Abstract

Monitoring land surface drought using remote sensing data is a challenge, although a few methods are available. Evapotranspiration (ET) is a valuable indicator linked to land drought status and plays an important role in surface drought detection at continental and global scales. In this study, the evaporative drought index (EDI), based on the estimated actual ET and potential ET (PET), is described to characterize the surface drought conditions. Daily actual ET at 4-km resolution for April–September 2003–05 across the continental United States is estimated using a simple improved ET model with input solar radiation acquired by Moderate-Resolution Imaging Spectroradiometer (MODIS) at a spatial resolution of 4 km and input meteorological parameters from NCEP Reanalysis-2 data at a spatial resolution of 32 km. The PET is also calculated using some of these data. The estimated actual ET has been rigorously validated with ground-measured ET at six Enhanced Facility sites in the Southern Great Plains (SGP) of the Atmosphere Radiation Measurement Program (ARM) and four AmeriFlux sites. The validation results show that the bias varies from −11.35 to 27.62 W m−2 and the correlation coefficient varies from 0.65 to 0.86. The monthly composites of EDI at 4-km resolution during April–September 2003–05 are found to be in good agreement with the Palmer Z index anomalies, but the advantage of EDI is its finer spatial resolution. The EDI described in this paper incorporates information about energy fluxes in response to soil moisture stress without requiring too many meteorological input parameters, and performs well in assessing drought at continental scales.

1. Introduction

Drought generally refers to an extended period of time where water availability for soil and vegetation is lower than expected in the ecological system. Classic approaches to land surface drought monitoring are predominately meteorological records or hydrological observation from point-based data. Standard metrics such as the Palmer drought severity index (PDSI; Palmer 1965) (a list of acronyms used in this paper is given in Table 1), the rainfall anomaly index (RAI; Van-Rooy 1965), and the standardized precipitation index (SPI; McKee et al. 1993, 1995) generally require a spatial distribution of precipitation or soil moisture. These data are in coarse spatial resolutions, which limit their utility to regional or continental scales. Drought metrics based on remote sensing methods improve spatial detail without requiring an extensive meteorological precipitation network and have become the most promising tool for drought monitoring of larger regions (Lambin and Ehrlich 1996; Ghulam et al. 2007b).

Table 1.

Acronyms.

Acronyms.
Acronyms.

Normalized difference vegetation index (NDVI)-based drought metrics have been extensively used for vegetation drought monitoring during the last several decades because of their sensitivity to surface vegetation cover change from climatic effects on dryness conditions. Kogan (1990, 1995) designed the vegetation condition index (VCI) to monitor crop water stress by using the statistical data of NDVI. Chen et al. (1994) developed the anomaly vegetation index (AVI) to study land surface dryness by analyzing annual NDVI dynamics. However, the accuracy of drought monitoring based on NDVI may be affected by the variability of vegetation cover in different ecosystems and time lags between drought occurrences at some consecutive time periods and NDVI change (Ghulam et al. 2007b; Qin et al. 2008). To overcome the deficiency of traditional NDVI-based drought metrics, other hybrid NDVI-based drought measures, such as the vegetation drought response index (VegDRI), integrate traditional climate-based drought indicators, and satellite-derived vegetation indices, and provide new potential for land surface drought assessment (Brown et al. 2008).

Because land surface temperature (LST) is highly related to canopy water content or soil moisture availability, the methods that combine NDVI and LST provide a better characterization of surface drought status by providing a two-dimensional spectral feature space. Based on these features, the temperature–vegetation index (TVX; Prihodko and Goward 1997) calculated by the ratio of LST to NDVI better correlates with canopy water (Nemani et al. 1993) or soil moisture (Carlson et al. 1994) in different land cover conditions. Subsequently, the vegetation health index (Kogan 1990, 1995, 1997; Seiler et al. 1998) and the temperature–vegetation dryness index (TVDI; Sandholt et al. 2002) were designed to estimate surface dryness. Han et al. (2006) used leaf area index (LAI) instead of NDVI to analyze the surface drought status under saturated NDVI and Ghulam et al. (2007a) replaced LST with albedo to further interpret the mechanism of the vegetation condition albedo drought index (VCADI).

Evapotranspiration, a key component of the land surface water budget and an indicator linked to land drought status, is a significant process that drives energy and water exchange between the atmosphere and land surface (Priestley and Taylor 1972; Wang et al. 2007). When drought occurs, stomata of stressed plants close to minimize water loss by transpiration, which leads to decreased latent heat flux; in order to keep an energy balance, sensible heat flux may increase. As a result of this process, leaf temperature will ultimately increase (McVicar and Jupp 1998). Based on this idea, Idso et al. (1981) and Jackson et al. (1981) proposed the crop water stress index (CWSI) by calculating the ratio of actual ET to potential ET (PET) for full canopy cover to study water stress in irrigated areas. Moran et al. (1994) used the Penman–Monteith (P–M) equation and CWSI in partial canopies by integrating the vegetation cover amount derived from surface reflectance data to form the water deficit index (WDI). Anderson et al. (2007a,b) developed a simple evaporative stress index (ESI) associated with the canopy and soil surface in a plant–soil system to assess the drought conditions of the continental United States by taking advantage of the Atmosphere–Land Exchange Inverse (ALEXI) model (see also Kustas and Norman 1999, 2000). However, with these metrics it is difficult to obtain the surface-to-air temperature difference because of the scarcity of local ground measurements. The surface-to-air temperature difference reflects the degree of stress-induced stomatal closure of the canopy leading to decreased evaporation in the canopy.

