Improving the estimation of land surface temperature for ... - iacm-forth

Improving the estimation of land surface temperature for the region of Greece: adjustment of a split window algorithm to account for the distribution of precipitable water

N. Chrysoulakis and C. Cartalis University of Athens, Dept. of Applied Physics, Building PHYS-V, Athens, 15784, Greece

Abstract. Land Surface Temperature (LST) may be estimated using thermal infrared data as acquired by the Advanced Very High Resolution Radiometer (AVHRR) onboard the NOAA satellites. For LST to be estimated, algorithms based on the "Split Window" method are applied, their performance depending greatly on the accurate definition of the precipitable water (PW) in the area of the satellite image. In this study, the precipitable water in the area of Greece as calculated from radiosonde data (00:00 UTC and 12:00 UTC) for a period of 29 years (1967-1995) for the Hellenikon Station (37o 54΄ N, 23o 44΄ E) is used to improve the performance of a split window algorithm for LST. The "PW adjusted" algorithm is applied in a series of NOAA/AVHRR images in order to estimate LST for South and Central Greece. A case study is presented, where results are compared with in situ temperature measurements from surface stations of the National Meteorological Service to demonstrate a very good agreement.

Land surface temperature estimation using AVHRR and radiosonde data

1. Introduction Land Surface Temperature (LST) may be estimated using data in the thermal infrared from the Advanced Very High Resolution Radiometer (AVHRR) on board the NOAA satellites. Prior to any estimation, thermal infrared data need to be corrected in order to compensate for: a) the atmospheric absorption resulting from the presence of water vapour in the atmospheric window 10,5 - 12,3 µm, b) the land surface emission coefficient, which varies depending the emitting surface and the wavelength (Casseles et al. 1997). The error in estimating LST, if the spectral dependence of the emission coefficient is not taken under consideration, may reach 3° K (Ottle and Stoll 1993). Estimating LST through AVHRR thermal infrared data, is accomplished using the "Split Window" method on channels 4 (10.5 - 11.3 µm) and 5 (11,5 - 12,5 µm), at the precondition that the variability of the emission coefficients is taken into account (Price 1984, Becker 1987, Becker and Li 1990a, Vidal 1991, Sobrino et al. 1991, Kerr et al. 1992, Ottle and Vidal Madjar 1992, Coll et al. 1994a, Prata 1993, Prata et al. 1995, Casselles et al. 1997). This method has been also proven successful for the estimation of sea surface temperature (McClain et al. 1985, Barton et al. 1989, Barton 1992). The algorithm selected in this study for the estimation of LST using AVHRR thermal infrared data, is a global "Split Window" algorithm developed by Coll et al. (1994a). It requires the brightness temperatures in AVHRR channels 4 and 5, the mean emissivities and the spectral emissivity difference in these channels. It also uses coefficients which depend on atmospheric moisture and the surface temperature.

These coefficients can be optimised according to the characteristics of a given area. The algorithm is referred to as the University of Valencia Model (UVM) and it is used by Caselles et al. (1997) for the Hydrologic Atmospheric Pilot Experiment data processing in the Sahel (HAPEXX-Sahel); its accuracy is ± 2 ° K. The algorithm is described by the relation: LST = T4 + [1 + 0.58(T4 - T5)] (T4 - T5) + 0.51 + α(1 - E) - β(∆ε) (1) where T4

is the radiance temperature for channel 4 of AVHRR,

T5

is the radiance temperature for channel 5 of AVHRR,

Ε

is the mean spectral emission coefficient for channels 4 and 5: E = (ε4 + ε5)/2 where: ε4 is the surface emission coefficient for channel 4, ε5 is the surface emission coefficient for channel 5,

and ∆ε is the difference of the emission coefficients for channels 4 and 5: ∆ε = ε4 - ε5 Coefficients α and β in equation (1) depend on the amount of atmospheric water vapour in the area of the satellite image and from the temperature of the surface under observation. They may be described as a function of the brightness temperature (T4) which is recorded in channel 4 of the AVHRR and the precipitable water (PW) in the area (Caselles et al. 1997).

