may be influenced by changes in solar zenith angle due to orbital drift (Kaufmann et al., 2000). Here we use the 8 km version of the Pathfinder data set (data set reference 1).

Continuing the time series with MODIS data

The Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the NASA Earth Observing System satellites offers vegetation index products beginning in September 2000. The MODIS NDVI product MOD13A2 (data set reference 2) has been called the “continuity index” (Running, 2002) because it was designed to continue the AVHRR Pathfinder NDVI data record. However, the MODIS data we use are significantly different from the Pathfinder data, with higher NDVI values.

At present (June 2006) the reasons for this marked difference are not clear. Until work is done that corrects for the difference, we must treat the data sets as two distinct data records (see Statistical Analysis for more information). Work comparing MODIS and AVHRR vegetation products for the conterminous United States shows good agreement with a linear relationship between the two and MODIS values being slightly higher (Gallo et al., 2004). Work validating MODIS vegetation indices at higher latitudes is ongoing (e.g. Verbyla, 2005).

The Greenness Indicator time series use the AVHRR PAL product and are kept up to date with the MODIS product because these products offer the best combination of accuracy, timeliness, and continuity (although continuity in a statistical sense has not yet been established).

Indicator Processing

Our indicator products (Seasonally Integrated NDVI, Nascence, and Peak NDVI) are computed from NOAA/NASA Pathfinder AVHRR Land (PAL) Data (data set reference 1) and from the MODIS/Terra Vegetation Indices 16-Day L3 Global 1km SIN Grid V004 (MOD13A2) data (data set reference 2).

Maps of satellite-derived greenness are usually composite images created over one to two weeks or more in order to compensate for cloud cover. If there is more than one image acquired over this time for which the grid cell is clear, then the highest value of NDVI for that grid cell within the compositing period is used.

The Pathfinder Composite Data Set of NDVI uses 4km AVHRR Global Area Coverage (GAC), and consists of three 10-day composites per month from 1981 through 2001 projected to a global 8 km x 8 km Goode's Interrupted Homolosine grid. The MOD13A2 uses MODIS Terra surface reflectances corrected for molecular scattering, ozone absorption, and aerosols and composite over 16 day periods, and projected to a global 1 km x 1 km grid.

At NSIDC, three yearly image products are created: peak NDVI, seasonally integrated NDVI (SINDVI), and date of nascence. Since snow, sparse vegetation and low sun angle can cause irregular NDVI results, data for winter are not used. Each year's 10-day or 16-day composite files that include data acquired on dates between the spring and fall equinoxes (nominally the growing season) are used. These are re-projected to an 8 km EASE-Grid (See All About EASE-Grid) prior to processing.

Peak NDVI images show the maximum value a grid cell obtains during a growing season, where valid ranges of NDVI are from -1.0 to 1.0.

Date of nascence can only be resolved to within a 10 or 16-day window using this data set. It is simply the sixth day within the 10-day compositing period (or the eighth day within a 16-day compositing period for MOD13A2) upon which a grid cell's NDVI value first exceeds 0.3.

Seasonally Integrated NDVI (SINDVI) is calculated by integrating the value of NDVI for file dates between the spring and fall equinoxes. So that the time period over which the integration takes place each year will be the same even though the compositing periods for MOD13A2 and PAL data files differ, interpolation is used to obtain a value of NDVI for each equinox date. Between these dates, 17 PAL files per year, or 11 MOD13A2 files per year, are required for a valid SINDVI value.

Long-term means are computed for all three products by averaging all the complete years in the PAL data record (1982-2001).

A year-by-year anomaly record is created for each product by subtracting each year's value from the long-term mean. These data are the basis for time series plots formed from small image subsets shown in the anomaly images: the North Slope of Alaska, the lowland tundra of Canada, and the boreal forest of Siberia. These areas are about 248 km square (31 x 31 8 km grid cells). Each boxed region's spatially averaged value of peak NDVI, SINDVI, and day of nascence is computed and plotted as a yearly time series. The same is done for each boxed region's spatially averaged anomaly (the anomaly is the difference from the long term, 1982-2001, mean value).

