B53D-01
North American Landscape Phenology: a 250m spatial resolution product derived from MODIS
Time series from remote sensing data provide the unique capacity to estimate landscape level phenology parameters for large regions through time. This presentation will describe a 250m x 250m spatial resolution phenology product for North America from 2001 to 2005. The phenology parameters are estimated from MODIS Normalized Difference Vegetation Index (NDVI) derived from the 8-day surface reflectance product. The algorithm for estimating phenology parameters is based on the TIMESAT software using enhancements that capitalize on the MODIS quality assurance data and an iterative procedure to fill gaps in space and time. The results provide the most spatially complete, highest resolution phenology product ever produced for the entire North American continent. These data are freely available through an on-line query and order system. The presentation will review the enhanced TIMESAT algorithm, provide a description of the phenology parameters included in the product, and an overview of the data distribution system. The advent of the US National Phenology Network will allow for coordination and disseminate of phenological observations at multiple scales and across scientific disciplines. The products presented here will be integrated into the NPN so that they can 1) promote landscape phenology research within NPN and 2) utilize NPN field and network data for validation studies. The presentation will describe our planned coordination with NPN. http://accweb.nascom.nasa.gov
B53D-02
Dynamic Phenological Patterns in Tropical Ecotonal, Disturbed, and Regenerating Forests
We examined patterns of optical- phenologic variability in intact, disturbed, and transitional evergreen broadleaf tropical forests in the Amazon, using space-borne hyperspectral (Hyperion) and moderate resolution satellite measurements from MODIS. Moderate resolution satellite data provide high frequency but poor spatial resolution data of limited spectral content, while hyperspectral data offer finer resolution and spectral detail but at infrequent time intervals. Our goal was to investigate landscape phenology patterns in complex tropical rainforests and assess the extent, magnitude, and synchrony of phenology patterns in response to disturbance and variations in climate. We found wide variability and large seasonality in spectral signatures in the tropical forests analyzed here, with much of the variation occurring in forest conversion and regenerating areas of varying age classes and type of secondary forest regrowth. Unique phenology responses to seasonal drought periods were observed across all vegetation types, both spectrally and temporally. Phenologic variability in moist tropical forests were largely driven by the availability of solar radiation, however, this enhanced dry season, greening signal became weaker with less developed regenerating forests, and was completely reversed in pasture and agriculture areas, where moisture limitations became the dominant control on phenology. Regenerating forests also appear to exhibit leaf flushing and green-up, prior to the start of the new rainy season. Thus, optical-phenology variations were observed in response to increases in physiologic activity as well as canopy structural changes. The more complex phenology patterns of the regenerating forests, with simultaneous browning and greening, were partly a result of the contribution of understory vegetation reflectances that mix with those of the forest canopy. This drying effect was more pronounced in the younger and more open forest canopies, relative to the older, closed, and more developed regenerating forests. The right combination of spectral, spatial, and temporal detail are needed for improved tropical forest phenology characterization for use in carbon and ecosystem production models.
B53D-03 INVITED
Spatio-Temporal Statistical Methods for Monitoring of Land Surface Phenology
The onset of greening, the start of senescence, the timing of the maximum of the growing season and the growing season length are frequently calculated phenology metrics based on satellite imagery. However, since it is complicated to validate the much coarser spatial resolution observations of land surface phenology even with networks of ground observations of phenology, it is often unclear as to what the land surface phenology metrics actually quantify. For example, in northern biomes, the greatest increase in a satellite derived vegetation index indicated as "start of the season" (SOS) in some methods, can often be due to snow melt. The end of the greenness (EOS) metric, on the other hand, could measure an extended period of cloudiness instead of actual vegetation senescence. Since the relationship between satellite and ground observations of phenological events is ambiguous, many techniques have been developed. Here I discuss and compare some of the more commonly applied methods to derive land surface phenology metrics: delayed moving average method, percent threshold method, quadratic models based on accumulated growing degree-days and the MOD12 phenology product, which relies on piecewise sigmoidal models. I demonstrate the methods using both NDVI and EVI time series from 2001 and 2007 derived from MODIS/Terra+Aqua Nadir BRDF-Adjusted Reflectance 16-Day L3 Global 0.05Deg CMG V005 (MCD43C4) data for North America north of 30°N. To remove snow-covered pixels, I use the snow and ice QA flags. To compare the methods I evaluate the spatio-temporal differences in phenological metrics and discuss the bias- variance dilemma that involves the trade-off between an over-fitted model that is too complicated and an over- smoothed model that is too simple. Finally, I discuss the related issue of the possibility of statistically comparing the land surface phenology metrics for two years of data which represent extremes in the polarity of the North Atlantic Oscillation (2000 and 2007).
B53D-04
Integrating ground observations of phenology with remotely sensed measurements: A 2007 growing season experiment at Sevilleta LTER
The use of satellites to monitor land surface phenology is important for understanding local and regional ecosystem variability, identifying change over time, and potentially predicting ecosystem response to short and long-term changes in climate. However, the relationship between how phenology is expressed on the ground and how it is interpreted from satellites is poorly understood because phenological stages do not always correspond well to changes in spectral reflectance. Rather than focusing on phenological stages (e.g., first leaf, first flower), the ground measurements in this study focus on changes in ecosystem greenness during the 2007 growing season. We collected bi-monthly measurements of community greenness in two perennial grasslands at the Sevilleta National Wildlife Refuge in central New Mexico, a Long Term Ecological Research (LTER) site. One site is dominated by blue grama grass (Bouteloua gracilis); the other is dominated by black grama grass (Bouteloua eriopoda). Grama grasses grow during the summer/fall time period, with onset of greenness typically occurring mid-July and peak greenness occurring in September. Bi-monthly ground measurements were collected from July 2, 2007 – October 4, 2007 within systematically arrayed 30x30 cm quadrats. Within each quadrat, we recorded percent green cover (grass or forb), percent non- photosynthetic cover, and percent soil. A nadir oriented digital photograph was also taken of each quadrat, from which a greenness index was calculated. Field sampling was timed within two days of an ASTER satellite image acquisition. Here, we compare three greenness measurements from ground sampling, digital photography, and ASTER satellite imagery for the 2007 growing season. We show the degree of correlation between the three measurements through time and draw inferences about how satellite imagery can be used to assess ecosystem phenology. This study is an important first step in furthering the linkage between remotely sensed phenologies and community-level ecosystem phenologies.