In particular, PW is the total amount of water vapour in a vertical direction between earth surface (or a surface in a given height) and the top of the atmosphere:

0

1 PW ° ³ M r dp g ps

(2)

where: Mr

is the mixing ratio,

g

is the acceleration of gravity,

p

is the pressure,

ps

is the surface pressure

Estimating the values of PW requires knowledge of the vertical distribution of atmospheric water vapour and therefore requires the use of radiosonde data. Consequently, defining the values of coefficients α and β for a given area is of considerable difficulty, a fact which is reflected in equation (1) regarding the estimation of surface temperature of this area. In this study, an effort is made to assess the typical monthly values of coefficients α and β for Greece, using a series of radiosonde data covering 29 years. From the resulting values for α and β, the algorithm expressed by equation (1) is applied on several NOAA/AVHRR images for the area of Greece.

2. Data and Methodology Due to the importance of PW in both the estimation of the energy balance and the atmospheric correction of satellite data, a number of researchers have attempted, in the past, to estimate its distribution in Greece (Asteriadis et al. 1992, Chrysoulakis et al. 1996, Cartalis and Chrysoulakis 1997), as well as to correlate it with the true rainfall height (Karakostas and Sioutas 1992). Daily values of PW were estimated on the basis of equation (2) with the use of radiosonde measurements of relative humidity and temperature from the surface to 500 hPa, at the National Meteorological Service at Hellenikon Station (37,54° N, 23,44° E) for the period 1967 – 1995. Following to the estimation of the daily values of PW, monthly mean values for the 29 year period, as well as standard deviations were calculated. It should be mentioned that for the estimation of the integral in equation (2) (integral limits in this case are from p = ps to p = 500 hPa), the method of dividing the atmospheric column in 11 different layers and summing the values of each layer was applied (Cartalis and Chrysoulakis 1997). Monthly mean values and standard deviations of PW for the 29 year period were further used for the calculation of the typical monthly values of coefficients α and β, which according to Caselles et al. (1997), are given by equations (3) and (4) below: α = (0,190 PW - 0.103)T4 - 67 PW + 107

(3)

β = (0,100 PW + 1,118)T4 - 68 PW - 163

(4)

where PW is in gr/cm2 and T4 in oK.

3. Results Tables 1 and 2 provide the monthly means and standard deviations of PW for the entire 29 years period, for the 00.00 and 12.00 GMT radiosondes respectively, as well as the detailed equations for coefficients α and β. These coefficients are given as a function of the brightness temperature recorded in channel 4 of AVHRR (T4) for every pixel. In Figures 1 to 4 the distribution of α and β for every month for night time and daytime radiosondes, for a typical T4 range (280o K - 330o K), is presented. Consequently, if the spectral emission coefficients ε4 and ε5 for channels 4 and 5 AVHRR are known for each pixel of the satellite image, equation (1) can be used either in combination with Table 1 or the corresponding Figures (1 and 2) to estimate LST for night time images, or in combination with Table 2 or the corresponding Figures (3 and 4) to estimate LST for daytime images. In order to estimate the LST sensitivity to PW, the distribution of the difference LST T4, for a typical T4 range (280° K - 330° K) was examined. Parameters ε and ∆ε were expressed using mean typical values of ε = 0,975 and ∆ε = - 0,005 (Caselles et al., 1997). The PW range (0.3 – 4.4 gr/cm2) has been determined using the minimum and maximum values of the PW data-set. Figures 5 to 8 were developed so as to demonstrate the relationship among LST - T4, PW and T4 - T5. In particular, these Figures present the distribution of the difference LST - T4 versus PW, for T4 - T5 values of 0.5° K, 1° K, 1.5° K and 2° K, respectively (where T5 is the brightness temperature recorded in channel 5 of AVHRR). The linear relationship between LSTT4 and PW is evident in all Figures. The Figures also yield that as PW increases, a linear reduction of the LST - T4 value is observed. Furthermore, the slope of this

linear distribution decreases with T4, which implies that the lower the T4, the stronger the dependence of LST- T4 on PW. For T4 = 280° K the decrement in LST - T4 is 2.23° K, as the PW increases from 0.3 to 4.4 gr/cm2, whereas for T4 = 330° K the decrement in LST - T4 is 1.16° K, as the PW increases from 0.3 to 4.4 gr/cm2. The result is that the underestimation in LST due to the atmospheric precipitable water is between 1.16 and 2.23° K. A bias in LST - T4 values is also evident in Figures 5 - 8. An increase of 0.5° K in T4 - T5 results in a bias of 1.23° K in LST - T4 values.