Also shown are early (1982-1996) and late (1997-2001) anomaly images. These are generated by averaging the first five years of data and subtracting the result from the long-term mean, and doing the same for the last five years of the Pathfinder data set. This is a way of showing large-scale temporal and spatial differences in the products from the beginning to the end of the data set time series.

Discussion

Overall, the anomaly images and trend maps on the greenness indicator pages support the findings of many researchers that north of 45 deg N, green-up is happening earlier, and greenness values are higher, than at earlier times in the satellite data record. The last year of the PAL time series shows lower values, however. Concerned that these may be an artifact of the degrading instrument orbit, we asked Dr. David Douglas, USGS, to compare NDVI from SPOT imagery with our PAL values for our areas of interest. His results gave a good match between the two sensors for a four year overlap period and lead us to conclude that the drop in values in 2001 was real.

Large-scale changes in greenness have been shown to be correlated with large-scale atmospheric variability patterns such as the North Atlantic Oscillation. (Gong and Shi, 2003, show that about 57% of the variance in NDVI on large spatial scales can be explained by atmospheric indices). Our maps show interesting smaller scale regional differences as well. The NDVI signature depends on interactions between vegetation, regional climate, soil substrate, and microclimate. While NDVI responds quickly to changes in air and soil temperature (biweekly variability in temperature can explain up to 50% of biweekly variability in NDVI according to Jia et al., 2004), the correlation between seasonal temperatures and seasonally integrated NDVI also depends on other factors, such as vegetation type (Walker et al., 2003). Serreze et al. (2000), in discussing vegetation changes along with other observed high-latitude changes, made mention of additional factors that may contribute to positive trends in NDVI. These included increasing carbon dioxide and nutrients that may be freed up as higher temperatures thaw previously frozen soil.

To summarize, if one is attempting to understand spatial patterns in greenness indices, ancillary data (such as land cover classification maps, elevation, and soil type) are needed to attribute changes in greenness to specific causes. Greenness becomes useful as a climate indicator when changes over relatively long time periods and large areas are measured, and higher NDVI values can be associated with warmer temperatures and larger amounts of above-ground phytomass (Walker et al., 2004).

One simple way to make sure that our greenness indicator time series are taken from relatively homogeneous areas that are responding similarly to climate change is to sample from a single biome. We have chosen three areas: a boreal forest biome in Siberia, a lowland tundra area in Canada, and a lowland tundra area on Alaska’s North Slope. Within biomes, elevation is an important source of phytomass variation (Walker et al., 2003). We chose areas with fairly uniform elevation.

NDVI and land cover change on the North Slope have been studied extensively, and NDVI has been shown to respond differently to changing climate depending on soil type, the acidity of the tundra, and vegetation type (Walker et al., 2003; Stow et al., 2003; Jia et al., 2004). We therefore divided our North Slope sample into areas of “typical tundra”, or bioclimate subzone 4, and “southern tundra”, or bioclimate subzone 5. The boundary between these subzones roughly marks a transition from north to south from more lakes, fewer shrubs and non-acidic tundra, to a greater number of shrubs and more acidic tundra (Walker et al., 2003).

Figure 1. Topography and bathymetry (left) and vegetation distribution (right) of the Arctic, with areas used in our analysis marked by squares. The images are from Vital Arctic Graphics, a publication of the United Nations Environmental Program, H. Ahlenius, Editor-in-Chief.

Statistical Analysis

Peak NDVI values within the regional subsets for 1982 through 2005 were explored using boxplots for homogeneity within and between regions, and analyzed for trends. Each region consists of 31 x 31 8km EASE-Grid grid cells centered at 69.4 N latitude and 155.3 W longitude (Alaska, EASE-Grid row 408, column 276); 61.1 N latitude and 95.6 W longitude (Canada, EASE-Grid row 146, column 473); and 67.2 N latitude and 116.8 E longitude (Siberia, EASE-Grid row 786, column 387).