Most recently, a simple improved ET estimation method based on satellite determination of surface net radiation, vegetation index, temperature, and soil moisture has been developed (Wang and Liang 2008). Some physical ET models, such as Two-Source Surface Energy Balance models (TSEB; Shuttleworth and Wallace 1985), the Surface Energy Balance Algorithm for Land model (SEBAL; Bastiaanssen et al. 1998a,b), the Simplified Surface Energy Balance approach (SSEB; Senay et al. 2007), and the Mapping ET at High Resolutions with Internal Calibration method (METRIC; Tittebrand et al. 2005; Allen et al. 2007a,b), require numerous meteorological input variables. However the simple improved ET model requires few meteorological parameters and is easy to operate for routine, long-term mapping of ET in large-scale applications (Wang and Liang 2008). The previous version of the ET model (Wang et al. 2007) was driven primarily by land surface net radiation (Rn), vegetation indices (VI), and temperature (T). Because soil moisture has a potentially important effect on ET (Detto et al. 2006; Gu et al. 2006; Krishnan et al. 2006), the improved ET model incorporates the influence of soil moisture into the ET parameterization by considering the diurnal land surface temperature range (DTSR) or diurnal air temperature range (DTAR) variable. This lays a foundation for developing the simple and applied surface drought methods based on the improved ET estimation.

In this paper, we describe the robust evaporative drought index (EDI) using shortwave products with high spatial resolution (4 km) generated by our laboratory covering the continental United States for April–September 2003–05, by integrating traditional drought indicators and energy radiation. Based on the improved ET model we analyze the validity of the EDI for regional land surface drought monitoring across the conterminous United States by qualitative examination of interannual temporal map patterns of EDI and other drought indicators. We also validate net radiation Rn and ET using extensive measurements from six ARM sites and four AmeriFlux sites. Ultimately the improved ET model incorporated into EDI may provide an optimal method for time-continuous land surface drought assessment over the continental United States.

2. Methods

a. Evaporative drought index

To highlight the soil moisture response to surface dryness, a simple EDI, following the evaporative stress index (Anderson et al. 2007b), is designed from the improved actual ET model and PET model, given by 1 minus the ET/PET ratio. Similarly, EDI is derived as:

 
formula

where ETs,c are the total modeled ET fluxes from both soil and canopy systems, respectively, and PETs,c are their respective associated whole potential ET fluxes. Values of EDI vary between 0 and 1. Generally EDI is sensitive to moisture stress and is comparable at different spatial scales. Higher EDI means more severe water stress or drying of the soil surface and lower EDI means less water stress or ample surface soil moisture.

b. The improved actual ET model

1) Model description

To reduce the input parameters of ET estimation, Wang and Liang (2008) used satellite measurements and extensive ground measurements at the Southern Great Plains (SGP) sites in the United States from January 2002 to May 2005 to develop a statistical model, given as follows:

 
formula

where Rn is net radiation, NDVI is the normalized difference vegetation index, Tad is the daytime average air temperature, and DTAR is the diurnal air temperature range.

This model indicates that the dominant factors driving seasonal variation of ET are surface net radiation, vegetation indices, temperatures, and diurnal air temperature range. Correlation coefficients between surface net radiation and ET are the highest, followed by the vegetation index, temperatures, and diurnal air temperature range (see Wang et al. 2007; Wang and Liang 2008). Considering the land cover types, which include grass, rangeland, pasture, crops, and forest, the coefficients may be representative for estimating ET over the continental United States except for the desert and glacier regions. Here ET over desert, snow, and ice are set to zero. In addition, the improved model is independently validated using AmeriFlux data and compared with Global Soil Wetness Project (GSWP)-2 ET products. The results show the improved model may supply useful information for regional and global application because the two independent datasets are in close agreement (Wang and Liang 2008). Because daily satellite land surface temperature products are not available under cloudy conditions, we select the daily NCEP-2 air temperature data and replace the DTSR with DTAR in this paper.

2) Model input data

The improved ET model has been successfully applied to global ET estimation at a spatial resolution of 1° × 1° (Wang and Liang 2008). The input data sources in this paper are described briefly below.

(i) Surface net radiation

We have produced three-yr (2003–05) incident daily photosynthetically active radiation (PAR) and daily shortwave solar radiation (Rs) products for the continental Unites States at 4-km spatial resolution from MODIS data (MOD02, MOD03, and MOD10) using a lookup table (LUT) method (Liang et al. 2006; Wang et al. 2010).

To use Rs as an input variable, we estimate daytime net radiation from shortwave radiation empirically (Wang and Liang 2009). This method is given by analyzing the relationship of Rn/Rs to daily minimum air temperature Tmin, diurnal temperature range DTR, normalized difference vegetation index NDVI, and relative humidity RH.

 
formula

In the above equation, NDVI is acquired from MODIS 16-day composite products and other parameters (Tmin, DTR, and RH) are extracted from NCEP-2 data with 32-km spatial resolution.

(ii) Vegetation index

Two global vegetation index products, NDVI and enhanced vegetation index (EVI), are available from MODIS. MODIS NDVI composite products with 1-km spatial resolution and 16-day temporal resolution are used here. Daily NDVI values are temporally interpolated from the 16-day averages (Van Leeuwen et al. 1999) using linear interpolation.

(iii) Air temperature and diurnal air temperature range

Both daytime average air temperature (Tad) and diurnal air temperature range (DTAR) are derived from NCEP-2 temperature data, covering April–September 2003–05. These data (available online at http://www.cpc.ncep.noaa.gov/products/wesley/reanalysis.html) on the Lambert Conformal Conic grid with 32-km spatial resolution are used in this paper, and Tad is extracted from average daytime NCEP-2 temperature. DTAR is calculated by the Tmax and Tmin acquired from NCEP-2 maximum and minimum temperature data. The relative humidity RH can be acquired through the average daytime NCEP-2 humidity data.

c. Method for calculating potential evapotranspiration

PET represents the ideal evaporation rate of capturing response to forcing variables if soil moisture is unlimited. To reduce model complexity and minimize meteorological data needs without decreasing the accuracy of PET estimates over continental or global areas, we adopt the Hargreaves method to estimate PET. This method has been verified as being nearly as accurate as the FAO-56 P–M method in estimating PET (Hargreaves et al. 1985; Hargreaves 1989, 1994; Hargreaves and Allen 2003; Vanderlinden et al. 2004; Khoob 2008), while at the same time being much simpler for practical use because it requires only two readily accessible parameters, temperature and solar energy. Therefore, PET can easily be estimated from NCEP-2 data. The Hargreaves model is expressed as follows:

 
formula

Here, PET is potential evapotranspiration (mm day−1), Ra is the extraterrestrial solar incident radiation (MJ m−2 day−1), Tmean is daily mean air temperature (°C), Tmax is daily maximum air temperature (°C), and Tmin is daily minimum air temperature (°C).