4. Case Study: Application on the broader region of South and Central Greece Coefficients α and β given in Tables 1 and 2 were used to estimate LST for the study area using equation (1). Due to lack of in situ measurements for the spectral emission coefficients ε4 an ε5 of AVHRR channels 4 and 5 for the given area, mean typical values were used to express parameters ε and ∆ε (Caselles et al., 1997): ε = 0,975 and ∆ε = - 0,005 A number of high resolution satellite images acquired by the AVHRR of NOAA (Local Area Coverage) were used. These images have a NOAA Level 1-b format and are structured in 10 bits (1024 grey levels) words. For demonstrational reasons, results relating to one image will be demonstrated (image from NOAA 14 on the 7th of March 1998 at 01:44 UTC). Consequently for the above date of reception and on the basis of Table 2, equation (1) is given as: LST = - 1,770 + 1,014 T4 + (T4 - T5) + 0,580 (T4 - T5)2

(5)

Initially the image was geometrically corrected using certain Geographical Control Points (Ground Control Points) which were given in a level 1-b format. The

Mercatoric system is selected for the map projection of the image with central meridian on 23,5o and latitude 37,5o. The spheroid model used for the projection is the International 1909 with European Datum, RE 50. For the estimation of brightness temperatures for channels 4 and 5 of the AVHRR, the procedure described in IGBP-DIS (1994) was followed. According to this procedure the initial digital values recorded by the satellite receiver for every pixel were first transformed in spectral radiances and after in brightness temperature values T4 and T5 for every pixel. LST values for the region of South and Central Greece were calculated following the application of equation (5). Temperature values estimated using the satellite image were compared to the temperature values from eight in situ observations from the National Meteorological Service Stations (March 7, 1998, 01:50 UTC). Table 3 shows the coordinates of the surface stations which provided available (same date and time) observations. It also shows the temperature values from in situ observations for each station and from the processing of the satellite data (NOAA temperature) as well as their differences (∆Τ), the latter reflecting the accuracy of the applied methodology. An examination of the values of ∆Τs’ shows a close agreement between the temperature values estimated using equation (5) and in situ observations. In particular, ∆Τ obtains the mean value of 0,75° K while only in one case it exceeds 1° K. Figure 9 shows the distribution of LST for the region of South and Central Greece following the application of equation (5), as well as the positions of the ground stations. Pixels that correspond to the coordinates of every station were referred on the satellite image.

5. Conclusion The adjustment of a split window algorithm to account for the value of PW, allows the use of the algorithm for the area of Greece for each month, for daytime or night time NOAA AVHRR images in the thermal infrared. Lack of information on PW, would require radiosondes for the provision of the site specific values of relative humidity and temperature. With the use of the monthly PW values, as resulting from the period 1967-1995, the algorithm depends mostly on the emission coefficients of the surface. A linear relationship between LST- T4 and PW was observed. The lower the T4, the stronger the dependence of LST- T4 on PW. A bias in LST - T4 values was also observed. An increase of 0.5° K in T4 - T5 results in a bias of 1.23° K in LST - T4 values. The application of the adjusted algorithm on a number of AVHRR images for the estimation of the Land Surface Temperatures for the region of South and Central Greece, was proven successful. Temperature values estimated from the AVHRR data in the thermal infrared are in close agreement with the respective temperature values of the National Meteorological Service. However the statistical sample of the ground stations is considered modest (eight stations) to express with certainty the capacity of the algorithm to give good results for the entire area under study. Finally for the purposes of this study, standard values for the surface emission coefficients were used, a fact which is considered an inherent difficulty of the algorithm in terms of its accuracy.

