Technical Note

Comparison of Split-Window Algorithms for the Retrieval of Surface Temperatures from AVHRR Data in Southern Québec

Jae-Dong JANG¹, Alain A. VIAU¹ and François ANCTIL²

¹ Department of Geomatics Sciences, Laboratoire de géomatique agricole et d’agriculture de précision (GAAP), Centre de Recherche en Géomatique (CGR), Université Laval, Québec (Québec), CANADA G1K 7P4

² Department of Civil Engineering, Université Laval, Québec (Québec), CANADA G1K 7P4

Manuscript was officially accepted as a technical note by

Canadian Journal of Remote Sensing

October 2003

Corresponding author : Jae-Dong Jang : phone: +1-418-656-2131 ext. 12821; fax: +1-418-656-5837; e-mail: jae-dong.jang@scg.ulaval.ca

Abstract

Land surface temperature is a significant parameter for energy and water cycles in the earth–atmosphere system. Many studies have proposed various algorithms, such as the split-window method, for retrieving surface temperatures from two different thermal infrared bands of satellite data. Each algorithm is developed for a limited study area and application. Here, as part of developing an optimal split-window method in southern Québec, a comparison of algorithms is carried out to evaluate the performance of each algorithm and examine correlation and differences between them. Furthermore, the deviation of surface temperature derived from various algorithms is compared with the brightness temperature of the thermal infrared band at 11 μm as a sensitivity test for atmospheric effects.

Résumé

Les échanges massiques (eau) et d’énergie à l’interface Terre atmosphère sont largement influencés par la température de surface de la Terre. Plusieurs études ont démontré la faisabilité d’extraire cette température d’observations satellitaires sur deux bandes différentes de fréquences infrarouges thermiques. Des algorithmes, tel que la méthode ‘split-window’, ont d’ailleurs été mis au point à cet effet pour des régions ou encore des applications spécifiques. L’objectif de cette étude consiste en le développement d’une méthode optimale d’extraction de type ‘split-window’ pour le sud du Québec. Une comparaison est d’abord effectuée en utilisant des images AVHRR. Les différentes températures de surface obtenues sont ensuite confrontées à la température de brillance de la bande infrarouge thermique (11 μm), afin d’évaluer leur sensibilité aux influences atmosphériques.

Introduction

Land surface temperature is a significant factor for understanding the boundary layer climate and for estimating daily evapotranspiration. Together with air temperature, land surface temperature is a key parameter of the energy and water cycles of the earth–atmosphere system.

In previous studies, land surface temperature has been retrieved using thermal infrared radiation emitted from the surface (Price, 1984; Becker and Li, 1990; Prata and Platt, 1991; Kerr et al. , 1992; Coll et al. , 1994b; Ulivieri et al. , 1994; Sobrino and Raissouni, 2000). The estimation of surface temperature (TS) has frequently been carried out with split-window equations using two thermal infrared bands, located at 11 μm and 12 μm, from the advanced very high resolution radiometer (AVHRR), Geostationary Operational Environmental Satellite (GOES) or Meteosat systems. It is difficult to retrieve precise TSs because the thermal infrared radiance is influenced by atmospheric absorption by water vapor and other gases. The split-window algorithm uses two adjacent thermal infrared channels, centered at 11 μm and 12 μm for AVHRR to retrieve surface temperatures because of their different atmospheric transmittances. The accuracy of the split-window algorithm depends on the magnitude of difference between the emissivities of the surface in the two bands (Becker, 1987). A difference of 0.01 between the two bands may cause an error of about 2 °C in TS retrieval (Qin and Karnieli, 1999). In practice, the ground emissivity cannot be derived from satellite images. Thus, the error in TS retrieval reaches 2 °C because the ground emissivity of the two channels may have a difference ranging from 0.002 to 0.008 (Sobrino et al. , 1991; Coll et al. , 1994a; Qin and Karnieli, 1999). The emissivity is close to one for dense vegetation canopies, and the density of vegetation can be derived from satellite images. As a vegetation index, the Normalized Difference Vegetation Index (NDVI) is calculated from visible and near infrared bands, and is related to the density of vegetation. Thus, the emissivity has been estimated using the relationship between emissivity and NDVI (Van de Griend and Owe, 1993; Cihlar et al. , 1997) in forest regions.

