COMPARISON OF EVAPOTRANSPIRATION IMAGES FROM …Reference ET Reference ET was calculated hourly using the 2005 ASCE-EWRI standardized Penman-Monteith method for the 0.5 m tall alfalfa

Pecora 17 – The Future of Land Imaging…Going Operational November 18 – 20, 2008 Denver, Colorado

COMPARISON OF EVAPOTRANSPIRATION IMAGES FROM MODIS AND LANDSAT ALONG THE MIDDLE RIO GRANDE OF NEW MEXICO

Richard G. Allen, Professor

Clarence W. Robison, Research Associate Ricardo Trezza, Visiting Professor Magali Garcia, Visting Professor

Jeppe Kjaersgaard, Asst. Professor University of Idaho – Kimberly Research and Extension Center

Kimberly, ID 83341 [email protected]

[email protected] [email protected] [email protected]

[email protected] ABSTRACT

Landsat images are highly preferred to images by MODIS for producing evapotranspiration (ET) maps for specific land use types because of their higher resolution (30 m). However, Landsat images are potentially available only each 16 days, and, with the expected failure of Landsat 5 sometime within the next five years and the questionable likelihood of a thermal sensor on Landsat 8 scheduled for launch in 2012, the prospect of images from Landsat that are useful for application with energy balance determination of evapotranspiration (ET) is less than bright. Therefore, more use of coarse resolution satellite images, such as 1 km thermal images from MODIS, will occur. One advantage of MODIS satellites is that images having view angle < ~15o are potentially available about each four to five days. Application of METRIC energy balance processes along the Middle Rio Grande of New Mexico using MODIS imagery indicates that one can successfully reproduce monthly and annual ET estimates that were obtained using Landsat imagery. However, spatial fidelity is highly degraded. This paper compares ET images for the Rio Grande region as produced by both MODIS and by Landsat.

BACKGROUND

The METRIC model (Mapping Evapotranspiration at high Resolution with Internalized Calibration) is a satellite-based image-processing model for calculating evapotranspiration (ET) as a residual of the surface energy balance. METRIC was developed by the University of Idaho (Allen et al., 2005a, 2007a,b, Tasumi et al., 2005a,b) for application to Landsat satellite imagery to maximize ET product resolution (30 m). METRIC has been converted to operate with coarser MODIS images (500 m) that have more frequent availability (Tasumi et al., 2006). METRIC uses as its foundation the pioneering SEBAL energy balance process developed in the Netherlands by Bastiaanssen (1995) and Bastiaanssen et al. (1998a,b), where near surface temperature gradients for estimating the sensible heat component of the surface energy balance are an indexed function of radiometric surface temperature, thereby eliminating the need for absolutely accurate surface temperature and the need for air temperature measurements. The surface energy balance is inversely and internally calibrated in METRIC using ground-based reference ET to reduce computational biases inherent to remote sensing based energy balance components and to provide congruency with traditional methods for ET. Slope and aspect functions and temperature lapsing are used in applications in mountainous terrain. The primary inputs to the METRIC model are short-wave and long-wave (thermal) images from satellite (e.g. Landsat and MODIS), a digital elevation model (DEM) and ground based weather data measured within or near the area of interest. ET “maps” (i.e., images) via METRIC provide the means to quantify ET on a field by field basis in terms of both the rate and spatial distribution.

METRIC has significant advantages over conventional methods for estimating ET from crop coefficient curves in that crop development stages do not need to be known with METRIC, nor does the specific crop type need to be known. In addition, the energy balance can detect reduced ET caused by water shortage. METRIC has significant advantages over many traditional applications of satellite-based energy balance in that its calibration is made using reference ET, rather than evaporative fraction. Use of reference ET for extrapolation of instantaneous ET to 24-hour


and longer periods compensates for regional advection effects that can cause ET to exceed daily net radiation in many arid or semi-arid locations (Allen et al., 2005b).

