IMPROVING PREDICTIVE CAPABILITIES OF ENVIRONMENTAL CHANGE WITH GLOBE DATA
This dissertation addresses two applications of Normalized Difference Vegetation Index (NDVI) essential for predicting environmental changes. The first study focuses on whether NDVI can improve model simulations of evapotranspiration for temperate Northern (> 35˚) regions. The second study focuses on whether NDVI can detect phenological changes in start of season (SOS) for high Northern (> 60˚) environments. The overall objectives of this research were to (1) develop a methodology for utilizing GLOBE data in NDVI research; and (2) provide a critical analysis of NDVI as a longterm monitoring tool for environmental change. GLOBE is an international partnership network of K-12 students, teachers, and scientists working together to study and understand the global environment. The first study utilized data collected by one GLOBE school in Greenville, Pennsylvania and the second utilized phenology observations made by GLOBE students in Alaska.
Results from the first
study showed NDVI could predict transpiration periods for environments like
Greenville, Pennsylvania. In
phenological terms, these environments have three distinct periods (QI, QII,
and QIII). QI reflects onset of the
growing season (mid March – mid May) when vegetation is greening up (NDVI <
0.60) and transpiration is less than 2mm/day.
QII reflects end of the growing season (mid September - October) when
vegetation is greening down and transpiration is decreasing. QIII reflects height of the growing season (mid
May – mid September) when transpiration rates average between 2 and 5 mm per
day and NDVI is at its maximum (>0.60).
Results from the second study showed that a climate threshold of 153 ±
22 growing degree days was a better predictor of SOS for Fairbanks than a NDVI
threshold applied to temporal AVHRR and MODIS datasets. Accumulated growing degree days captured the
interannual variability of SOS better than the NDVI threshold and most closely
resembled actual SOS observations made by GLOBE students. Overall, biweekly composites and effects of
clouds, snow, and conifers limit the ability of NDVI to monitor phenological
changes in Alaska. Both studies did show
that GLOBE data provides an important source of input and validation
information for NDVI research.
While the Earth surface temperature has increased steadily over the 20th century and is projected to continue, the effects on the hydrological cycle (amount and timing of precipitation, evapotranspiration, discharge rates, and extreme events) have differed across the globe (McCarthy et al. 2001). There have been higher precipitation levels and more flooding in mid and high latitudes and annual fresh water discharge to the Arctic Ocean has increased (Karl and Knight 1998; Jacobs et al. 2001; Milly et al. 2002; Peterson et al. 2002). As temperatures continue to rise, further hydrological changes are expected with more variability between regional and local environments (Jackson et al. 2001; Palmer and Raisanen 2002).
Two critical parameters needed for predicting these changes are evapotranspiration (ETP) and soil moisture. Two-thirds of the precipitation that falls over land each year comes from evapotranspiration from soil and vegetation and soil moisture is a major driver of the ETP process (Chahine 1992). In biomes that have distinct winter seasons, start of spring phenological events, specifically timing of budburst and green-up of leaves, coincides with transpiration. Seasons leave annual signatures that reflect the dynamic nature of the hydrologic and carbon cycles and link the different spheres (atmosphere, biosphere, hydrosphere, and pedosphere/lithosphere) of the Earth system. Furthermore, observed shifts in earlier spring phenological events, both plant and animal responses, have been attributed to anthropogenic climate change (Root et al., 2003; Root et al., 2005). Thus signatures, such as global plant waves of green-up and green-down, can monitor natural and anthropogenic fluxes in the environment.
Given these documented temperature, precipitation, and phenological changes, a corresponding increase in evapotranspiration would be expected. However, with the exception of Israel, in the past 50 years pan evaporation rates have decreased in India, former Soviet Union, and the United States (Peterson et al. 1995; Chattopadhyay and Hulme 1997; Cohen et al. 2002). This pan evaporation paradox has been explained by global decreases in solar irradiance and increases in cloud cover and aerosol levels (Ramanathan et al. 2001; Ohmura and Wild 2002; Roderick and Farquhar 2002). Brutsaert and Parlange (1998) contend that pan evaporation is only a good indicator of potential evaporation in environments with ample land-surface moisture supply and that the decreasing pan evaporation rates actually indicate an increase in terrestrial evaporation. Additionally, extrapolating daily or seasonal canopy transpiration rates from single leaf porometer measurements, the most widely used instrument for measuring stomatal conductance, is difficult and requires simulation modeling (Pearcy et al., 1992). Therefore, as the Earth continues to warm, monitoring start of season (SOS) and accurately simulating transpiration becomes increasingly more important for predicting effects of climate change on carbon and hydrologic cycles.
