PROJECT SUMMARY
Atmosphere-biosphere interactions are a key component of the Earth’s physical system. Understanding this interaction is a crucial part of efforts to improve global change simulation models, and monitor variations in the carbon budget. While atmospheric data generally are available for most land areas, no such network exists to collect comparable information about biosphere activity. Recent satellite information can contribute to the development of global biospheric databases. In order to realize this potential, these remotely sensed data must be carefully calibrated with surface information. This approach is the most straightforward way of deducing the physical basis of, and developing a historical context for, the satellite’s observations. This study will combine models and techniques produced in a previously funded project with surface vegetation data sets, surface energy balance and carbon flux data, and new remote sensing products, to further develop a physical interpretation of the first appearance of spring foliage, commonly called the “green wave”, in mid-latitudes. The green wave is particularly important because it is one of the crucial biospheric variables necessary for accurate modeling of processes such as water balance and net primary productivity. This project will first adapt existing empirical models to explore 20th century green wave geographic and temporal variability in portions of North America, Europe, and Asia. Next, the relationship between satellite information and surface data will be refined for selected land cover types (biome), using high resolution phenology products developed by the EROS Data Center and Michael White. These results will facilitate interconnection of the various atmospheric and biospheric changes associated with the green wave phenomenon, forming the conceptual basis for a future mid-latitude green wave model. Therefore, this project will demonstrate that satellite bioclimatology is more effective at providing needed information for atmosphere-biosphere simulation models and global change monitoring, when used in concert with surface phenological data and models.
RESEARCH FINDINGS
In the first phase of the project, existing “Spring Indices” multiple regression-based models were adapted to explore 20th century Spring “green wave” geographic and temporal variability in portions of North America, Europe, and Asia. These models were developed to serve as bases of comparison between native plant species and satellite-derived phenology. The Spring Indices (SI) models were first slightly modified to remove region-specific information used to optimize performance in eastern North America, and then compared to the original test data set to assess stability and accuracy. Using input data from western North America, Estonia, Germany, and China, processing for each region was as follows: 1) assessment and removal of locations where the lilac-honeysuckle indicators could not be grown successfully, based on annual chill-hour and growing degree-day totals; 2) calculation of the SI “suite of measures” for selected station-years, which includes SI composite chill date, SI first leaf date, SI first bloom date, last 28 F (-2.2 C) frost date, damage index (SI first leaf date minus last frost date), and the average annual temperature; 3) comparison of the model output to local phenological data to assess performance; 4) selection of stations with sufficient data for computation of 1961-1990 normals; 5) conversion of data to annual departures-from-normal; and 6) assessment of regional trends in the spring indices and average annual temperature.
Spring seasons across North America over the 1900-1997 period were examined using the Spring Indices suite of measures and actual lilac phenological data. Regional differences were detected, as well as an average five to six day advance toward earlier springs, over a thirty-five-year period from 1959-1993. Driven by seasonally warmer temperatures, this modification agrees with earlier bird nesting times, and corresponds to a comparable advance of Spring plant phenology in Europe, both described by other researchers. These results also align with trends toward longer growing seasons, reported by recent carbon dioxide and satellite studies. North American Spring warming is strongest regionally in the northwest and northeast portions of the continent. Meanwhile, slight autumn cooling is apparent in the central USA.
In preparation for application throughout global mid-latitudes, the Spring Indices were tested against existing regional lilac bloom data sets in the western USA, Estonia, Germany, and China. The results indicate that the SI models are performing in other regions with error levels similar to those displayed in its developmental region (eastern USA). A further model test in Germany showed that annual departures of the spring index first bloom model over the 1959-1993 period correlated with comparable multi-species plant data at the +0.9 level. Analyses similar to those done in North America have also been completed in China. Results show a five day advance toward earlier last spring frost dates in China, but no changes in first leaf or first bloom dates. The last frost date changes appear strongest in the northeastern portion of the country. Analyses in the former USSR were hampered by the data set “in hand” ending in 1989, and the lack of lilac phenological observations for comparison.
In the second phase, the relationship between satellite-derived and surface phenology data were refined for the deciduous broadleaf forest and mixed woodland land cover types, by comparisons to two high resolution phenology products developed by the EROS Data Center and researcher Michael White for the conterminous USA. Surface phenology data sets were aggregated and assigned to 10 km x 10 km (100 pixel) “windows” for further analysis and linkage to the satellite-derived data. Differences between the satellite SOS dates and the surface phenological dates were calculated for each case, and then the cases were stratified by general BATS land cover type. The satellite and spring index phenology dates were further compared to native species dates at Harvard Forest, MA to assess the representativeness of both measures.
The remote-sensing analyses utilize two sets (developed by Reed and White) of improved satellite-derived start-of-season (SOS) dates, obtained for 10×10 pixel “windows” around the stations. The analysis period runs from 1990-1999, and the new techniques give actual “day of year” output for each station-year, rather than the previous bi-week values. Both techniques perform reasonably well (median absolute errors of about 6 days) in tests conducted using the deciduous broadleaf tree and mixed woodland land cover types. However, White-SOS dates are closer in time to SI Bloom dates, while Reed-SOS dates are considerably earlier and errors” show a relationship to time of year. White-SOS tracks overall yearly trends better, but with somewhat greater average “error.” “Error” in both techniques varies considerably by year. While the Reed SOS and White SOS offer improvements over earlier techniques, their temporal resolution is still hampered by the NDVI biweekly satellite-derived data sets from which they are derived.
In conclusion, this project has shown that continental-scale phenological models, developed in North America ,can be applied effectively to other regions in the Northern Hemisphere, and provide an effective means for monitoring one aspect of global change in mid-latitudes. The remote sensing-related results further validate the feasibility and desirability of developing time lines that interconnect the various atmospheric and biospheric changes associated with the green wave phenomenon. Such connections will form the conceptual basis for improved future mid-latitude green wave modeling studies, and may help clarify net ecosystem exchange and energy budget data obtained from eddy covariance tower sites.