APPLICATION OF METRIC TO THE MIDDLE RIO GRANDE

The METRIC model was applied to Landsat imagery for 2002 for irrigated areas along the Middle Rio Grande

valley from near Algodones south to near San Acacia and to MODIS imagery for years 2002, 2005 and 2007 for the same region. The MODIS image dates in 2002 are exactly the same dates as those processed using Landsat imagery to provide opportunity for cross-comparisons. All MODIS image dates were selected to limit satellite sensor view angles to less than about 15o for the areas of interest, along with cloud free conditions for the MRG area. Thirteen Landsat and MODIS dates were processed for year 2002, spanning the period Jan. 14 to Nov. 7, twenty one MODIS dates were processed for year 2005, spanning the period Jan. 6 to Dec. 24, and twenty six MODIS dates were processed for year 2007, spanning the period Jan. 28 to Dec. 30. Specifics of the application of METRIC with MODIS imagery are given in Tasumi et al., (2006). The following sections briefly describe the calibration of METRIC to MODIS images, which is done uniquely for each image date. The calibration process for METRIC is detailed in Allen et al., (2007a) and involves the selection of ‘dry’ and ‘wet’ pixels and conditions where the sensible heat flux components of the energy balance are calibrated. Dry (hot) Pixel Selection

The dry, hot pixel for METRICMODIS calibration was selected from a 2.5 km by 2.5 km dry, flat desert area SE of Albuquerque shown in Figure 1. This location had relatively uniform vegetation and was selected to represent the condition where ET ~ 0 for periods without recent antecedent precipitation. This same location was used for all three years and was selected based on results from prior Landsat-based applications. The estimated ET for the hot pixel was set to a positive value for some dates according to antecedent precipitation using a daily surface soil water balance driven by weather data from the MRGCD weather stations, as described in a following section. The ET values assigned to the ‘hot pixel condition’ were adjusted for some MODIS images to force the METRIC process to produce ET values targeted for selected dry areas within the MRG valley.

Figure 1. Hot pixel location used to calibrate METRIC for all MODIS images. Wet (cold) Pixel Selection The wet (cold) calibration pixel of METRIC, when applied with Landsat imagery, is identified as a fully vegetated pixel having a near minimum surface temperature and where the assumption that its ET is approximately equal to 1.05 x alfalfa reference ETr can be made (Allen et al., 2005b, Tasumi et al., 2005a,b, Allen et al., 2007a). These types of pixels are relatively easy to identify with Landsat imagery where the 30 m resolution for short wave reflectance and 60 to 120 m resolution for surface temperature permits sampling from within individual fields. With MODIS, where resolution for short wave is 500 m and for thermal is 1000 m, it is nearly impossible to identify a single pixel having uniform, full vegetation cover and maximum ET, especially in the Middle Rio Grande


application where agricultural and riparian areas follow a relatively thin river corridor. Therefore, alternative methods were explored for calibrating METRIC to MODIS images for a wet condition.

Two methods were tested. (1) Determination of relative ET to assign to the ‘cold’ pixel based on the normalized difference vegetation index (NDVI), and (2) using the ‘SEBAL approach’ of Bastiaanssen et al. (1998a,b) where one assumes that the skin temperature of a nearby, and generally shallow (< 4 m deep), water body is the same temperature at which the sensible heat flux, H, to the air becomes zero.

Tasumi et al., (2006) developed a Kc vs NDVI relationship for the MRG area by integrating METRIC-based ETrF (ETrF represents the ‘fraction of reference ET’ and is synonymous with Kc) derived from Landsat images into MODIS scale data. The METRICLandsat-based ETrF products were degraded to 1 km and correlated to MODIS-based NDVI based on at surface reflectances. A relationship was produced for application to the higher end of NDVI values where:

ETrF = 1.513 NDVI – 0.145 (1)

2002. The ETrF-NDVI relationship was applied to the average NDVI of five 500 m sized pixels in each MODIS image having the highest NDVI value for the 2002 MODIS applications. The result was estimated ETrF associated with the surface temperatures associated with these five selected pixels.

2005. For 2005, the ETrF-NDVI relationship was applied to the average NDVI from the twenty 500 m pixels in the MODIS image having the highest NDVI values. The mean NDVI of these 20 pixels was used with the equation to estimate the ETrF associated with the mean surface temperature associated with these pixels. The equation was applied to all images. The ETrF average for the ‘cold’ pixel condition ranged from about 0.8 to 0.9, indicating a 15 to 25% reduction in ET relative to ET expected for a fully vegetated, fully watered condition. This decrease was due to the absence of MODIS thermal pixels, which are 1000 m on a side, that have full (dense and green) vegetation cover.