The objective of this study is to compare various split-window algorithms for retreving surface temperatures for southern Québec with better accuracy. The satellite images are corrected for radiometric, geometric and atmospheric error to retrieve accurate surface radiation. The emissivity is estimated from NDVI values calculated from reflectances of corrected visible and near infrared bands. TSs are retrieved using various split-window algorithms from two thermal infrared bands with the derived emissivity. A description of the study area and data sets is provided in Section 2. The pre-processing of satellite images for radiometric, geometric and atmospheric corrections is presented in Section 3. In Section 4, the comparison of split-window algorithms is explained. The summary is presented in the final section.

Study area and data set

Our experimental study area is between latitudes 45°N and 50°N, and longitudes 76°W to 67°W in southern Québec province. There are large boreal regions in the north, broadleaf forests in the south, and agricultural regions mostly around Montreal city and along the river (Figure 1). The reproductive activity of the forest in the study area generally occurs in the period from May to October, but there is snow on the mountains in May and October.

The NOAA series of meteorological satellites has played a significant role in remote sensing for the last two decades. These satellites are placed in sun-synchronous orbits at an average altitude of 833 km. They have the advantage of covering the same area at least twice in each 24-hour period: once during the day and again at night. The acquired images have advantages because of their field of view of ±55.4° and their wide swath (2800 km) and their spatial resolution of 1.1 km at the nadir.

Figure 1. The study area in southern Québec province.

The data set of NOAA-14 AVHRR was used in this comparison study. The scan times of the NOAA-14 image over the study area in daytime were around 3 pm local time. To estimate emissivity and TS, only the images from the daytime pass were employed. Three AVHRR level 1b images were obtained from the Satellite Active Archive (SAA) of NOAA for the fifth of June, the 25^th of July and the 27^th of August 2000, respectively. The five bands of AVHRR image consist of visible (0.58–0.68 μm), near infrared (0.725–1.10 μm), middle infrared (3.55–3.98 μm), and two thermal infrareds (10.3–11.3 μm and 11.5–12.5 μm).

Pre-processing

The pre-processing of satellite image data generally consists of radiometric, geometric, and atmospheric corrections. As the first step, the radiometric correction is a process that converts digital numbers derived by sensors to the top of atmosphere (TOA) reflectances (in optical bands) or brightness temperatures (TBs, in thermal bands). In this study, radiometric correction was carried out for the optical bands using a simple formula (Rao and Chen, 1999). The reflectances were derived from channels 1 and 2, and the TBs from Channels 3, 4 and 5.

For geometric correction, we used a two-step process, correcting the panoramic effect and then mapping using a second-order polynomial using the earth location data of AVHRR level 1b. Pixel spacing is 1.1 km by 1.1 km directly under the satellite path, and increase gradually in size to about 6 km in across track by 2 km in along track at the edges of the swath, the so-called panoramic effect. The bigger pixels are split into several pixels the same size as pixels at the nadir. Consequently, the panoramically corrected image has more pixels in each column than the original image. To correct other sources of geometric distortion, the earth location data of every 40th pixel in the line header of level 1b data were used as ground control points for the second-order mapping polynomial. There was, however, some disagreement between the image after correction by the polynomial and the reference map, because the earth location data exhibit spatiotemporal variation (Krasnopolsky and Breaker, 1994). Thus, each image corrected by the polynomial was compared with the reference map and then further corrected by shifting manually to the approximate real location.

The atmospheric correction derives the reflectance at the top of the canopy (TOC) from TOA reflectance by removing the atmospheric effects. In this study, version 2 of the Simplified Method for the Atmospheric Correction (SMAC) was employed as a tool for the atmospheric correction of the visible and near infrared bands. SMAC is a simple algorithm consisting of empirical equations derived by comparing with results from the 5S model (Rahman and Dedieu, 1994). The performance of SMAC is much faster than that of 5S, but retaining comparable accuracy. As input parameters for SMAC, the water vapor content and ozone content were assumed constant at 2.3 g/cm² and 0.319 cm-atm, respectively, and a constant value of 0.07 was used for the aerosol optical depth at 550 nm radiation wavelength for clear sky conditions in eastern Canada (Teillet, 1992; Fedosejevs et al. , 2001).

