PROJECT SUMMARY
Enhancing accuracy of water and carbon flux estimates at the regional, continental, and global scales is a critical part of improving understanding of the interactions between global climate change and land surface biospheric processes. Current approaches that scale between regional estimates with data from remote sensing, eddy covariance flux, and intensive plant- and stand-level flux measurements assume that plot level estimates of extremely small areas are representative of the region. The timing of leaf flush and subsequent expansion (phenology) during spring, which in tree species is highly sensitive to climate change, in turn has a profound impact on lower atmospheric energy-mass exchange through differential carbon assimilation and transpiration totals across the landscape. However, spatial variations in the timing of spring phenology at the community level have not been systematically measured and analyzed. If large, these variations may reflect gradients in plant growth during spring leaf development that could foster systematic errors in seasonal fluxes of equal or greater magnitude than those during the height of the growing season. Thus, phenological data collected in a spatially explicit manner offer considerable opportunities for gauging landscape-level spatial variations crucial for accurate scaling-up of flux measurements to larger areas or downscaling regional-scale atmospheric circulation models. In this project, spatial variability of spring phenological data will be measured and analyzed at the community level, and then compared to the microenvironment and remote sensing measurements.
INTELLECTUAL MERIT
This project addresses issues that are significant for advancement across the fields of climatology, plant physiological ecology, and remote sensing. The nature of phenological variability in space and time has not been previously recorded over a large area and combined with supporting measurements. Understanding stand-level spatial patterns of spring plant phenological development (and the processes that drive them) may be the key knowledge needed to improve landscape level estimates of evapotranspiration and carbon accumulation derived from moderate resolution remote sensing data. The results will contribute to broader understanding of landscape variability, atmosphere-biosphere interactions, and the type of measurements that are necessary to accurately scale-up flux measurements to regional and continental areas.
BROADER IMPACTS
Students will be closely involved with the research tasks in this project. One to two students will use the project data to support development of their doctoral dissertations. Students will collect and analyze the phenological data during each of the spring field campaigns. I will work to enhance the phenology components of an existing alternative high school program that serves “at risk” American Indian students of the Lac Du Flambeau Reservation, and help these students connect their observations to the project’s scientific measurements. The detailed spatially concentrated phenological measurements and spatial analyses that proceed from this study will lay the foundation for future work that explicitly links phenological variations with plant physiological responses. The spatially concentrated phenological measures produced by this study will provide concurrent remote sensing studies being conducted at the ChEAS site with a record of spring plant development and growth that contains vastly more information about species differences, spatial variability, and precise event timing that has typically been recorded in the past. These measures will present new opportunities for comparison and validation of the satellite-derived measures, and may lead to consideration of adoption of similar phenological monitoring schemes at additional sites. Overall, the results of this project will also contribute to better understanding of the impacts of climate change on the biosphere, which will increase knowledge of potential future changes, and may allow for better planning relative to societal impacts of biospheric changes.
RESEARCH FINDINGS
The results from our first round of analyses, which involved building a ground observation-derived “pixel” (termed “landscape phenology”) for comparison to satellite-derived measures of the onset of spring (from MODIS vegetation indices) were completed using 2006 and 2007 observations. The results of these analyses show that satellite-derived measures compare well with surface phenology measures, especially when the coarse resolution of the satellite measures is taken into consideration. We placed light sensor under selected groves of trees and then compared these measurements to reference sensors placed in the open. The idea was to develop a general understanding of how our manual phenology observations compared to an objective measure of light interception by the developing canopy (more directly comparable to satellite and flux tower measurements). The initial results are encouraging, and further data will be collected in 2009. We expect this to help in reducing the need for manual observations in future studies without loss of data integrity. The expansion of the South study area to a full 600 x 600m and addition of a similar-sized North study area added a much larger sample of trees, but also a number of different microenvironments and plant communities. The North study area has a larger contrast in air temperature accumulations than the South, so we expected that this might lead to detectable variations in phenology. Also, maples (red and sugar, the dominant species in the North study area) seem to be more adaptable to different microenvironments than trembling aspen, balsam fir, and alder (the dominant species in the South study area). Indeed, variations in phenology apparently driven by differences in micro-environment (primarily air temperature) show up for red maple, sugar maple in the North study area, and aspen in the expanded South study area. These variations should allow development of co-variation phenological models that account for both environmental factors and genetic variability, as originally proposed by this study. However, red maple in the South study area shows an inverse relationship (colder air temperatures related to earlier phenology) suggesting other environmental factors are dominant in this case.
We have continued to compare our manual phenology observations to light interception levels recorded by HOBO pendant light sensors placed under known deciduous canopy types. Our goal is to develop a general relationship, such that an approximation of average manual phenology can be reliably reproduced from the light sensor data, in anticipation of the future when manual spring phenology observations will no longer be recorded at this study site. Also, these analyses will provide a useful sequence for application of our methods to other locations. For 2008 red and sugar maple observations, we have evidence to suggest that the light sensor data do not start to change noticeably until the trees reach the 400-level on the manual phenology scale. We are also exploring the use of remote-sensing filtering/smoothing software for processing of the light sensor data, given that some of the same decision rules should hold (e.g., levels should not go down).