B53D-05
Remote sensing data assimilation for a prognostic model of vegetation phenology
Since vegetation and climate interact dynamically, a two-way coupling of vegetation phenology will lead to a better representation of the terrestrial water and carbon cycle in weather forecasts and climate predictions. We investigate four existing prognostic phenology schemes for use in global climate simulations: The C/N and DGVM scheme as part of the Community Land Model Version 3 (NCAR), the GSI scheme as part of the Simple Biosphere Model Version 2.5 (CSU/ETH), and the Joint UK Land Environment Simulator (UK Metoffice). We document their highly variable performance in various climatic environments. Our aim is to provide a vegetation phenology scheme which can prognose a more realistic seasonal-to- interannual variability of transpiring leaves in response to climatic forcings on a global scale. To achieve this goal, MODIS-derived phenological states are assimilated into the above (existing) phenology schemes by use of the Ensemble Kalman Filter. Through simultaneous states and parameter estimation the large uncertainty of biome- dependent phenological parameters (e.g. minimum temperature required for leaf-out) can be reduced. However, this step involves a careful quantification of uncertainties in the cloud- and aerosol-contaminated satellite data. Next we show how the constrained parameter set improves the seasonal variability of predicted phenological states for a global range of ecosystem types in various climate zones. Interannual variability of a extra-tropical climatic region is validated by use of a reconstructed start-of-season dataset covering deciduous temperate broadleaf vegetation in the Swiss lowland plateau. The currently sparse coverage of long and consistent intra- seasonal validation datasets for such models justifies new autonomous ground-based observational networks and reconstruction of phenological states from proxy datasets. Upon completion of this NASA Energy and Water Cycle sponsored research project a generally-applicable prognostic phenology scheme including a satellite data assimilatoin framework will be provided to the climate modeling community. It will be suitable for two applications: a) processing a global phenological reanalysis dataset (assimilation mode) b) predicting phenological states in future climate simulations (prognostic mode)
B53D-06
Constraining spatial patterns and secular trends of springtime phenology with contrasting models based on plant phenology gardens and carbon dioxide flux networks
Shifts in the timing and distribution of spring phenological events are a central feature of global change research. Most evidence, especially for multi-decade records, indicates a shift towards earlier spring but with frequent differences in the magnitude and location of trends. Here, using two phenology models, one based on first bloom dates of clonal honeysuckle and lilac and one based on initiation of net carbon uptake at eddy covariance flux towers, we upscaled observations of spring arrival to the conterminous US at 1km resolution. The models shared similar and coherent spatial and temporal patterns at large regional scales but differed at smaller scales, likely attributable to: use of cloned versus extant species; chilling requirements; model complexity; and biome characteristics. Our results constrain climatically driven shifts in 1981 to 2003 spring arrival for the conterminous US to between -2.7 and 0.1 day/23 years. Estimated trend differences were minor in the biome of model development (deciduous broad leaf forest) but diverged strongly in woody evergreen and grassland areas. Based on comparisons with the normalized difference vegetation index (NDVI) and a limited independent ground dataset, predictions from both models were consistent with observations of satellite-based greenness and measured leaf expansion. First bloom trends, which were mostly statistically insignificant, were also consistent with NDVI trends while the net carbon uptake model predicted extensive trends towards earlier spring in the western US that were not observed in the NDVI data, showing the implication of model application outside the biome range of initial development.
B53D-07
Spring phenology in taiga and tundra
According to several studies, the onset of spring has tended to get earlier in the last few decades. However, most studies analyze the phenological variations either for a short time period (since 1982 with satellite observations), or for a restricted region using ground observations. Ground observations, satellite observations and modeling were analysed jointly to study phenological variations in boreal Eurasia in 1936-2005, and 1920-2005 in Central Siberia. The results show that the trend that is observed by remote sensing is essentially due to a shift at the end of the 1980's, related to a shift in the spring temperature. In West Siberia and European Russia, the trend to an earlier spring has existed since as early as 1940. In contrast, the central and eastern parts of Siberia display successive trends with opposite signs, and the trend observed by remote sensing is due to both very early leaf appearance in the 1990's and to very late leaf appearance in 1983-1984, showing that the trend must not be extrapolated to predict future phenology. The green-up model was then applied over the low arctic tundra region. It reproduces the green-up dates estimated using remote sensing correctly, with a RMS difference of 5 days, and it reproduces the ground observations of dwarf birch leaves in Alaska with less than 3 days error, showing that a model based on air temperature is able to predict low arctic phenology. The model was applied over the whole low arctic region from 1958 to 2002. In North East Canada and North East Russia, no remarkable trend is found in the timing of green- up, whereas a ten day advance is recorded in the last few decades in North Alaska and in North West Siberia.