Exploratory data analysis shows that over the 20 years in the PAL series, the distributions of peak NDVI for Canada and Siberia subsets are fairly symmetrical, while that from Alaska is skewed (Figure 2). When the Alaska region is broken into coastal and inland subsets (Figure 3), it is clear that these regions have distinctly different NDVI distributions, corresponding to different biomes (and elevation ranges) as remarked on in the Discussion section.

Figure 2. Boxplots for all PAL data years by regional subset. The circles on the box plots represent the outmost 5% of data values, while the box with median line encompass the interquartile range of 50% of the data values.

Figure 3. Boxplots for the Alaska region (left) and for 15 x 15 grid cell subregions corresponding to more inland (left) and coastal (right) areas.

The distribution of peak NDVI values at each region was plotted as a time series (Figure 4 shows the series for Siberia), and linear regression analysis, using the median value for each year, performed. (For a quick overview of linear regression see Linear Regression for Trend Analysis: Assumptions and Limitations in the Sea Ice Index indicator documentation.)

Of the five regions modeled using the PAL data set, only the Siberia subset and the coastal North Slope subset show slight but significant (at 95% confidence level) trends. These are 0.0044 and 0.0038 NDVI unit increases per year respectively.

Figure 4. The distribution of peak NDVI values for each year for the Siberia subset. Series for other regions are not shown.

When the MODIS series is added, the median value falls outside the regression model prediction interval for each of the six successive years. Figure 5 shows results for Siberia. The plot for the coastal North Slope subset looks similar. This leads us to conclude that MODIS and PAL peak NDVI are not similarly distributed, and therefore MODIS peak NDVI should not be treated as a seamless continuation of the PAL peak NDVI data set. While this analysis was done for peak NDVI only, it is clear that SINDVI and nascence from MODIS should not be treated as seamless continuations of the same products from AVHRR.

Figure 5. Like Figure 4, but with the confidence interval for the linear regression over the PAL series added (red lines), and the MODIS peak NDVI distributions added (represented by orange box plots).

This analysis proves statistically what is evident by differencing MODIS and AVHRR indicator products for the one season (in 2001) for which the Pathfinder and MODIS products overlap, as shown by the difference images in Figure 6.

Figure 6. Difference (MODIS product minus AVHRR product) images for peak (left), nascence (middle), and SINDVI (right), for the 2001 growing season

Authors

Florence Fetterer and Matt Savoie developed the Greenness Indicators products and Web pages. Florence Fetterer did development research and wrote the documentation, and Matt Savoie did the programming and provided the statistical analysis.

Acknowledgments

We thank David C. Douglas, Research Wildlife Biologist, USGS Alaska Science Center, Juneau, and Alfredo R. Huete, Department of Soil, Water and Environmental Science at the University of Arizona, Tucson, for reviewing this site and offering suggestions for improvements. Dr. Huete and Dr. Douglas provided expert information concerning use of the MODIS and AVHRR NDVI products in high latitudes. Dr. Douglas also provided the animations of NDVI and melt onset over sea ice.

This site was created as a contribution to NOAA's Arctic Research Office SEARCH program, John Calder, program manager; and as an addition to the NOAA Arctic Change Indicator Web site, with funding from NOAA's Oceanic and Atmospheric Research office.

References

Armstrong, R.L., M.J. Brodzik, K. Knowles, and M. Savoie. 2005. Global Monthly EASE-Grid Snow Water Equivalent Climatology. Boulder, CO: National Snow and Ice Data Center. Digital media.

Gallo, K., L. Ji, B. Reed, J. Dwyer, J. Eidenshink. Comparison of MODIS and AVHRR 16-day normalized difference vegetation index composite data. Geophysical Research Letters, 31, L07502,doi:10.1029/2003GL019385, 2004.

Gong, D., and P. Shi, Northern hemispheric NDVI variations associated with large-scale climate indices in spring, Int. J. Remote Sensing, 24 (12), 2559-2566, 2003.

Jia, G.J., H.E. Epstein, and D.A. Walker, Controls over intra-seasonal dynamics of AVHRR NDVI for the Arctic tundra in northern Alaska, Int. J. Remote Sensing, 25 (9), 1547-1564, 2004.