References

ASTERIADIS, G., CONTADAKIS, M. E, KALKITSIS, X. I., and TSIOUMIS, A. K., 1992, Estimation of Precipitable Water over Greek Area, using ground Meteorological Observations. In Proceedings of 1st National Conference on Meteorology, Climatology and Atmospheric Physics, 21 - 23 May, Aristotelean

University

of

Thessaloniki,

(Thessaloniki,

Greece:

Aristotelean University of Thessaloniki), pp. 487 - 494. BARTON, I. J., 1992, Satellite-derived sea surface temperatures - A comparison between operational, theoretical and experimental algorithms. Journal of Applied Meteorology, 31, 432 - 442. BARTON, J. I., ZAVODY, M. A., O’ BRIEN, M. D., CUTTEN, R. D., SAUNDERS, W. R., LLEWELLYN - JONES, T. D., 1989, Theoretical Algorithms for satellite - derived sea surface temperature. Journal of Geophysical Research, 94, 3365 - 3375. BECKER, F., 1987, The impact of spectral emissivity on the measurement of land surface temperature from a satellite. International Journal of Remote Sensing, 8, 1509 - 1522. BECKER, F., and Z.-L., LI, 1990a, Towards a local split window method over land surfaces. International Journal of Remote Sensing, 11, 369 - 394. CARTALIS, C., and N. CHRYSOULAKIS, 1997, Estimation of precipitable water in Greece on the basis of radiosondes and satellite data. Toxicological and Environmental Chemistry, 58, 163-171.

CASELLES, V., COLL, C., and E. VALOR, 1997, Land surface emissivity and

temperature determination in the whole HAPEX - Sahel area from AVHRR data. International Journal of Remote Sensing, 18, No 5, 1009 1027. CHRYSOULAKIS, N., PROEDROU M., and C. CARTALIS, 1996, Estimation of Precipitable Water fluctuations over Athens area. In Proceedings of 3rd National Conference on Meteorology, Climatology and Atmospheric Physics, 25 – 27 September, National Observatory of Athens, (Athens, Greece: National Observatory of Athens), pp. 165 - 170. COLL, C., CASELLES, V., SORBINO, J. A., and E., VALOR, 1994a, On the atmospheric dependence of the split-window equation for land surface temperature. International Journal of Remote Sensing, 15, 105 - 122. IGBP-DIS, 1994, Satellite Fire Detection Algorithm. Workshop Technical Report, (Chris Justice, NASA/GSFC, USA / Pete Dowty, University of Virginia, USA - April 1994). KARAKOSTAS, T. S. and M. B. SIOUTAS, 1992, Correlation of Precipitable Water and Convective Thunderstorms Precipitation height in central Macedonian area. In Proceedings of 1st National Conference on Meteorology, Climatology and Atmospheric Physics, 21 - 23 May, Aristotelean University of Thessaloniki, (Thessaloniki, Greece: Aristotelean University of Thessaloniki), pp. 503 – 507. KERR, Y. H., LAGOUARDE, J. P., and J., IMBERNON, 1992, Accurate land surface temperature retrieval from AVHRR data with the use of an improved splitwindow algorithm. Remote Sensing of Environment, 41, 197 - 209.

McCLAIN, E. P., PICHEL, W. G., and C. C., WALTON, 1985, Comparative performance of AVHRR-based multichannel sea surface temperatures. Journal of Geophysical Research, C6, 11587 - 11601. OTTLE, C., and D., VIDAL-MADJAR, 1992, Estimation of land surface temperature with NOAA -9 data. Remote Sensing of Environment, 40, 27 - 41. OTTLE, C. and M. STOLL, 1993, Effect of atmospheric absorption and surface emissivity on the determination of land surface temperature from infrared satellite data. International Journal of Remote Sensing, 14, 2025 - 2037. PRATA, A., CASELLES, V., COLL, C., OTTLE, C., and J., SORBINO, 1995, Thermal remote sensing of land surface temperature from satellites: current status and future prospects. Remote Sensing Reviews, 12, 175 - 224. PRATA, A., 1993, Land Surface Temperatures Derived From the Advanced Very High Resolution Radiometer and the Along-Track Scanning Radiometer 1. Theory. Journal of Geophysical Research, 98, 16689 - 16702. PRICE, J. C., 1984, Land surface temperature measurements from the split window channels of the NOAA -7 AVHRR. Journal of Geophysical Research, 89, 7231-7237. SOBRINO, J. A.., COLL, C., and V., CASELLES, 1991, Atmospheric correction for land surface temperature using NOAA -11 AVHRR channels 4 and 5. Remote Sensing of Environment, 38, 19 - 34. VIDAL, A., 1991, Atmospheric and emissivity correction of land surface temperature measured from satellite using ground measurements or satellite data. International Journal of Remote Sensing, 12, 2449 - 2460.

