Vol. 81, No. 1, January 4, 2000, p. 1.
Young Students, Satellites Aid Understanding of Climate-Biosphere Link
Michael A. White, Mark D. Schwartz, and Steven W. Running
For more information, contact Michael A. White, NTSG, School of Forestry, University of Montana, Missoula, MT 59801 USA; E-mail: mike@ntsg.umt.edu
Copyright 2000 American Geophysical Union
Data collected by young students from kindergarten through high school are being combined with satellite data to develop a more consistent understanding of the intimate connection between climate dynamics and the terrestrial biosphere. Comparison of the two sets of data involving the onset of budburst among trees and other vegetation has been extremely encouraging.
Surface-atmosphere interactions involving exchanges of carbon, water, and energy are strongly affected by interannual variability in the timing and length of the vegetation growing season, and satellite remote sensing has the unique ability to consistently monitor global spatiotemporal variability in growing season dynamics. But without a clear picture of how satellite information (Figure 1) relates to ground conditions, the application of satellite growing season estimates for monitoring of climate-vegetation interactions, calculation of energy budgets, and large-scale ecological modeling is extremely limited. The integrated phenological analysis of field data, satellite observations, and climate advocated by Schwartz [1998], for example, has been primarily limited by the lack of geographically extensive and consistently measured phenology databases.
 |
Kindergarten through 12th grade students gathering data in a joint NASA/National Oceanic and Atmospheric Administration/National Science Foundation program is designed to be simultaneously educational for the students and useful for the scientific community. The program, Global Learning and Observations to Benefit the Environment (GLOBE), involves atmospheric, soil, biologic, and hydrologic measurements.
Designed in 1998, the GLOBE budburst protocol was widely implemented as an optional special measurement for 1999. Students permanently marked two branches of two trees of the dominant upper-canopy species. They were cautioned to select native species away from irrigation and fertilization sources.
Budburst monitoring took place at the students' GLOBE biology study sites, at quantitative land cover sample sites, and in some cases on school grounds. The intention was to select trees most representative of the dominant overstory species. The trees were observed daily until budburst.
Fifty-one schools, 26 of them in the United States and most of the rest in Western Europe, participated in 1999. Some reported data for a single species and some for up to six species. Some students also investigated climate-vegetation interactions with temperature and moisture bioclimatic indices. Budburst data was extracted for all 26 participating U.S. schools. For schools with more than one reported date, the mean date of budburst was calculated. Collected data goes on the Web.
Figure 1 shows the satellite-derived onset of greenness (analogous to the start of the growing season) for the conterminous United States derived from a modified version of an algorithm presented by White et al. [1997]. Yet the usefulness of any such satellite information is fundamentally determined by an understanding of how ground vegetation conditions correspond to satellite estimates.
Student observations do show that dates of leaf budburst correspond remarkably well with satellite growing season estimates. Spring 1999 budburst dates of dominant upper-canopy species, measured by students participating in the GLOBE program, tended to occur at around half the maximum annual satellite-measured greenness (Figure 2).
 |
The data strongly suggest that over many sites in the continental United States, onset of greenness occurs at approximately the initiation of upper-canopy growth in the deciduous broad leaf forest biome. Lower-canopy vegetation activity is therefore highly likely to be responsible for the satellite signal prior to onset.
Variation in vegetation phenology, especially the timing of springtime leaf and shoot growth, which can vary by more than 1 month from year to year, is an easily detectable signal of vegetation responses to both short- and long-term climatic variability. (Phenology is the study of recurring biological cycles and their connection to climate.)
Long-term vegetation phenology records of the initiation and completion of the growing season reveal strong climatic influences on the length and timing of the growing season [Menzel and Fabian, 1999] with enormous implications for many fields of geophysical research.
The timing of continental leaf growth patterns is related to many aspects of lower-atmospheric meteorology, including lapse rates, humidity, and wind direction [Schwartz, 1992]. Net carbon assimilation in eastern U.S. deciduous forests is also extremely sensitive to small variations in the timing of spring growth [Goulden et al., 1996; White et al., 1999].
Also, the presence or absence of a photosynthetically active canopy exerts a strong control on radiation partitioning into sensible and latent heat fluxes. This in turn has major implications for weather and climate modeling. There are few, if any, other easily detectable signals of vegetation-climate interactions with such broad implications.