2007. For 2007, the cold pixel selection was assisted by use of a statistical method that selected the lower 20 to 40 % of surface temperatures from the population of pixels within the agricultural corridor of the MRG system that contained the 10% highest values for NDVI. As an independent check, the procedure of 2005 was also applied, where NDVI was averaged for the twenty 500 m pixels in the MODIS image having the highest NDVI values. This mean NDVI was then used with the ETrF-NDVI relationship to estimate the ETrF associated with the mean surface temperature associated with those selected pixels. The statistical approach has the desirable attribute of being nearly automatable. As in 2002 and 2005, the maximum ETrF average for the cold pixel condition ranged from about 0.8 to 0.9, indicating a 15 to 25% reduction in ET relative to ET expected for a fully vegetated, fully watered condition. The resulting ETrF assigned to the cold condition, relative to NDVI, and resulting calibrations was similar among years as shown in Fig. 2. ETrF/NDVI > 1.5 resulted from antecedent rainfall events. The ETrF/NDVI for the hot pixel calibration changed substantially with image and year due to effects of antecedent soil moisture.

Figure 2. Comparison of ETrF/NDVI for the wet (cold) calibration condition for application of METRICMODIS to the MRG region during years 2002, 2005 and 2007.

ETrF/NDVI at Cold Calibration Pixels - MRG: METRIC-MODIS

0.0

0.5

1.0

1.5

2.0

2.5

3.0

0 50 100 150 200 250 300 350 400DOY

ETrF

/NDV

I

2007 - Cold Pixels 2005 Cold-Pixels 2002 - Cold Pixels


Reference ET Reference ET was calculated hourly using the 2005 ASCE-EWRI standardized Penman-Monteith method for

the 0.5 m tall alfalfa reference basis, applied to automated weather stations operated by the Middle Rio Grande Conservancy District (MRGCD). Two weather stations, Angostura and Jarales, were selected due to their completeness of record, quality of data, and location within the corridor. A strict quality control analysis based on ASCE-EWRI (2005) was applied to all weather data and corrections or adjustments to data were made. Hourly ETr was used in the calibration of METRIC to each satellite image date and was summed to 24-hour totals during computation of daily, monthly and annual ET grids. Locations of the Angostura and Jarales MRGCD stations are shown in Fig. 3. The ETr from the two stations was averaged prior to usage for calibration. Water Balance for the Hot Pixel

A soil water balance was used to estimate the soil water content for the hot pixel on the image dates. The soil water balance was based on the FAO-56 procedure (Allen et al, 1998). Precipitation data was obtained from the Jarales weather station because of its proximity to the hot pixel. Results for 2007 are shown in Fig. 4.

Figure 3. “False color” Landsat images for Paths 34 and 35 of Landsat showing locations of weather stations along the Middle Rio Grande Vegetated areas along the MRG river show as the bright red vertical strip in the image center.

Albuquerque Intl. Airport

Santa Fe Airport

Espanola NWS293031

San Acacia

Boys Ranch

JaralesLosLunas

AlbuquerqueCandelaria Farms

Angostura

Pena Blaca

Jemez

Alcalde

Path 34

Path 35


2007 Water balance for bare soil. Precipitation from Jarales

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1/9/

07

1/30

/07

2/20

/07

3/13

/07

4/3/

07

4/24

/07

5/15

/07

6/5/

07

6/26

/07

7/17

/07

8/7/

07

8/28

/07

9/18

/07

10/9

/07

10/3

0/07

ETrF

bas

ed o

n ET

r0.00

6.00

12.00

18.00

24.00

30.00

36.00

42.00

Prec

ipita

tion

(mm

)

Figure 4. Soil water balance for bare soil. Precipitation data from Jarales, 2007.