The difference between Ra and Rs is that Ra is top-of-atmosphere solar radiation, which depends solely on solar zenith angle; while surface solar radiation Rs depends on other clouds, aerosols, and atmospheric water vapor content. Bibliographical resources (Mimikou and Baltas 2002) provide Ra as a function of season and latitude in tabular form.

d. Aggregation methods

Considering Rs is the major contributor to net radiation and actual ET, we both aggregate the 1-km NDVI from MODIS products and disaggregate 32-km NCEP-2 data into 4-km resolution. Subpixel NDVI information at the 1-km scale has been used to estimate aggregate land surface parameters for the 4-km resolution improved ET model grids. Aggregate values (υ) of NDVI were calculated as a weighted average of values expected for each pixel:

 
formula

where ni is the number of 1-km pixels of NDVI pixel i within a given 4-km grid cell. Mean air temperature (Tad) and daytime average air temperature (DTAR) at 32-km resolution from NCEP-2 data are then disaggregated into a 4-km grid using a nearest neighbor technique.

e. Anomalies analysis and map generation

To assess the monthly temporal change of surface dryness, monthly composites of daily ET and PET are calculated from the equation below based on the improved ET model and Hargreaves equation;

 
formula

where υ(m, y, i, j) is an output variable for month m, year y, day d, pixel location i, j, and the value on total day n. After filling the continental grids by compositing, we can fuse the daily flux signature for monthly drought monitoring.

To highlight the difference of drought indices for continual years, and to be comparable with the Palmer Z index, the EDI map is shown as anomalies in monthly composited values compared to multiyear average values for the period of n years,

 
formula

The map generation procedure for the drought metrics anomalies is as follows. For the radiation flux (Rs, Rn, ET, PET, and EDI) map, daily radiation flux products with a sinusoidal projection from MODIS and NCEP-2 grid data on a Lambert Conformal Conic projection are all transformed into latitude–longitude projections, and then estimated ET with Eq. (2) and the estimated PET with Eq. (4) are integrated into monthly products using Eq. (6). The next step calculates the average monthly products (EDI and Palmer Z) and obtains the monthly anomalies map by Eq. (7). In the anomalies map, differences between EDI and Palmer Z can be presented by executing short- or long-term surface drought mapping and soil moisture mapping as well.

f. Comparison drought indices

Incorporating antecedent precipitation, moisture supply and demand, the Palmer drought indices can reflect drought impacts over different time periods and sectors (Palmer 1965). They include three primary indices used widely throughout the world. Among these indices, the Palmer moisture anomaly index (Z index) indicates the extent of soil moisture departure from the mean for each month during a short term. The PDSI, the Palmer hydrologic drought severity index (PHDSI) and the standardized precipitation index (SPI) are also developed for monitoring long-term surface drought conditions. Of these, the Palmer Z index is the most comparable to EDI, considering it has been one of the most widely used in the United States, even though the SPI has advantages of statistical consistency and the ability to reflect both short-term and long-term drought impacts. However, Palmer indices are in very low spatial resolution and cannot meet requirement of certain users.

3. Results analysis

a. Validation of the surface net radiation estimation

From the record of the National Climatic Data Center (NCDC) and the Drought Monitor (more information available online at http://drought.unl.edu/dm), the surface dryness conditions were extremely severe in 2003, slightly alleviated in 2004 and wetter in 2005 across the continental United States, as shown in Fig. 1. In 2003, extreme or severe drought conditions covered more than 20% of the continental United States. Although the drought conditions were alleviated by 2004, the areas of surface dryness still covered 10%–20% of the United States. The dry extent diminished quickly in 2005, when the areas experiencing extreme drought were less than 15%.

Fig. 1.

Percentage area of (top) dry and (bottom) wet conditions based on the Palmer drought index in the United States from January 1996 to December 2006. (Figure from NOAA; more information available online at http://www.ncdc.noaa.gov/oa/climate/research/2006/ann/drought-summary.html)

Fig. 1.

Percentage area of (top) dry and (bottom) wet conditions based on the Palmer drought index in the United States from January 1996 to December 2006. (Figure from NOAA; more information available online at http://www.ncdc.noaa.gov/oa/climate/research/2006/ann/drought-summary.html)

Rigorous validation of the surface radiation flux derived from remote sensing data described here requires time-continuous flux measurements at the scale of at least 4 km. Up to now such time-continuous flux measurements may not exist at continental scales. However, many towers of different regional areas may sample the variability of radiation flux in the heterogeneous landscape at a 4-km scale, thus we select six Energy Balance Bowen Ratio sites of the Enhanced Facility of the Atmosphere Radiation Measurement Program supported by the U.S. Department of Energy and four AmeriFlux sites for direct validation of the continental scale flux. Table 2 summarizes conditions of the six ARM and four AmeriFlux sites. The land cover types represented vary from grassland, pasture, to cropland and forest, with locations shown in Fig. 2. ET collected at AmeriFlux sites is measured by the eddy covariance (ECOR) method and corrected by the Twine et al. (2000) method because of the energy imbalance problem.