With rare exceptions, ground phenological records are of limited duration or geographical extent. Large, consistently measured data sets of native species phenology are especially lacking, leaving the task to satellite remote sensing [Reed et al., 1994].
Data obtained from optical remote sensing, though, are limited by cloud contamination, sensor calibration, and in particular, an inadequate understanding of how satellite observations relate to vegetation developmental status. For example, incorporating satellite estimates of a fully active canopy that in fact is still emerging could result in severe errors in ecological and climate models.
Satellite estimates of the timing and length of the growing season must therefore be carefully interpreted with consistently obtained observations of ground phenology. Since most variation in growing season length occurs in spring, budburst observations are the most useful tool for testing satellite algorithms.
On the satellite side, the normalized difference vegetation index (NDVI) is computed from Advanced Very High Resolution Radiometer (AVHRR) measurements of radiation in near-IR and red wavelengths with global daily coverage at a 1.1 km spatial resolution. NDVI is a commonly used metric of ecosystem level greenness and photosynthetic activity calculated as the difference between near-IR and red reflectances divided by their sum. Mathematically, NDVI can range from -1 to 1 but in practice ranges from about 0.05 to 0.7 for land surfaces not covered by snow or clouds.
White et al. [1997] created a ratio of NDVI ranging from zero to one and identified the onset of greenness as the date at which the ratio exceeds 0.5. On average, the 0.5 level corresponds to the period of maximum NDVI increase in the spring and maximum NDVI decrease in the fall. The main advantage is that the algorithm predicts the start of the growing season at half the maximum greenness regardless of the absolute magnitude of site NDVI, thus making the use of constant and arbitrary NDVI thresholds unnecessary.
Based on limited field data from White et al. [1997], the 0.5 level seems to correspond to initial overstory leaf expansion, an appropriate initiation of the growing season for large-scale ecological models that often do not include an understory. This research is meant to develop a more complete understanding of the relationship between the satellite methodology and ground conditions.
Originally designed to operate on a full year of NDVI data, the technique was modified to extract site-specific NDVI thresholds useful for real-time monitoring. For each budburst location, we extracted 1989-1997 AVHRR biweekly composite data from the Earth Resources Observation Systems Data Center. (Compositing reduces cloud contamination and data volume by extracting channel data at the date containing the maximum NDVI within the compositing period [Holben, 1986]).
Since cloud contamination exists even in composited data, we screened the data to remove cloudy periods. For each year, we found the minimum and maximum NDVI and computed the NDVI at the half-maximum level (NDVIhalfmax). We calculated the mean NDVIhalfmax for each GLOBE budburst pixel based on the 8-year record. (The record for 1994 was incomplete and not used). NDVIhalfmax, while an absolute threshold, is still sensitive to vegetation conditions at each site.
The satellite-derived onset of greenness for the continental United States was computed as the week in which the 1999 weekly composite time series exceeded historical NDVIhalfmax (Figure 1). For the GLOBE sites, the same threshold method was used on daily time series created from cloud-screened weekly composite data interpolated with a spline fit. Composited satellite data were available only for weeks 13-26; onset of greenness was therefore detectable after week 13. GLOBE budburst data outside this range of dates could not be used and were discarded. Onset of greenness dates were then compared with the recorded dates of budburst for the GLOBE sites.
 |
Figure 3 shows application of the procedure to two sites. In the lower line in Figure 3, observed budburst was recorded prior to an increase in the NDVI data. Onset of greenness was predicted 12 days later, leading to one of the largest errors in the data set.
In the upper line in Figure 3, onset of greenness and budburst occurred within 2 days during a period of rapid NDVI increase. For every site except one, budburst was recorded during a period of increasing NDVI (positive slope for the three composite periods closest to budburst). Both panels show that period 16 (April 16-22) was cloud-contaminated, as it was for many eastern sites. We also extracted the NDVI at which budburst was measured (NDVIbud).
NDVIhalfmax was positively correlated with the NDVIbud, but the slope of the relationship was less than one (Figure 2a). For points to the right of the 1:1 line (Figure 2a), this indicates that budburst was recorded at an NDVI higher than NDVIhalfmax. Prediction errors (date of predicted onset of greenness, date of observed budburst), on the other hand, were negatively correlated with NDVIbud (Figure 2b). When NDVIbud was lower than NDVIhalfmax, errors were positive.