MODIS products used in METRIC applications

Table 1 lists the MODIS (Terra) products that were used in METRIC applications. MODIS radiances, geolocation fields, and cloud mask products were obtained from the NASA Level 1 and Atmosphere Archive and Distribution System (LAADS) web site. Land surface temperature (LST) version 5, used for 2005 and 2007, was obtained from the Earth Observing System Gateway. Version 4 LST was used for 2002.

Table 1. MODIS Terra products used for the 2005 METRIC application

MODIS product name Product Version

Product Description Obtained from:

MOD02HKM 5 Calibrated Radiances http://ladsweb.nascom.nasa.gov

MOD03 5 Geolocation Fields http://ladsweb.nascom.nasa.gov

MOD11_L2 5 Land Surface Temperature

Provided by Zheming Wang (MODIS LST Team Leader).

MOD35 5 Cloud mask http://ladsweb.nascom.nasa.gov

Application of METRIC

METRICMODIS was applied with unique calibration to each image. In some desert areas, where surface temperature was sometimes much larger than the hot pixel temperature due to shielding of the soil surface by dry brush, the value for dT was constrained to that for the hot pixel. In addition, soil heat flux (G) under these conditions was reduced to account for the fact that in a dry and hot desert soil, thermal conductance is often reduced by the crusting of the soil and lower density and associated particle to particle contact (Allen et al. 2007a). A number of mountainous areas were cloudy in many images. Fortunately, the Rio Grande corridor was clear in most of images, which facilitated interpolation of daily to monthly and seasonal ET.

A pixel located in the agricultural area shown on Fig. 5 was selected to illustrate the evolution of ETrF during 2007 as shown in Fig. 6. Many of the ETrF values corresponded to the general ETrF vs. NDVI relation (black dots). Three gray points in Fig. 6 have high values of ETrF due to significant precipitation events; thus these dates do not follow the general ETrF-NDVI trend.


Figure 7 shows the variation of ETrF in a desert area from January to September 2007. Negative values for

ETrF reflect uncertainties in the estimation of near-zero ET as a residual of the surface energy balance. These negative values probably represent nearly zero evapotranspiration. ETrF rose above about 0.10 during July and August, reflecting ET and vegetation growth stimulated by the summer monsoon.

Figure 5. Location of an agricultural pixel used to plot ETrF evolution in Figure 6. Left: MODIS image and right: Landsat 7 image used as a reference.

ETrf FOR A PIXEL IN A VEGETATED AREA

0

0.2

0.4

0.6

0.8

1

1.2

0 50 100 150 200 250 300 350

DAY OF THE YEAR

ETR

f

0

40

80

120

160

PREC

IPITATIO

N (m

m)

Figure 6. Variation of ETrF, obtained from METRIC, from Jan. to Sept. 2007 (black symbols). High values of ETrF from precipitation events are shown in gray color.

COMPARISON WITH LANDSAT APPLICATIONS

ET images derived from METRICMODIS were compared with those derived from METRICLandsat for year 2002. In Fig. 8, METRICMODIS results, averaged over the MRG valley (from approximately Algodones south to approximately San Acacia), were close to those for METRICLandsat. Annual ET over the valley was estimated at 1045 mm by the METRICMODIS application and 1067 mm by the METRICLandsat application for 2002. The difference was 2%. Other investigators also report close agreement between ET maps generated using SEBAL with


MODIS and using SEBAL with Landsat images in the Middle Rio Grande Valley (Hong et al., 2005) and in the White Volta Basin (Compaoré et al., 2007).

TYPICAL ETrf FOR DESERT AREAS

-0.2

0

0.2

0.4

0.6

0.8

0 50 100 150 200 250 300 350

DAY OF THE YEAR

ETrF

0

40

80

120

160

PREC

IPITATIO

N (m

m)

Figure 7. Variation of ETrF, obtained from METRIC, from January to September 2007, obtained by averaging several pixels in a desert area.