Table 2.

Brief description of six ARM sites of the SGP and four AmeriFlux sites throughout the United States. The ET is collected by the Energy Balance Bowen Ratio (EBBR) method at ARM sites and by ECOR method at the AmeriFlux sites.

Brief description of six ARM sites of the SGP and four AmeriFlux sites throughout the United States. The ET is collected by the Energy Balance Bowen Ratio (EBBR) method at ARM sites and by ECOR method at the AmeriFlux sites.
Brief description of six ARM sites of the SGP and four AmeriFlux sites throughout the United States. The ET is collected by the Energy Balance Bowen Ratio (EBBR) method at ARM sites and by ECOR method at the AmeriFlux sites.
Fig. 2.

Locations of test sites across the United States (six ARM sites and four AmeriFlux tower sites).

Fig. 2.

Locations of test sites across the United States (six ARM sites and four AmeriFlux tower sites).

For derivation of actual ET, surface net radiation (Rn) estimation is essential and its accuracy affects that of ET. Equation (3) accurately estimates Rn using MODIS and NCEP-2 data. Estimated Rn has been validated with the ground-measured Rn, and results are plotted in Fig. 3. At the six ARM sites (Fig. 3a), the bias of estimated Rn is −7.81 W m−2 (−3% in relative value), the root-mean-square error (RMSE) is 42.11 W m−2 (11% in relative value) and the correlation coefficient is about 0.93. The negative bias may partly come from the error of NCEP Reanalysis-2 datasets deviating from meteorological observations. Figure 3b shows the bias of estimated Rn at the four AmeriFlux sites to be 11.17 W m−2 (9% in relative value). The RMSE is 51.74 W m−2 (17% in relative value) and the correlation coefficient is about 0.86. The positive bias may result from the underestimated ground observations of Rn at the AmeriFlux sites. Though these results may not be as robust as those obtained using meteorologically observed data, owing to the coarse spatial resolution of air temperature and relative humidity products derived from NCEP-2 data, their accuracy is sufficient for estimating ET.

Fig. 3.

Scatterplots of estimated daytime Rn calculated with Eq. (3) and the ground-measured Rn at (a) 6 ARM sites and (b) 4 AmeriFlux sites for April–September 2003–05.

Fig. 3.

Scatterplots of estimated daytime Rn calculated with Eq. (3) and the ground-measured Rn at (a) 6 ARM sites and (b) 4 AmeriFlux sites for April–September 2003–05.

b. Validation of actual ET estimation

Daily actual and potential evapotranspiration are calculated during April–September 2003–05 because this period encompasses the agricultural crop growing cycle. Figure 4 shows monthly composites (except for April 2004, because of missing data) of daily actual ET derived with the improved ET model for this time interval. The estimated actual ETs are validated by the six ARM sites (EF02, EF07, EF09, EF12, EF18, and EF20) and four AmeriFlux sites (Blodgett, Bondville, Black Hills, and Walker Branch). Figure 5 gives the scatterplots of the times series of the ground-measured and estimated daily ET calculated from Eq. (2) with Rn, NDVI, daytime-averaged air temperature and diurnal air temperature range (DTAR) extracted from NCEP-2 data. The bias varies from −11.35 to 27.62 W m−2, the correlation efficient ranges from 0.65 to 0.86 and the RMSE varies from 45.32 to 64.81 W m−2. The biases of all four AmeriFlux sites are positive, with the worst validation results from Blodgett, which show strong bias. The cause of this bias (not evident at other sites) is difficult to determine. The positive bias of the AmeriFlux sites may partially be caused by the energy imbalance issue pertinent to the eddy covariance method, which may result in underestimation of ET (Wang et al. 2007). Wang and Liang (2008) reported a bias of −1.90 W m−2 and RMSE of 28.6 W m−2 using the same improved ET model from meteorological observations of ARM data for 12 sites at SGP. During the estimation of ET, the cumulative error of Rn and ET retrieval, plus the coarse resolution of temperature from NCEP-2 data may reduce the accuracy of ET, hence results not be as quite as good as that reported by Wang and Liang (2008). However the estimation is acceptable for assessing the surface drought conditions at continental scales.

Fig. 4.

Monthly composites of daily ET for April–September 2003–05 (April 2004 is represented by 11-day composites because of missing data).

Fig. 4.

Monthly composites of daily ET for April–September 2003–05 (April 2004 is represented by 11-day composites because of missing data).

Fig. 5.

Scatterplots of estimated ET [using the improved ET model of Eq. (2)] and ground-measured ET for each of 10 sites.

Fig. 5.

Scatterplots of estimated ET [using the improved ET model of Eq. (2)] and ground-measured ET for each of 10 sites.

From Fig. 4 monthly composites of daily ET under both clear and cloudy sky conditions, similar spatial patterns and temporal evolution trends across the continental United States are revealed. In April, enhanced ET (blue) appears in the southeast and then grows to the north and Midwest as vegetation greens up. By May, the difference of ET between east and west over the continental United States is obvious and a marked discontinuity occurs at the midregion. From June to August, the area of high ET expands as the crops start to grow in the Midwest corn belt. After the crops are harvested, the actual ET trails off gradually until September. However, interannual variability in spatial patterns exists at the same month in different years, which may be attributed to the differences in temperature, precipitation, and vegetation growth conditions from year to year. For the regions of no or sparse vegetation cover, surface evaporation occurs in the top 2–5 cm of the soil surface layer. Conductivity in deeper layers is negligible (Anderson et al. 2007a,b). Therefore a high ET rate over regions of minimal vegetation cover is closely linked to the current precipitation and cannot be sustained for the long-term. In contrast, healthy vegetation can sustain long-term ET because vegetative roots can extract moisture from deeper soil layers.