This pattern is at least partially explained by the natural progression of species phenology versus the aggregate picture obtained from satellites. For example, data for maples tend to fall to the left of the 1:1 line (Figure 2a), indicating that maples initiate growth early in canopy development, leading to low NDVIbud and positive errors (Figure 2b). The further a point is from the 1:1 line, the more the site phenology diverges from the overall phenological trend seen from satellites.
The mean absolute error for the comparison was 5.2 days while budburst standard deviation was 11.9 days. This indicated that using the satellite algorithm to predict the onset of greenness was significantly more accurate than using the mean date of budburst, a necessary condition for any satellite monitoring technique. Even when using the GLOBE database consisting of numerous native species of varying phenologies, prediction bias was only 0.20 days.
Likely sources of error include selection of trees not representative of general canopy development; monitoring of a vegetation type, such as crops, not representative of the dominant land cover; and misregistration of satellite data. Also, computing site-specific NDVIhalfmax was a better method than choosing arbitrary thresholds. A range of constant absolute NDVI threshold from 0.3 to 0.45 produced significantly larger errors and biases, with the best threshold at the mean NDVIhalfmax of 0.37.
GLOBE budburst data are also available for many other sites outside the continental United States. Our data selection was guided by the availability of a processed satellite data set. In future years, the same comparison should be carried out for GLOBE sites throughout the world with a similar processed data stream from the soon-to-be-launched Terra satellite. In particular, the analysis should be expanded to evergreen forests and grasslands. The current GLOBE budburst activities do not address grassland phenology, but a greenup measurement protocol is under development for implementation in 2000.
GLOBE is also pursuing a lilac budburst protocol as an expansion of a measurement network originally maintained by the U.S. Department of Agriculture [Schwartz and Marotz, 1988]. In the lilac protocol, schools make a minimum 5-year commitment to planting, maintaining, and monitoring cloned individuals of Red Rothomagensis (a lilac shrub, Syringa chinensis).
Lilac budburst data, along with other phenological stages, will be reported to GLOBE in the same way as native species data. The two protocols are related and highly complementary. The native species data provide information about vegetation activity as seen from satellites while the lilac protocol provides a genetically identical response to climate dynamics and weather systems.
Educators interested in participating in the phenology protocols can visit GLOBE on the Web (http://www.globe.gov) and go to the “Learn About GLOBE” section. Contact Mark Schwartz (mds@uwm.edu) for more information on the lilac program.
We wish to thank Dixon Butler for his support of the budburst protocol and Rebecca Boger, Pete Jackson, and Ian Sprod for technical assistance and for processing the GLOBE budburst data.
Goulden, M. L., J. W. Munger, S.-M. Fan, B. C. Daube, and S. C. Wofsy, Exchange of carbon dioxide by a deciduous forest: Response to interannual climate variability, Science, 271, 1576-1578, 1996.
Holben, B. N., Characteristics of the maximum-value composite images from temporal AVHRR data, Int. J. Remote Sens., 7, 1417-1434, 1986.
Menzel, A., and P. Fabian, Growing season extended in Europe, Nature, 397, 659, 1999.
Reed, B. C., et al., Measuring phenological variability from satellite imagery, J. Veg. Sci., 5, 703-714, 1994.
Schwartz, M. D., Phenology and springtime surface-layer change, Mon. Weather Rev., 120, 2570-2578, 1992.
Schwartz, M. D., Green-wave phenology, Nature, 394, 839-840, 1998.
Schwartz, M. D., and G. A. Marotz, Synoptic events and spring phenology, Phys. Geogr., 9, 151-161, 1988.
White, M. A., S. W. Running, and P. E. Thornton, The impact of growing-season length variability on carbon assimilation and evapotranspiration over 88 years in the eastern U.S. deciduous forest, Int. J. Biometeorol., 42 139-145, 1999.
White, M. A., P. E. Thornton, and S. W. Running, A continental phenology model for monitoring vegetation responses to interannual climatic variability, Global Biogeochem. Cycles, 11, 217-234, 1997.
