Predicting riparian evapotranspiration from MODIS vegetation indices and meteorological data
Introduction
As in other arid and semiarid regions, western U.S. rivers service agricultural, urban, and environmental needs; yet they are inherently variable in flow, both seasonally and interannually (Hartmann et al., 2002). Evapotranspiration (ET) by riparian vegetation can account for 20–30% of total river discharge, as along the Rio Grande in New Mexico (Dahm et al., 2002), so river water budgets require estimates of riparian ET. However, obtaining accurate estimates has been challenging because a riparian floodplain is a complex mosaic of species associations with different canopy characteristics and percent vegetation cover (%C) (Dahm et al., 2002, Goodrich et al., 2000, Zamora et al., 2001, Congalton et al., 1998 and U.S. Dept. of Interior, 2000).
Eddy covariance flux towers have been established to measure ET on several western rivers, including the Lower Colorado River (Craig Westenberg, United States Geological Survey, pers. comm., DeMeo et al., 2003), the San Pedro River (Goodrich et al., 2000, Scott et al., 2000, Scott et al., 2003, Scott et al., 2004), and the Middle Rio Grande (Cleverly et al., 2002, Dahm et al., 2002). The towers provide ET and carbon dioxide fluxes integrated over several thousand square meters (approx. 50 m × 50 m) (Cooper et al., 2003). Hence, they can provide ET estimates for typical mixed-species associations along a river. They have been deployed in riparian plant associations dominated by mesquite (Prosopis spp.; Scott et al., 2000, Scott et al., 2004, Scott et al., 2003), giant sacaton (Sporobolus wrightii; Scott et al., 2000), cottonwood (Populus spp.; Cleverly et al., 2002, Dahm et al., 2002), and saltcedar (Tamarix ramosissima; Cleverly et al., 2002). Eddy covariance is considered to be the most accurate method for measuring ET at scales of 100 m–1 km (Rana & Katerji, 2000), but the towers must be calibrated and verified for each site.
The current challenge is to find reliable methods to scale these point estimates of ET to larger river stretches and to incorporate these values into riparian water budgets (Coonrod & McDonnell, 2001, Cooper et al., 2000, Cooper et al., 2003, Dahm et al., 2002, Goodrich et al., 2000, Kustas et al., 2002, Prueger et al., 2001). Two approaches have been taken to predict ET from remote sensing data: physical models (Gillies et al., 1997) and empirical models that relate ET to vegetation index (VI) measurements over a growing season (Choudhury et al., 1994).
Physical models attempt to predict ET from the surface energy balance equation:where Rn is net radiation, G is soil heat flux, H is sensible heat flux density, λ is the latent heat of vaporization for water, and E (mm d−1) is the rate of water vapor flux from the surface to the atmosphere (Monteith & Unsworth, 1990), which is equivalent to ET in this study. Surface energy fluxes depend strongly on λE (Gillies et al., 1997, Moran et al., 1994). The simplified method (Carlson et al., 1995) is based on an equation in the form:where Rn24 and ET24 are net radiation and ET (mm d−1) integrated over a 24 hour period, respectively, To13 and Ta13 are surface radiant temperature and air temperature measured near the time of local maximum (e.g., 1300 h local time), respectively. B and the exponent n are pseudo constants that need to be determined for each application. Carlson et al. (1995) showed that these constants are functions of %C as estimated by Normalized Difference Vegetation Index (NDVI). In practice, Rn can be obtained either from ground data or by calculation, Ta13 can be approximated by the maximum daily air temperature over a canopy as measured from meteorological stations, and NDVI and To13 can be obtained from satellite sensors. This approach has been applied widely to ET measurements and has a potential accuracy of 1.5 mm d−1 (about 20–30% of actual ET) in semiarid regions (Carlson et al., 1995). It requires an accurate measure of radiometric land surface temperature (LST) and NDVI by remote sensing techniques.
VI-based methods for estimating ET are modifications of the crop coefficient method (Jensen & Haise, 1963) for estimating water demand by irrigated crops. Crop coefficients (Kc) are empirical ratios relating crop ET (ETc) to a calculated reference-crop ET (ETo) that is based on atmospheric water demand (Jones, 1983) over a crop cycle or to actual ET measurements, as in the present study. A Kc curve gives the seasonal distribution of Kc as a function over time or a time-related index, such as growing degree-days. In this form, however, Kc cannot account for variations in crop growth from field to field, as affected by soil type, nutrition, uneven water distribution, or other agronomic factors.