Figure 6 shows monthly composites of PET using the Hargreaves method [Eq. (4)] for May 2003–05. The total trends of PET distribution are basically consistent with each other and smoother than the actual ET is for this month, but its lower resolution decreases detailed information of PET at regional scales.

Fig. 6.

Monthly composites of PET for May 2003–05.

Fig. 6.

Monthly composites of PET for May 2003–05.

c. EDI interannual anomalies variability

To highlight the extreme dry or wetter conditions during the six months of interest for 2003–05, monthly EDI deviations (ΔEDI) of the 3-yr average are mapped (Fig. 7). The EDI anomaly (ΔEDI) is selected because it reduces the effects on inaccuracies of PET and emphasizes the difference of EDI variation by removing their identical features. Many anomalies of EDI in Fig. 7 are arbitrary because only 3 years of daily shortwave solar radiation (Rs) products with high resolution are available to establish the climatological baseline; April 2004 provides only 11-day composites because of missing data. These fields will improve as more years of data are available for calculating the anomalies. Despite this, the anomalies of EDI may reflect the variation of drought over the continental United States and provide a good example for assessing the surface drought and fluxes conditions.

Fig. 7.

Monthly composites of anomalies of EDI (ΔEDI) compared with anomalies in the Palmer Z index (ΔZ) for April–September (a) 2003, (b) 2004, and (c) 2005.

Fig. 7.

Monthly composites of anomalies of EDI (ΔEDI) compared with anomalies in the Palmer Z index (ΔZ) for April–September (a) 2003, (b) 2004, and (c) 2005.

d. Comparisons with Palmer Z index

The EDI anomaly (ΔEDI) is compared with the Palmer Z index anomaly (ΔZ; data derived from the NCDC, more information available online at http://www.ncdc.noaa.gov/pub/data/) displayed in Fig. 7. In general, there is good agreement between EDI and the Palmer Z index. Both are independent methods for surface dryness detection because the input parameters for EDI include surface net radiation, air temperature, and NDVI, while the Palmer Z index needs primarily antecedent precipitation data.

In 2003, both ΔEDI and ΔZ manifest the extremely dry conditions in west and central Texas in April and May. The wetter regions are located in the southeast and central parts of the United States in both April and May. NCDC reported that in 2003 Texas experienced the driest March–May in its 109-yr record. The dryness conditions of southern regions including Texas became wetter in June and overall the trends of the ΔEDI map is wetter than that of ΔZ. Wisconsin and Michigan are dryer than other larger contiguous regions, as shown obviously in ΔEDI. This is consistent with the report of “abnormal dryness around Lake Michigan” from Drought Monitor records; Anderson et al. (2007b) analyzed this trend as well. By July, the Palmer Z index indicates that the southwest regions range from wet to extremely dry, in accordance with the Drought Monitor records. Although ΔEDI also records this trend, it is to a lesser extent than ΔZ and Drought Monitor documents. The extreme dry conditions prevail in Utah in July while the ΔEDI map displays the opposite conditions; this demonstrates the limitation of EDI because of the complexity of terrain in Utah. In August, along the Midwest regions of the United States–Canada border, dry conditions are indicated by both indices. EDI demonstrates the severe drought stress in Kansas and Oklahoma, which was mitigated to some extent in September.

In 2004, extraordinarily dry conditions emerged from California into Washington State and then extended further into the Southeast, mainly Florida, as indicated in both indices in April. The Palmer Z index also shows that New Mexico, Colorado, and West Texas undergo wetter than average conditions, while ΔEDI reflects the opposite trend because though heavy rains have primarily taken place 20 days before, ΔEDI is the composite of only the last 11 days (because of missing data), and consequently it misses these conditions. In May, both indices highlight the dryer conditions in California and central regions from North Dakota to Oklahoma. By June, the dry conditions recorded by NCDC in Kansas and Oklahoma stand out in the ΔEDI, while the Palmer Z index does not capture this detail. But in July, the pattern of ΔEDI is not consistent with the ΔZ in both North Dakota and South Dakota. This bias of EDI may result from noise in MODIS Insolation and NCEP-2 products. Fortunately, ΔEDI and ΔZ indicate the same trends of the dryer regions occurring in the Southwest and Southeast. These dry conditions persist into August, and the Central Plains are wetter in the ΔEDI map versus the ΔZ map. In September, the extremely dry conditions prevail in the mid-Atlantic and central regions of the United States. Slight differences do exist in both the ΔEDI and ΔZ maps. Michigan shows dryer conditions in the ΔZ map but wetter in the ΔEDI map because of underestimated PET using the Hargreaves equation.

In 2005, dryness occurred throughout New Mexico, Kansas, Oklahoma, and Texas during April, as seen in both ΔEDI and ΔZ. NCDC records show that these states experienced the driest year in a long time. However, by May, these regions became wetter because of the heavy monthly precipitation, and dry conditions shift to the mid-Atlantic region. Unlike ΔZ index, ΔEDI shows dry conditions in the Northwest consistent with the NCDC reports. In June, both indices show extremely dry conditions prevalent from the Southwest and Southern Plains to the Great Lakes, then eastward to the Northeast. However EDI shows the extremely dry state in Wisconsin, which is not consistent with the Drought Monitor records. This indicates that EDI may have been in error partially caused by the accuracy of the NCEP-2 data. In July, ΔEDI and ΔZ show that a band of extreme drought stretched almost continuously from the Texas Gulf Coast to upper Michigan, while parts of the Northeast experienced short-term dryness. From August to September, extreme drought arrives at much of the Eastern seaboard and Northwest regions. From the ΔEDI and ΔZ map, Georgia, Maryland, and South Carolina experience the driest conditions during this period, which is confirmed by the Drought Monitor. These three states had the driest September in the 111-yr record despite rains from tropical systems such as Hurricane Rita. In September, the ΔEDI composite picks up signatures of drought from Michigan southward into Louisiana, but ΔZ shows the opposite conditions. The results of these discrepancies between the EDI and Palmer Z maps illustrate the positive dry bias of EDI, which may come from the bias of the actual ET and PET methods themselves.