ET produced by the METRICMODIS method performed better (relative to the METRICLandsat application) than did a test application of the SEBAL calibration method applied to MODIS images, where the temperature for the ‘wet’ pixel condition was taken as water temperature (in this case the temperature of Elephant Butte Reservoir). In the METRIC applications, hourly alfalfa reference ETr is used to calibrate the energy balance process. In the SEBAL method, sensible heat flux (H) is assumed to be zero for at water surface temperature. The ‘hot pixel’ for calibrating METRIC and SEBAL was the same. The MODIS-SEBAL calibration approach and application tended to underestimate for the six dates processed. Comparisons for Specific Locations

Fig. 9 shows Landsat and MODIS based ETrF for selected pixels having high NDVI and low NDVI, analyzed by sampling over 2 km areas. The graph provides an idea of general values for the ETrF for the cold/hot pixel selections. At locations having relatively high NDVI, MODIS ETrF was close to that from Landsat. Fig. 9 shows MODIS ETrF for image no. 9 (8/10/2002) to be underestimated. Otherwise, comparisons were quite good. Agreement was also very good for the locations having low NDVI. ETrF was high for the first two images due to antecedent rainfall. Impact of MODIS Resolution

Fig. 10 is similar to Fig. 9, but where the analysis was conducted by sampling 500 m pixels, rather than aggregated to 2 km sized areas. At the 500 m scale, METRICMODIS consistently underestimated ETrF at high-NDVI locations relative to MODISLandsat because the MODIS thermal band has 1 km resolution that is compromised by pixels neighboring a specific 500 m pixel.


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1 2 3 4 5 6 7 8 9 10 11 13

Image number

Kc

(MR

G V

alle

y A

vg.)

MODIS (METRIC)Landsat (METRIC)MODIS (SEBAL)

Figure 8. Average Kc (i.e., ETrF) for the MRG valley, calculated by METRICMODIS, METRICLandsat and MODIS-SEBAL for 12 images during 2002. Image no.13 was not included due to many small clouds over MRG. The Kc is based on the ASCE-EWRI (2005) standardized Penman-Monteith alfalfa reference equation.

00.10.20.30.40.50.60.70.80.9

1

1 2 3 4 5 6 7 8 9 10 11 13

Image number

Kc

(MR

G V

alle

y)

High Vege (MODIS) High Vege (Landsat) Low Vege (MODIS) Low Vege (Landsat)

Figure 9. MODIS and Landsat based average ETrF (i.e,. Kc for locations having high NDVI (0.50 or higher) and low NDVI (0.20 and lower) along the Middle Rio Grande valley corridor during 2002, sampled with 2 km resolution.

No significant differences between METRICMODIS and METRICLandsat were found for low-NDVI locations at the 500 m scale. This was most likely due to the presence of relatively large clusters of 500 m pixels having similar vegetation and land cover and surface temperature, so that the surface temperature value was not compromised by different surface temperatures of neighboring pixels.


00.10.20.30.40.50.60.70.80.9

1

1 2 3 4 5 6 7 8 9 10 11 13

Image number

Kc

(MR

G V

alle

y)

High Vege (MODIS) High Vege (Landsat) Low Vege (MODIS) Low Vege (Landsat)

Figure 10. MODIS and Landsat based average Kc for high NDVI (0.50 or higher) and low NDVI (0.20 and lower) pixels in MRG, analyzed at 500 m resolution. Aggregation to the ET-Toolbox Grid

The ET produced from MODIS was aggregated into 4 km sized grid cells of the ET-Toolbox grid managed by the U.S. Bureau of Reclamation and MRGCD as part of their river management programs (www.usbr.gov/pmts/rivers/awards/ettoolbox.pdf). An example of monthly ET and ETrF computed by interpolating between satellite image dates is shown in Fig. 11.

SUMMARY

In summary, the modifications to METRIC to utilize MODIS imagery were successful and useful ‘maps’ of ET and ETrF (i.e., Kc) were created on a daily, monthly and annual basis. The application of METRIC with MODIS is challenging due to the large number of quality issues associated with MODIS products and the coarse grid size of thermal data (1 km).


Figure 11. Monthly ET and Kc (i.e., ETrF) aggregated to 4 km sized grid cells of the ET-Toolbox grid along the

Middle Rio Grande during 2007.