Kaufmann, R.K., L. Zhou, Y. Knyazikhin, N. Shabanov, R.B. Myneni, and C.J. Tucker, Effect of orbital drift and sensor changes on the time series of AVHRR vegetation index data, IEEE Transactions on Geosciences and Remote Sensing, 38 (6), 2584-2597, 2000.

Myneni, R. B., C. D. Keeling, C. J. Tucker, G. Asrar and R. R. Nemani. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature, 386, 698-702. 1997.

National Research Council. Climate Data Records from Environmental Satellites. National Academies Press, Washington, DC. 116 pp. 2004.

Running, S.W. New Satellite Technologies Enhance Study of Terrestrial Biosphere. Eos, Transactions of the American Geophysical Union Vol 83, 41,458-460 458, 2002.

Serreze, M. C., J. E. Walsh, et al. Observational evidence of recent change in the northern high-latitude environment. Climatic Change 46(1/2),159-207, 2000.

Shabanov, N.V, Zhou, L., Knyazikhin, Y., Myneni, R.B., and Tucker, C. J., Analysis of interannual changes in northern vegetation activity observed in AVHRR data during 1981 to 1994. IEEE Trans. Geosci. Remote Sens., 40, 115-130, 2002.

Shippert, M. M., D. A. Walker, N. A. Auerbach, and B. E. Lewis, Biomass and leaf-area index maps derived from SPOT images for Toolik Lake and Imnavait Creek areas Alaska, Polar Rec., 31, 147Ð 154, 1995.

Stow, D., S. Daeschner, A. Hope, D. Douglas, A. Petersen, R. Myneni, L. Zhou and W. Oechel. Variability of the seasonally integrated normalized difference vegetation index across the north slope of Alaska in the 1990s. Int. J. Remote Sensing, 24 (5), 1111-1117, 2003.

Stow, D. A. and 23 others. Remote sensing of vegetation and land-cover change in arctic tundra ecosystems. Remote Sensing of Environment 89, 281-308, 2004.

Tucker, Analysis of interannual changes in northern vegetation activity observed in AVHRR data from 1981-1994, IEEE TGRS, 40 (1), 115-130, 2002.

Verbyla, D.L., Assessment of the MODIS leaf area index product (MOD15) in Alaska, Int. J. Remote Sensing, 26 (6), 1277-1284, 2005.

Walker, D. A., et al., Phytomass, LAI, and NDVI in northern Alaska: Relationships to summer warmth, soil pH, plant functional types, and extrapolation to the circumpolar Arctic, J. Geophys. Res., 108(D2), 8169, doi:10.1029/2001JD000986, 2003.

Wang, M. and J. Overland. Detecting Arctic climate change using Köppen climate classification. Climatic Change 67(1): 43-62, 2004.

Zhou, L., C.J. Tucker, R.K. Kaufimann, D. Slayback, N.V. Shabanov, and R.B. Myneni. Variations in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999. J. Geophys. Res., 106, 20,069-20083. 2001.

Data set reference 1

• Dataset_Creator: Peter Smith and Tong Zhu
• Dataset_Title: NOAA/NASA AVHRR Pathfinder Land 10-day and Monthly FTP Binary Data
• Dataset_Release_Place: Greenbelt, Maryland, USA
• Dataset_Publisher: NASA Goddard Space Flight Center Distributive Active Archive Center
• Online_Resource: http://daac.gsfc.nasa.gov/data/dataset/AVHRR/01_Data_Products/04_FTP_Products/index.html

Data set reference 2

• Dataset_Creator: Land Processes Distributed Active Archive Center
• Dataset_Title: MODIS/Terra Vegetation Indices 16-Day L3 Global 1km SIN Grid V004
• Dataset_Release_Date: 2003
• Dataset_Release_Place: Sioux Falls, South Dakota, USA
• Dataset_Publisher: U.S. Geological Survey
• Data_Presentation_Form: Remote Sensing Image

Climate data record: “a time series of measurements of sufficient length, consistency, and continuity to determine climate variability and change.” (National Research Council, 2004)