As an alternative, Kc can be adjusted throughout the crop cycle to take into account changes in the fraction of absorbed solar radiation (fARs) by the plant canopy (estimated by VIs) as the crop develops. A time-series of VI measurements is correlated with measured ETc or ETo to develop a VI–Kc curve over the crop cycle. Once calibrated, these VI-based Kc curves can provide close estimates of ETc within 10% of measured values among fields with different growth characteristics (e.g., Hunsaker et al., 2003).
Choudhury et al. (1994) used a heat balance and radiative transfer model to study relations among transpiration coefficients (Tc) and VIs. They provided a theoretical basis for estimating transpiration from nonstressed crops from VI and Ta data. From the relationship between ET and LAI and between LAI and VI, they developed an equation in the form:The term [1−(VImax−VI)/(VImax−VImin)]n converts VI to a scaled value (0–1) and is derived from the light extinction curve through a canopy as estimated by VIs. The exponent n depends on the crop and the VI used. The effects of soil evaporation and crop stresses added scatter and uncertainty into the ET estimates.
The present research builds on previous studies conducted on the main shrub and tree species [saltcedar (T. ramosissima), cottonwood (Populus fremontii), and willow (Salix gooddingii)] along low-elevation western U.S. rivers such as the Lower Colorado River. Previous to this work, models of ET were developed based on stem sap flow measurements (assumed to be equivalent to transpiration) and were correlated with To−Ta and micrometeorological data (Nagler et al., 2003). As expected from Eq. (2), sap flow was negatively correlated with To−Ta for all three species, although there was considerable scatter in the data. Transpiring canopies maintained leaf temperatures (Tc) about 2 °C lower than Ta at midday, while soil temperature was as much as 15° higher than Ta. Under nonstressed conditions, rates of sap flow were similar (P>0.05) among species as a function of leaf area (Nagler et al., 2003). In a following research, canopy models were developed relating VI measurements to fARs, leaf reflectance, LAI, and canopy leaf angles for each species (Nagler et al., 2004). Due to differences in canopy characteristics, LAI by itself was not a good predictor of fARs across species (Nagler et al., 2001, Nagler et al., 2004). However, NDVI was directly proportional to fARs across species. NDVI and other VIs were also proportional to %C even in mixed species associations with plants of different LAIs (Nagler et al., 2001, Nagler et al., 2004). Based on these results for riparian species, it appeared feasible to attempt to predict ET across species by either physical or VI-based models. The present study also builds on eddy covariance methods developed to measure the ET of riparian plant associations along the Middle Rio Grande River (Cleverly et al., 2002, Dahm et al., 2002).
We explored methods that can be used to scale riparian ET from flux towers to larger river stretches by using Enhanced Vegetation Index (EVI) and land surface temperature (LST) data from the Moderate Resolution Imaging Spectrometer (MODIS) sensor on the EOS-1 Terra satellite. MODIS provides daily coverage of most of the globe, and products include radiometrically corrected, reflectance data at 250 m resolution for red and NIR bands, at 500 m resolution for the blue band, and at 1-km and 5-km resolutions for the thermal bands (Huete et al., 2002, Wan et al., 2004).
The ground data set consisted of ET and micrometeorological data as previously published from four flux towers on the Middle Rio Grande in New Mexico (Fig. 1) (Cleverly et al., 2002, Dahm et al., 2002). The flux towers encompassed two vegetation associations: Rio Grande cottonwood (Populus deltoides ssp. Wislizennii) dominated, and saltcedar (T. ramosissima) dominated; and two land types: occasionally flooded or unflooded. Four years of ET data were gathered at each site. The specific objectives were (1) to determine if MODIS VIs and LSTs could be combined into an equation similar to Eq. (2) to predict ET; (2) to develop empirical models relating VIs and micrometeorological to ET according to Eq. (3); and (3) to determine if the relationship between ET and VI is species dependent. The last condition is important due to the heterogeneous nature of the riparian zone, as high-resolution imagery contains mixed riparian associations within a pixel (Nagler et al., 2001, Nagler et al., 2004). The hypothesis, based on previous research (Nagler et al., 2001, Nagler et al., 2003, Nagler et al., 2004), was that a robust relationship between ET and VI could be established across sites, years, and species associations.