Although EDI has a slight dry bias, there is generally good correspondence of the ΔEDI and ΔZ spatiotemporal patterns. The higher resolution of the EDI can provide a wider range of sensitivity to drought conditions than that of the Palmer Z index.

4. Conclusions and discussion

A strategy has been presented for drought monitoring by computing time-continuous fluxes over the continental United States. We describe the evaporative drought index based on the improved actual ET model. EDI incorporates energy flux information for sensitivity to soil moisture stress and has demonstrated good performance in drought mapping and assessment at continental scales.

In this paper we produced the monthly composites of EDI (except for April 2004) for April–September 2003–05. During the compositing of EDI, we obtained the daily actual ET using the improved ET model with surface net radiation (Rn), air temperature (Tad), vegetation index (VI), and diurnal air temperature range (DTAR) from MODIS shortwave solar radiation products and NCEP-2 data. The daily net radiation and actual ET was verified with ARM and AmeriFlux ground-measured data. The results showed that the bias of estimated Rn at the six ARM sites is −7.81 W m−2 and the bias of estimated Rn at the four AmeriFlux sites is 11.17 W m−2. The bias of actual ET at all validation sites varies from −11.35 to 27.62 W m−2 and the correlation coefficient varies from 0.65 to 0.86. Subsequently we compared the EDI with Palmer drought metrics and found it to agree well with the Palmer Z index, which indicates the deviations in precipitation from norms. EDI at 4-km resolution using MODIS and NCEP-2 data has a significantly finer resolution than that of the Palmer Z index. Hence, it can capture more detailed information for drought monitoring applications.

The EDI is based on satellite determination of surface net radiation, vegetation index, temperature, and diurnal air temperature range, and consequently its long-term climatological record is limited by the time span of the geostationary satellite record (Keyantash and Dracup 2002; Anderson et al. 2007a,b). The bias and noise of EDI products may stem from 1) the errors of MODIS Insolation products and NCEP Reanalysis-2 data; and 2) the biases of the ET and PET models. This work illustrates the need to increase understanding of the impact of the ET estimation model. The actual ET method is being improved under a range of many more land cover types and climate conditions.

In terms of land surface drought evaluation, we define the surface dryness as wet (EDI ≤ 0.2), normal (0.2 < EDI ≤ 0.4), moderate (0.4 < EDI ≤ 0.6), severe (0.6 < EDI ≤ 0.8), and extreme drought (EDI > 0.8) according to the relationship between EDI and the 0–10-cm soil moisture. However, in this paper we only compare EDI anomalies with Palmer Z index anomalies for April–September 2003–05 to demonstrate that EDI is a good indicator of drought conditions. EDI can provide an enhanced range of sensitivity to detect surface drought. The application of its drought classes in different ecosystems needs to be explored in the future.

Acknowledgments

We thank the anonymous reviewers for their critical and helpful comments and suggestions. We also thank Dr. Xiaotong Zhang and Dongdong Wang at University of Maryland, College Park, for their help. Evapotranspiration, net radiation, and corresponding meteorological observations were obtained from the ARM Program of the U.S. Department of Energy (http://www.archive.arm.gov/) and AmeriFlux network (http://public.ornl.gov/ameriflux/data-get.cfm). MODIS satellite data were obtained online (http://ladsweb.nascom.nasa.gov/ and https://wist.echo.nasa.gov/api/). Air temperature data were obtained from the NCEP–NCAR Reanalysis Project (CDAS) (http://www.cpc.ncep.noaa.gov/products/wesley/reanalysis.html). Palmer Z index products were obtained from the NCDC (http://www1.ncdc.noaa.gov/pub/data/). This work was in part supported by NASA (Grant NNX08AC53G), the Natural Science Fund of China (Grant 40771148), the RandD Special Fund for Public Welfare Industry of China (Meteorology): (GYHY200806022), the High-Tech Research and Development Program of China (2008AA121806 and 2009AA12Z128), and Chinese Fellowship Program (2008601058).