Captions

Table 1. Monthly means () and standard deviations (σ) of PW and monthly values of coefficients α and β for the 29 years period, for the 00:00 UTC radiosondes.

Table 2. Monthly means () and standard deviations (σ) of PW and monthly values of coefficients α and β for the 29 years period, for the 12:00 UTC radiosondes.

Table 3. Geographical coordinates of National Meteorological Station surface stations from which temperature observations are taken. Temperature values from in situ observations for each station (in situ temperature), temperature values resulting from the processing of the satellite data (NOAA Temperature) and their difference (∆Τ).

Figure 1. Distribution of coefficient α for each month, for a typical T4 range (280o K 330o K). Curves were developed using the values from Table 1 for the 00:00 UTC radiosondes.

Figure 2. Distribution of β coefficient for every month, for a typical T4 range (280o K - 330o K). Curves were developed using the values from Table 1 for the 00:00 UTC radiosondes.

Figure 3. Distribution of α coefficient for every month, for a typical T4 range (280o K - 330o K). Curves were developed using the values from Table 2 for the 12:00 UTC radiosondes.

Figure 4. Distribution of β coefficient for every month, for a typical T4 range (280o K - 330o K). Curves were developed using the values from Table 2 for the 12:00 UTC radiosondes.

Figure 5. Distribution of the difference LST – T4 with PW, for a typical T4 range (280o K - 330o K), for T4 –T5 =0.5° K. The lower the T4, the stronger the dependence of LST- T4 on PW.

Figure 6. Distribution of the difference LST – T4 with PW, for a typical T4 range (280o K - 330o K), for T4 –T5 =1° K. The lower the T4, the stronger the dependence of LST- T4 on PW.

Figure 7. Distribution of the difference LST – T4 with PW, for a typical T4 range (280o K - 330o K), for T4 –T5 =1.5° K. The lower the T4, the stronger the dependence of LST- T4 on PW.

Figure 8. Distribution of the difference LST – T4 with PW, for a typical T4 range (280o K - 330o K), for T4 –T5 =2° K. The lower the T4, the stronger the dependence of LST- T4 on PW.

Figure 9. Distribution of Land Surface Temperature for the region of South and Central Greece (7 March 1998, 01:44 UTC). Numbers are shown the position of the Meteorological Stations used in this study.

00.00 UTC

(gr/cm2)

σ

α ( ° K)

β ( ° K)

JAN

2,161 2,398 2,233 1,842 1,924 1,818 1,723 1,643 2,025 2,067 2,045 2,210

0,790 0,775 0,668 0,738 0,631 0,630 0,649 0,652 0,651 0,730 0,661 2,247

0,308 T4 - 37,757 0,353 T4 -53,662 0,321 T4 -42,642 0,247 T4 -16,422 0,263 T4 -21,926 0,242 T4 -14,830 0,224 T4 -14,830 0,209 T4 -3,0709 0,282 T4 -28,644 0,290 T4 -31,519 0,286 T4 -29,999 0,317 T4 -41,043

1,334 T4 - 309,918 1,358 T4 - 326,060 1,341 T4 - 314,875 1,302 T4 - 288,264 1,310 T4 - 293,850 1,300 T4 - 286,648 1,290T4 - 280,152 1,282 T4 - 274,714 1,320 T4 - 300,669 1,325 T4 - 303,586 1,322 T4 - 302,044 1,339 T4 - 313,253

FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

12.00 UTC

(gr/cm2)

σ

α ( ° K)

β ( ° K)