REFERENCES Allen, R.G., Pereira, L.S., Raes, D. and Smith, M., 1998. Crop Evapotranspiration: Guidelines for Computing Crop

Water Requirements. United Nations FAO, Irrigation and Drainage Paper 56. 1998. Rome, Italy. 300 p. http://www.fao.org/docrep/X0490E/X0490E00.htm (accessed Feb. 5, 2007)

Allen, R.G., Tasumi, M, Morse, A.T. and Trezza, R., 2005a. A Landsat-based Energy Balance and Evapotranspiration Model in Western US Water Rights Regulation and Planning. J. Irrigation and Drainage Systems. 19:251-268.

Allen, R.G., Tasumi, M. and Trezza, R., 2005b. Benefits from Tying Satellite-based Energy Balance to Reference Evapotranspiration. Earth Observation for Vegetation Monitoring and Water Management. (D’Urso, G., A. Jochum, and J.Moreno (ed.)). Amer. Instit. Physics; ISBN-10: 0735403465. p. 127-137.

Allen, R.G., Tasumi, M. and Trezza, R., 2007a. Satellite-based energy balance for mapping evapotranspiration with internalized calibration (METRIC) – Model. ASCE J. Irrigation and Drainage Engineering. 133(4): 380-394

Allen, R.G., Tasumi, M. Morse, A.T., Trezza, R., Kramber, W., Lorite I. and Robison, C.W., 2007b. Satellite-based energy balance for mapping evapotranspiration with internalized calibration (METRIC) – Applications. ASCE J. Irrigation and Drainage Engineering. 133(4): 395-406

ASCE – EWRI, 2005. The ASCE Standardized reference evapotranspiration equation. ASCE-EWRI Standardization of Reference Evapotranspiration Task Comm. Report, ASCE, ISBN 078440805, Stock Number 40805, 216 pages.

Bastiaanssen, W.G.M., 1995. Regionalization of surface flux densities and moisture indicators in composite terrain: A remote sensing approach under clear skies in Mediterranean climates. Ph.D. Diss., CIP Data Koninklijke Bibliotheek, Den Haag, the Netherlands. 273 p.

Bastiaanssen, W.G.M., Menenti, M., Feddes, R.A. and Holtslag, A.A.M., 1998a. A remote sensing surface energy balance algorithm for land (SEBAL): 1. Formulation. J. Hydrology, 212-213, p. 198-212.

Bastiaanssen,W.G.M., Pelgrum, H., Wang, J., Ma, Y., Moreno, J., Roerink G.J. and van der Wal, T., 1998b. The Surface Energy Balance Algorithm for Land (SEBAL): Part 2 validation, J. Hydrology , 212-213: 213-229

Compaoré, H., J.M.H. Hendrickx, S.-h. Hong, J. Friesen, N.C. van de Giesen, C. Rodgers, J. Szarzynski, and P.L.G. Vlek, 2007. Evaporation mapping at two scales using optical imagery in the White Volta Basin, Upper East Ghana. Physics and Chemistry of the Earth, Parts A/B/C In Press:doi:10.1016/j.pce.2007.04.021.

Hong, S.-h., J.M.H. Hendrickx, and B. Borchers, 2005. Effect of scaling transfer between evapotranspiration maps derived from LandSat 7 and MODIS images. Proc. Int. Society for Optical Engineering, SPIE 5811:147-158.

Tasumi, M., Trezza, R. and Allen, R.G., 2006. METRIC Manual for MODIS Image Processing Version 1.0 (April, 2006). University of Idaho R&E Center, Kimberly, Idaho. 67 p.

Tasumi, M., Allen, R. G., Trezza, R., and Wright, J. L., 2005a. Satellite-based energy balance to assess within-population variance of crop coefficient curves, J. Irrig. and Drain. Engrg, ASCE 131(1):94-109.

Tasumi, M., Trezza, R., Allen, R.G. and Wright, J. L., 2005b. Operational aspects of satellite-based energy balance models for irrigated crops in the semi-arid U.S. J. Irrigation and Drainage Systems. 19:355-376.

Documents

COMPARISON OF EVAPOTRANSPIRATION IMAGES FROM …Reference ET Reference ET was calculated hourly using the 2005 ASCE-EWRI standardized Penman-Monteith method for the 0.5 m tall alfalfa