REFERENCES

REFERENCES
Allen
,
R. G.
,
M.
Tasumi
, and
R.
Trezza
,
2007a
:
Satellite-based energy balance for mapping evapotranspiration with internalized calibration (METRIC): Applications.
J. Irrig. Drain. Eng.
,
133
,
395
406
.
Allen
,
R. G.
,
M.
Tasumi
, and
R.
Trezza
,
2007b
:
Satellite-based energy balance for mapping evapotranspiration with internalized calibration (METRIC): Model.
J. Irrig. Drain. Eng.
,
133
,
380
394
.
Anderson
,
M. C.
,
J. M.
Norman
,
J. R.
Mecikalski
,
J. A.
Otkin
, and
W. P.
Kustas
,
2007a
:
A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation.
J. Geophys. Res.
,
112
,
D10117
.
doi:10.1029/2006JD007506
.
Anderson
,
M. C.
,
J. M.
Norman
,
J. R.
Mecikalski
,
J. A.
Otkin
, and
W. P.
Kustas
,
2007b
:
A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 2. Surface moisture climatology.
J. Geophys. Res
,
112
,
D11112
.
doi:10.1029/2006JD007507
.
Bastiaanssen
,
W. G. M.
,
M.
Menenti
,
R. A.
Feddes
, and
A. A. M.
Holtslag
,
1998a
:
A remote sensing surface energy balance algorithm for land (SEBAL). I. Formulation.
J. Hydrol.
,
212/213
,
198
212
.
Bastiaanssen
,
W. G. M.
,
H.
Pelgrum
,
J.
Wang
,
Y.
Ma
,
J. F.
Moreno
, and
G. J.
Roerink
,
1998b
:
A remote sensing surface energy balance algorithm for land (SEBAL). II. Validation.
J. Hydrol.
,
212/213
,
213
229
.
Brown
,
J. F.
,
B. D.
Wardlow
,
T.
Tadesse
,
M. J.
Hayes
, and
B. C.
Reed
,
2008
:
The Vegetation Drought Response Index (VegDRI): A new integrated approach for monitoring drought stress in vegetation.
GIScience Remote Sens.
,
45
,
16
46
.
Carlson
,
T. N.
,
R. R.
Gillies
, and
E. M.
Perry
,
1994
:
A method to make use thermal infrared temperature and NDVI management to infer surface soil water content and fractional vegetation cover.
Remote Sens. Rev.
,
9
,
161
173
.
Chen
,
W.
,
Q.
Xiao
, and
Y.
Sheng
,
1994
:
Application of the anomaly vegetation index to monitoring heavy drought in 1992 (in Chinese).
Remote Sens. Environ.
,
9
,
106
112
.
Detto
,
M.
,
N.
Montaldo
,
J. D.
Albertson
,
M.
Mancini
, and
G.
Katul
,
2006
:
Soil moisture and vegetation controls on evapotranspiration in a heterogeneous Mediterranean ecosystem on Sardinia, Italy.
Water Resour. Res.
,
42
,
W08419
.
doi:10.1029/2005WR004693
.
Ghulam
,
A.
,
Z.
Li
,
Q.
Qin
, and
Q.
Tong
,
2007a
:
Exploration of the spectral space based on vegetation index and albedo for surface drought estimation.
J. Appl. Remote Sens.
,
1
,
013529
.
doi:10.1117/1.2784792
.
Ghulam
,
A.
,
Q.
Qin
, and
Z.
Zhan
,
2007b
:
Designing of the perpendicular drought index.
Environ. Geol.
,
6
,
1045
1052
.
Gu
,
L. H.
, and
Coauthors
,
2006
:
Direct and indirect effects of atmospheric conditions and soil moisture on surface energy partitioning revealed by a prolonged drought at a temperate forest site.
J. Geophys. Res.
,
111
,
D16102
.
doi:10.1029/2006JD007161
.
Han
,
L.
,
P.
Wang
,
H.
Yang
,
S.
Liu
, and
J.
Wang
,
2006
:
Study on NDVI–Ts space by combining LAI and evapotranspiration.
Sci. China. Ser. D.
,
49
,
747
754
.
Hargreaves
,
G. H.
,
1989
:
Accuracy of estimated reference crop evapotranspiration.
J. Irrig. Drain. Eng.
,
115
,
1000
1007
.
Hargreaves
,
G. H.
,
1994
:
Defining and using reference evapotranspiration.
J. Irrig. Drain. Eng.
,
120
,
1132
1139
.
Hargreaves
,
G. L.
, and
R. G.
Allen
,
2003
:
History and evaluation of Hargreaves evapotranspiration equation.
J. Irrig. Drain. Eng.
,
129
,
53
63
.
Hargreaves
,
G. L.
,
G. H.
Hargreaves
, and
J. P.
Riley
,
1985
:
Irrigation water requirements for Senegal river basin.
J. Irrig. Drain. Eng.
,
111
,
265
275
.
Idso
,
S. B.
,
R. D.
Jackson
, and
P. J.
Pinter
,
1981
:
Measuring yield reducing plant water potential depressions in wheat by infrared thermometry.
Irrig. Sci.
,
2
,
205
212
.
Jackson
,
R. D.
,
S. B.
Idso
,
R. J.
Reginato
, and
P. J.
Pinter
,
1981
:
Canopy temperature as a crop water stress indicator.
Water Resour. Res.
,
17
,
1133
1138
.
Keyantash
,
J.
, and
J. A.
Dracup
,
2002
:
The quantification of drought: An evaluation of drought indices.
Bull. Amer. Meteor. Soc.
,
83
,
1167
1180
.
Khoob
,
A. R.
,
2008
:
Comparative study of Hargreaves’s and artificial neural network’s methodologies in estimating reference evapotranspiration in a semiarid environment.
Irrig. Sci.
,
26
,
253
259
.
Kogan
,
F. N.
,
1990
:
Remote sensing of weather impacts on vegetation in non-homogeneous areas.
Int. J. Remote Sens.
,
11
,
1405
1419
.
Kogan
,
F. N.
,
1995
:
Application of vegetation index and brightness temperature for drought detection.
Adv. Space Res.
,
15
,
91
100
.
Kogan
,
F. N.
,
1997
:
Global drought watch from space.
Bull. Amer. Meteor. Soc.
,
78
,
621
636
.
Krishnan
,
P.
,
T. A.
Black
,
N. J.
Grant
,
A. G.
Barr
,
E. H.
Hogg
,
R. S.
Jassal
, and
K.
Morgenstern
,
2006
:
Impact of changing soil moisture distribution on net ecosystem productivity of a boreal aspen forest during and following drought.
Agric. For. Meteor.
,
139
,
208
223
.
Kustas
,
W. P.
, and
J. M.
Norman
,
1999
:
Evaluation of soil and vegetation heat flux predictions using a simple two-source model with radiometric temperatures for partial canopy cover.