JAN

1,802

FEB

2,173

MAR

2,055

APR

1,809

MAY

1,832

JUN

1,845

JUL

1,669

AUG

2,004

SEP

2,022

OCT

2,013

NOV

1,944

0,591 0,731 0,602 0,646 0,629 0,672 0,564 0,979 0,711 0,691 0,691

DEC

2,054

0,712

0,239 T4 -13,756 0,310 T4 - 38,590 0,287 T4-30,658 0,241 T4-14,215 0,245 T4 -15,767 0,248 T4 -16,638 0,214 T4 - 4,838 0,278 T4 - 27,243 0,281 T4 - 28,479 0,280 T4 - 27,902 0,266 T4 - 23,220 0,287 T4 - 30,622

1,298 T4-285,558 1,335 T4-310,763 1,323 T4-302,712 1,299 T4-286,024 1,301 T4-287,599 1,303 T4-288,484 1,285 T4-276,507 1,318 T4-299,247 1,320 T4-300,501 1,319 T4-299,916 1,312 T4-295,164 1,323 T4-302,676

In Situ NOAA Latitude Longitude Temperature Temperature ∆Τ (° K) (° K) (° K) 283 282 1 1 ATHENS/HELLENIKON 37° 44' Ν 23° 44' Ε 36° 48' Ν 27° 04' Ε 285 284 1 2 KOS AIRPORT 37° 42' Ν 26° 54' Ε 282 282 0 3 SAMOS AIRPORT 36° 24' Ν 25° 29' Ε 287 286 1 4 SANTORINI ISLAND 38° 08' Ν 21° 25' Ε 286 284 2 5 ARAXOS 38° 19' Ν 23° 33' Ε 277 276 1 6 TANAGRA 38° 54' Ν 24° 33' Ε 281 281 0 7 SKIROS ISLAND 39° 03' Ν 26° 36' Ε 285 285 0 8 MITILINI Ground Station

00.00 UTC 65

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

α ( ° Κ)

60

55

50

45 280

285

290

295

300

305

310

315

Brightness Temperature in AVHRR Channel 4 ( ° K)

320

325

330

00.00 UTC 150 140 130

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

120

β ( ° Κ)

110 100 90 80 70 60 50 280

285

290

295

300

305

310

315

Brightness Temperature in AVHRR Channel 4 ( ° K)

320

325

330

12.00 UTC 66 64 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

62

α ( ° Κ)

60 58 56 54 52 50 48 280

285

290

295

300

305

310

315

Brightness Temperature in AVHRR Channel 4 ( ° K)

320

325

330

12.00 UTC 150 140 130

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

120

β ( ° Κ)

110 100 90 80 70 60 50 280

285

290

295

300

305

310

315

Brightness Temperature in AVHRR Channel 4 ( ° K)

320

325

330

LST dependence on T4 and PW for T4 - T5 =0.5 ° K 3,9

T4

LST - T4 (° K)

3,4

280 285 290 295 300 305 310 315 320 325 330

2,9

2,4

1,9

1,4 0,3

0,8

1,3

1,8

2,3 PW (gr/cm2)

2,8

3,3

3,8

4,3

LST dependence on T4 and PW for T4 - T5 =1 ° K

4,8 T4 280 285 290 295 300 305 310 315 320 325 330

LST - T4 (° K)

4,3

3,8

3,3

2,8

2,3 0,3

0,8

1,3

1,8

2,3 PW (gr/cm2)

2,8

3,3

3,8

4,3

LST dependence on T4 and PW for T4 - T5 =1.5 ° K 6,5

6

T4 280 285 290 295 300 305 310 315 320 325 330

LST - T4 (° K)

5,5

5

4,5

4

3,5 0,3

0,8

1,3

1,8

2,3 PW (gr/cm2)

2,8

3,3

3,8

4,3

LST dependence on T4 and PW for T4 - T5 =2 ° K

7,5 T4 280 285 290 295 300 305 310 315 320 325 330

LST - T4 (° K)

7

6,5

6

5,5

5 0,3

0,8

1,3

1,8

2,3 PW (gr/cm2)

2,8

3,3

3,8

4,3