Section snippets
Overview of research approach
Correlation and regression analyses were conducted on the following data sets: ET data from four tower sites over 4 years; NDVI, EVI, and LST data from MODIS; and micrometeorological data collected at the tower sites. All data sets were compiled as 16-day composite values corresponding to the standard composite VI products produced from MODIS data by the EROS data center (n=209 observations per site). A correlation matrix was constructed relating ET as the dependent variable to VIs and
ET, NDVI, and EVI values over 4 years
Time plots of ET and VIs at 16-day intervals over the study are in Fig. 2. ET and VIs had regular annual cycles, with peak values occurring between Days 130 and 290 each year. There was a close correspondence between ET and VI patterns. A two-way ANOVA showed that ET differed by sites (F=11.5, P<0.001) but not years (F=0.8, P>0.05). The unflooded saltcedar site had significantly lower ET (P<0.001) than the other three sites. NDVI and EVI ANOVA results followed the same pattern as ET.
Correlation matrix of ET, VIs, and meteorological variables
ET was
Accuracy of eddy covariance data
Flux measurements made over tall, irregular canopies, like these, require careful calibration to produce reliable data (Rana & Katerji, 2000, Scott et al., 2004). Eddy covariance results at the saltcedar sites reported in this study have been validated in several studies. Cooper et al. (2003), working at the Bosque del Apache site, compared ET estimated by an eddy covariance tower set over a saltcedar stand with a lidar-based spatial analysis of the moisture field over the same stand. Over a
Acknowledgements
This research was supported by the National Aeronautics and Space Administration (Earth Science Enterprise, Carbon Cycle Science, Applications Program, Grant #00-OES-08, NAG13-2001).
References (51)
- et al.
A new look at the simplified method for remote-sensing of daily evapotranspiration
Remote Sensing of Environment
(1995) - et al.
Relations between evaporation coefficients and vegetation indices studied by model simulations
Remote Sensing of Environment
(1994) - et al.
Seasonal estimates of actual evapo-transpiration from Tamarix ramosissima stands using three-dimensional eddy covariance
Journal of Arid Environments
(2002) - et al.
Spatial source-area analysis of three-dimensional moisture fields from lidar, eddy covariance and a footprint model
Agricultural and Forest Meteorology
(2003) - et al.
Spatial and temporal properties of water vapor and latent energy flux over a riparian canopy
Agricultural and Forest Meteorology
(2000) - et al.
The effect of applied water on the water use of saltcedar in a desert riparian environment
Journal of Hydrology
(1997) - et al.
Seasonal estimates of riparian evapotranspiration using remote and in situ measurements
Agricultural and Forest Meteorology
(2000) - et al.
Overview of the radiometric and biophysical performance of the MODIS vegetation indices
Remote Sensing of Environment
(2002) - et al.
Evaluating the effects of sub-pixel heterogeneity on pixel average fluxes
Remote Sensing of Environment
(2000) - et al.
Estimating crop water deficit using the relation between surface-air temperature and spectral vegetation index
Remote Sensing of Environment
(1994)
Assessment of vegetation indices for riparian vegetation in the Colorado River delta, Mexico
Journal of Arid Environments
Leaf area index and Normalized Difference Vegetation Index as predictors of canopy characteristics and light interception by riparian species on the Lower Colorado River
Agricultural and Forest Meteorology
Measurement and estimation of actual evapotranspiration in the field under Mediterranean climate: A review
European Journal of Agronomy
Transpiration of cottonwood/willow forest estimated from sap flux
Agricultural and Forest Meteorology
Interannual and seasonal variation in fluxes of water and carbon dioxide from a riparian woodland ecosystem
Agricultural and Forest Meteorology
The understory and overstory partitioning of energy and water fluxes in an open canopy, semiarid woodland
Agricultural and Forest Meteorology
The water use of two dominant vegetation communities in a semiarid riparian ecosystem
Agricultural and Forest Meteorology
Correcting eddy-covariance flux underestimates over a grassland
Agricultural and Forest Meteorology
MODIS vegetation index compositing approach: A prototype with AVHRR data
Remote Sensing of Environment
Photosynthesis, respiration and conservative carbon use efficiency of four field grown crops
Agricultural and Forest Meteorology
Irrigation water management: Irrigation water needs
Training Manual 3
Using remote sensing and GIS to compute evapotranspiration in the Rio Grande bosque
Mapping and monitoring agricultural crops and other land cover in the Lower Colorado River Basin
Photogrammetric Engineering and Remote Sensing
Evapotranspiration at the land/water interface in a semi-arid drainage basin
Freshwater Biology
Cited by (0)
- 1
Tel.: +1 505 277 9341; fax: +1 505 277 5355.
- 2
Tel.: +1 520 621 3228; fax: +1 520 621 5401.
- 3
Tel.: +1 805 893 4541; fax: +1 805 893 2578.