Normalized difference vegetation index (NDVI), with its spatial and temporal extent, has been an instrumental tool for monitoring inter- and intra-annual transpiration and phenological changes on the Earth’s surface (Running and Nemani, 1988; Lloyd, 1990; Reed et al., 1994; Myneni et al. 1997; Suzuki et al., 1998; and Tucker et al., 2001; Shabanov et al., 2002; Piao et al., 2006). NDVI, derived initially from Advanced Very High Resolution Radiometer (AVHRR) data, is the normalized difference between surface reflectance of infrared and red bands and has a near linear relationship with photosynthesis and transpiration (Sellers, 1985). Tucker and Sellers (1986) found NDVI provided information on transpiration capacities of plant canopies in addition to photosynthesis, the more common application. Early applications of NDVI were for classifying land cover and monitoring vegetation dynamics (Tucker et al., 1985; Justice et al., 1986; Townsend and Justice, 1986). Subsequent applications have expanded to include: 1) land cover mapping and change detection (DeFries et al., 1998; Hansen et al., 2000; Loveland et al., 2000; Sturm et al., 2001, Tape et al., 2006); 2) identifying fire disturbances (Kasiscke and French, 1995; Goetz et al., 2006); 3) monitoring transpiration (Running and Nemani, 1988; Suzuki et al.; 1998); and 4) developing phenological metrics to detect changes in growing season length (Lloyd, 1990; Reed et al., 1994; Myneni et al. 1997; Tucker et al., 2001).
Zhou et al. (2001) found a positive correlation between temporal NDVI patterns and surface temperatures in Northern latitudes. Such correlations increase confidence in temperature driven simulations, such as ETP, and in turn NDVI can be used for determining the seasonal onset of transpiration periods. However, correlations between photosynthesis, transpiration, and NDVI are strongest in climates where radiation is the primary control of these two processes (e.g. Seattle) and weakest in water stressed environments (e.g. Tucson) (Running and Nemani, 1988). Furthermore, a high correlation between NDVI and evapotranspiration was found for Siberia but was less apparent over tropical regions of the Indo-China peninsula (Suzuki et al., 1998). Increases in NDVI have also been attributed to longer growing season in Northern latitudes (Myneni et al., 1997; Tucker et al., 2001; Zhou et al., 2001; Shabanov et al., 2002; Piao et al., 2006).
In spite of its wide spread use, NDVI is a surrogate measurement of plant photosynthetic activity and the translation of the actual signal requires careful consideration (Tucker, 1979; Shabanov et al., 2002). Much of the current NDVI research lacks field validation and as a result it is difficult to interpret what observed changes in NDVI mean. NDVI, like all satellite data, characterizes a pixel and rarely is the pixel homogenous, especially when the pixel size is large. Heterogeneity of the landscape, coupled with atmospheric effects from aerosols and clouds, compound the interpretation of the NDVI signal making field validation essential. GLOBE, through its suite of data collection and extensive worldwide network of students, provides a means to validate such research.
For my dissertation I developed two distinct research projects to determine if, and how, GLOBE measurements could support satellite data and algorithm evaluation. Both projects incorporated GLOBE and satellite data, specifically NDVI. Furthermore, each project provided a critical analysis of a different application of NDVI. In the first study NDVI was used to monitor transpiration and in second to monitor vegetation phenology. NDVI results were validated with GLOBE measurements as well as external data sources n both studies. While the research differed in each project, the overall objectives of my dissertation were to (1) develop a methodology for utilizing GLOBE measurements in NDVI research; and (2) provide a critical analysis of NDVI as a long-term monitoring tool for environmental change.
My first research project investigated this NDVI-transpiration relationship to determine whether NDVI could accurately predict transpiration periods for a Northern temperate climate. This project utilized a suite of measurements (atmosphere, land cover and soils) made by one GLOBE school in Greenville, Pennsylvania during 1998 through 2001. Students’ measurements were used to initialize and validate a soil-vegetationatmosphere transfer (SVAT) model (GAPS). GAPS, General-Atmosphere-Plant-Soil Simulator, a SVAT model developed by Riha et al. (2003), was used to simulate soil water and energy fluxes at this location from 1998 through 2001. Model outputs were compared to corresponding temporal NDVI time series data derived from SPOT 4 Vegetation.
My second research project investigated whether NDVI could accurately detect phenological changes, specifically start of season (SOS), in high Northern latitude (> 60˚) environments. Vegetation phenology, the relationship between climate and terrestrial plant growing seasons, has become increasingly important in climate change research. Furthermore, mid- and high Northern latitude regions have experienced the largest temperature increases during the most recent warming (1976 to present) (Houghton et al., 2001). This research encompassed phenology metrics derived from multi-temporal
AVHRR and Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI data.
These metrics, in addition to climate based metrics, were used to detect start of season for
Alaska. Results from both methods were validated with phenology observations from GLOBE students in Alaska for 2001 through 2004.