Using all tree phenology data gathered in the springs of 2006, 2007, and 2008, we explored the impact of decreasing the sampling frequency from every other day, to every 4, 6, and 8 days. One of the issues to be resolved was simple cost and logistics. Initially I made an ad hoc assumption when developing this project that daily observation would not be necessary. In so doing, the cost was halved by moving to phenology observations only every other day. However, that effort still requires four observers to cover the full expanded study area with its 888 trees in two 625 x 625 study areas. So the goal was to see how further decreases in sampling frequency would affect error rates. We first calculated the standard error associated with observations of all species at the every other day level. Next we calculated how much error was introduced by larger sampling intervals, if we filled in the “missing” observation times with linear interpolations of the existing data. These results show that average absolute errors are not much greater for taking observations every four days, than the standard errors associated with every two day observations. Error rates nearly double when the observation period is further increased to every six days. Thus, we feel that these results justify reducing phenology observations to every four days, as will be done for our last spring phenological observation campaign (supported by a supplement), since it will cut the number of needed observers from four to two. Also, since the observers will still go our every two days, we will still have observations on half of the trees in the combined study areas every two days, thus further increasing our options to account for any increased error.
One of the first results of our work (in the initial study area) demonstrated that changes in air temperature across the study area were small, and did not seem to have a differential impact on tree phenology. We are now convinced this was primarily because in that location when the microenvironment changed (from aspen upland to alder swamp) a species change also occurred. Thus, we did not have examples of trees of the same species existing in highly contrasting microenvironments. As our work continued into the expanded study areas, especially in the north 625 x 625 area, we recognized a very different situation. First of all, we had much greater topographic variation across the study area, which resulted in much greater microclimatic differences in temperature, but also the maple species appeared across in all of these microenvironments, thus showing a much more plastic/tolerant response to environmental differences that aspen of alder. So after first noting the microclimatic differences in the north study area in spring 2008, we made some adjustments to the HOBO temperature sensor locations in spring 2009 to verify the 2008 patterns. We then used all these data to produce a map of average departures of growing degree hour accumulations at each site, and analyzed these results with a simple clustering routine to produce a four-level “regionalization” of the north study area.
While average phenological dates for “adjacent” microclimate regions (i.e., Colder/Slightly Colder, Warmer/Slightly Warmer, etc.) were not always significantly different, further separated micro-regions always were (at the .05 α level or better). Also, the standard error of the phenological values within each micro-region ranged from 0.34 to 1.26 days (average of 0.54); with only of the sixteen values being more than 0.62, and two being less than 0.41). So it is our contention that the similarity of these standard error values across the different micro-regions represents the generally uniform level of phenological variation caused by essentially random genetic differences between the individual trees. Since our regionalization has also produced significant differences in average phenologies that can be associated with accumulated air temperature differences, we are confident that our analyses have separated the effects of presumed genetic differences from environmental (air temperature) effects on tree phenology. We need to continue to refine these analyses in order to determine the best way to present these results for ease of interpretation and replicability. We were fortunate to have considerable variation among the spring phenology sequences over the five observation years. Our first spring (2006) began quite early, but was slowed by a substantial cold snap in mid-May (including accumulating snow). Spring 2007 also started early and stayed warm throughout the observation period, thus also ending quite early. In contrast, spring 2008 was the coolest of the years, starting about two weeks later than 2006 and 2007. Spring 2009 was the one “intermediate” year, and was perhaps the only one that was close to the longer-term average spring sequence when compared to century-scale climate records (Schwartz et al. 2006). Lastly, spring 2010 turned out to be the earliest of the five years in the study, even though (as in 2006) it was slowed by an early-May accumulating snow storm (however, unlike 2006, the period of colder temperatures was several days shorter, thus having less delaying effect on the tree phenologies).
Comparison of the effects of increasing sampling interval (by removing observations days from the full sequence to create “degraded” four, six, and eight day sampling series) were quite revealing, both collectively and by species. Average phenological measurement uncertainty showed a minimal increase (less than a 50% increase) as the sampling interval rose to four days, but nearly tripled when the interval was further increased to six days. Additionally, species differences were present and consistent, with speckled alder showing the least change, sugar maple showing the most, and the other three major species (balsam fir, red maple, and trembling aspen) displaying similar responses.
The benefits of high spatial and temporal resolution monitoring of phenological changes in the spring, which have been demonstrated by this project, can also be realized in autumn. Indeed, we know much less about autumn phenology than spring, and its impact on the growing season is increasingly being realized to be equally important. For example, at the Park Falls/WLEF flux tower, the time when carbon flux goes to zero (end of photosynthetic activity) varied by 30 days over the 1997-2009 period. Thus, for a five-week period (from Sept. 20th to Oct. 22nd, 2010, two observers monitored leaf coloring levels and leaf fall levels for the deciduous trees in the study areas. The protocol was similar to that used in the spring, except that coloring and fall levels were recorded simultaneously. The values were 800, 810, 850, and 890 (for less than 10%, 10-50%, 50-90%, and more than 90% or leaves colored) and 900, 910, 950, and 990 (for less than 10%, 10-50%, 50-90%, and more than 90% or leaves fallen). Transitions were rapid, and part of the initial coloring period was missed. However, the rapid transition in both coloring and leaf fall was captured, and the manual observations were verified with light sensor data. Further study over additional years appears justified.