Agric. For. Meteor.
,
94
,
13
25
.
Kustas
,
W. P.
, and
J. M.
Norman
,
2000
:
A two-source energy balance approach using directional radiometric temperature observations for sparse canopy covered surfaces.
Agron. J.
,
92
,
847
854
.
Lambin
,
E. F.
, and
D.
Ehrlich
,
1996
:
The surface temperature-vegetation index for land cover and land cover change analysis.
Int. J. Remote Sens.
,
17
,
463
487
.
Liang
,
S.
,
T.
Zheng
,
R.
Liu
,
H.
Fang
,
S. C.
Tsay
, and
S.
Running
,
2006
:
Estimation of incident photosynthetically active radiation from Moderate Resolution Imaging Spectrometer data.
J. Geophys. Res.
,
111
,
D15208
.
doi:10.1029/2005JD006730
.
McKee
,
T. B.
,
N. J.
Doesken
, and
J.
Kleist
,
1993
:
The relationship of drought frequency and duration to time scales.
Preprints, Eighth Conf. on Applied Climatology, Anaheim, CA, Amer. Meteor. Soc., 179–184
.
McKee
,
T. B.
,
N. J.
Doeskin
, and
J.
Kleist
,
1995
:
Drought monitoring with multiple time scales.
Preprints, Ninth Conf. on Applied Climatology, Dallas, TX, Amer. Meteor. Soc., 233–236
.
McVicar
,
T. R.
, and
D. L. B.
Jupp
,
1998
:
The current and potential operational uses of remote sensing to aid decisions on drought exceptional circumstances in Australia: A review.
Agric. Syst.
,
3
,
399
468
.
Mimikou
,
M.
, and
E.
Baltas
,
2002
:
Technical Hydrology.
2nd ed. (in Greek), National Technical University of Athens, 230 pp
.
Moran
,
M. S.
,
T. R.
Clarke
,
Y.
Inoue
, and
A.
Vidal
,
1994
:
Estimating crop water deficit using the relation between surface-air temperature and spectral vegetation index.
Remote Sens. Environ.
,
49
,
246
263
.
Nemani
,
R.
,
L. L.
Price
,
S. W.
Running
, and
S. N.
Goward
,
1993
:
Developing satellite derived estimates of surface moisture status.
J. Appl. Meteor.
,
32
,
548
557
.
Palmer
,
W. C.
,
1965
:
Meteorological drought.
U.S. Weather Bureau Research Paper 45, NOAA, 58 pp
.
Priestley
,
C. H. B.
, and
R. J.
Taylor
,
1972
:
On the assessment of surface heat flux and evaporation using large-scale parameters.
Mon. Wea. Rev.
,
100
,
81
92
.
Prihodko
,
L.
, and
S. N.
Goward
,
1997
:
Estimation of air temperature from remotely sensed surface observations.
Remote Sens. Environ.
,
60
,
335
346
.
Qin
,
Q.
,
A.
Ghulam
,
L.
Zhu
,
L.
Wang
,
J.
Li
, and
P.
Nan
,
2008
:
Evaluation of MODIS derived perpendicular drought index for estimation of surface dryness over northwestern China.
Int. J. Remote Sens.
,
7
,
1983
1995
.
Sandholt
,
I.
,
K.
Rasmussen
, and
J.
Andersen
,
2002
:
A simple interpretation of the surface temperature-vegetation index space for assessment of surface moisture status.
Remote Sens. Environ.
,
79
,
213
224
.
Seiler
,
R. A.
,
F.
Kogan
, and
J.
Sullivan
,
1998
:
AVHRR-based vegetation and temperature condition indices for drought detection in Argentina.
Adv. Space Res.
,
21
,
481
484
.
Senay
,
G. B.
,
M.
Budde
,
J. P.
Verdin
, and
A. M.
Melesse
,
2007
:
A coupled remote sensing and simplified surface energy balance approach to estimate actual evapotranspiration from irrigated fields.
Sensors
,
7
,
979
1000
.
Shuttleworth
,
W. J.
, and
J. S.
Wallace
,
1985
:
Evaporation from sparse crops—An energy combination theory.
Quart. J. Roy. Meteor. Soc.
,
111
,
839
855
.
Tittebrand
,
A.
,
A.
Schwieben
, and
F.
Berger
,
2005
:
The influence of land surface parameters on energy flux densities derived from remote sensing data.
Meteor. Z.
,
14
,
227
236
.
Twine
,
T. E.
, and
Coauthors
,
2000
:
Correcting eddy-covariance flux underestimates over a grassland.
Agric. For. Meteor.
,
103
,
279
300
.
Vanderlinden
,
K.
,
J. V.
Giraldez
, and
M.
Van Meirvenne
,
2004
:
Assessing reference evapotranspiration by the Hargreaves method in southern Spain.
J. Irrig. Drain. Eng.
,
130
,
184
191
.
Van Leeuwen
,
W.
,
A.
Huete
, and
T.
Laing
,
1999
:
MODIS vegetation index compositing approach: A prototype with AVHRR data.
Remote Sens. Environ.
,
69
,
264
280
.
Van-Rooy
,
M. P.
,
1965
:
A rainfall anomaly index independent of time and space.
Notos
,
14
,
43
48
.
Wang
,
D.
,
S.
Liang
,
R.
Liu
, and
T.
Zheng
,
2010
:
Estimation of daily-integrated PAR from sparse satellite observations: Comparison of temporal scaling methods.
Int. J. Remote Sens.
,
31
,
1661
1677
.
Wang
,
K.
, and
S.
Liang
,
2008
:
An improved method for estimating global evapotranspiration based on satellite determination of surface net radiation, vegetation index, temperature, and soil moisture.
J. Hydrometeor.
,
9
,
712
727
.
Wang
,
K.
, and
S.
Liang
,
2009
:
Estimation of daytime net radiation from shortwave radiation measurements and meteorological observations.
J. Meteor. Appl. Climatol.
,
48
,
634
643
.
Wang
,
K.
,
P.
Wang
,
Z.
Li
,
M.
Cribb
, and
M.
Sparrow
,
2007
:
A simple method to estimate actual evapotranspiration from a combination of net radiation, vegetation index, and temperature.
J. Geophys. Res.
,
112
,
D15107
.
doi:10.1029/2006JD008351
.

Footnotes

Corresponding author address: Yunjun Yao, Dept. of Geography, University of Maryland, College Park, College Park, MD 20742. Email: boyyunjun@163.com