Split-window algorithm

The emissivity used in split-window algorithms is a critical parameter for the accuracy of TS. Dense vegetation canopy has a high emissivity, with the value 0.98 given as representative (Chen et al. , 1989), while soil and rock areas have about 0.95 (Salisbury and D'Aria, 1992). NDVI values indicating the density of vegetation can be used as a proxy indicator for estimating emissivities. In this study, the emissivity was estimated by an empirical algorithm based on the relationship between emissivity and the logarithm of NDVI (Van de Griend and Owe, 1993; Cihlar et al. , 1997) in the range [0.955, 0.985]. The emissivity of channel 4, ε₄, and the emissivity difference of channels 4 and 5, ε₄ − ε₅ (Δε)_, are as follows.

ε₄ = 0.9897 + 0.029 ln(NDVI) (1)

Δε = 0.01019 + 0.01344 ln(NDVI) (2)

The split-window algorithms have been developed by many researchers, and six different algorithms were evaluated in this work and are presented in Table 1. The three daytime AVHRR images were employed as representative data for the evaluation of split-window algorithms. The evaluation of the algorithms was carried out by comparisons using root mean square error (RMSE) and correlation coefficients (Table 2). Total atmospheric water vapour content was estimated by Sobrino et al. (1999) method for S&R split-window algorithm. In the relationships between algorithms, correlation coefficients were generally higher than 0.97, although RMSEs varied from 0.36 to 3.65 °C . Considering the results of RMSE and correlation coefficients, the correlations between the Price and other algorithms were less with higher RMSE, while those between other algorithms were higher with less RMSE. The six algorithms produced quite large difference between some of them for retrieving surface temperatures. Thus the algorithms need to be tested for understanding the difference. The atmosphere is a key factor to modify the radiance from the Earth surface by water vapour or aerosol. In next step, the six algorithms were investigated for atmospheric effects.

* TS is surface temperature, T₄ and T₅ are TBs of AVHRR channel 4 and 5, ε=(ε₄ + ε₅)/2, T₀=273.15 (K), and W is total atmospheric water vapour content. ε² is the square of ε.

Each algorithm was further tested for relative sensitivity by analyzing the variation of the deviation between the TS and Channel 4 TB (Figure 2). The deviations were computed by averaging the differences between the TSs and TBs for each 1.5 K span of TB. Channel 4 TB (11.0 μm) was chosen instead of Channel 5 (12 μm), because there is more atmospheric effect (absorption by water vapour) in Channel 5 than in Channel 4 (Ouaidrari et al. , 2002). The Price method produced the largest deviation from Channel 4 TB, while other methods have smaller deviations but similar trend each other. The deviations of temperatures estimated by Coll were much less sensitive to the variation of Channel 4 TB than those of other algorithms, so the Coll algorithm may exhibit a lesser atmospheric effect on the accuracy. Additionally, the coefficients in the Coll equation were optimized for the mid-latitude region where our test area is located. The Coll algorithm may be proposed as a split-window algorithm for estimating TS in southern Québec with better accuracy.

Figure 2. Variations of the deviations of surface temperature (TS) from brightness temperature (TB) of AVHRR Channel 4.

Summary

Several split-window algorithms have been analyzed to find an optimal algorithm for retrieving surface temperature from AVHRR images in southern Québec. In this comparison study, RMSE and correlation coefficients were used to understand the relationship between algorithms. The correlation coefficients are generally above 0.97, and that between Ulivieri and P&P is more correlated than others. In testing sensitivity to atmospheric effects, the TS of the Coll algorithm is much less influenced by water vapour absorption, and the algorithm was calibrated for the mid-latitude region in the previous work. Therefore, the Coll algorithm may be used as a split-window method for retrieving TSs in southern Québec with better accuracy because it takes better account of atmospheric effects.

Acknowledgment

This study was done in the laboratory of agricultural geomatics and precision agriculture in Laval University and supported by funds from SOPFEU. The images of NOAA-14 AVHRR were derived from the website of NOAA’s Satellite Active Archive ().

References

Becker, F. 1987. The impact of spectral emissivity on the measurement of land surface temperature from a satellite. International Journal of Remote Sensing, Vol. 8, pp. 1509-1522.

Becker, F., and Li, Z.-L. 1990. Towards a local split window method over land surfaces. International Journal of Remote Sensing, Vol. 11, pp. 369-393.

Chen, J. M., Yand, B. J., and Zhang, R. H. 1989. Soil thermal emissivity as affected by its water content and surface treatment. Soil Science, Vol. 148, pp. 433-435.

CIFQ (Conseil de l’industrie forestière du Québec), 2003. Forèt. URL : http://www.cifq.qc.ca.

Cihlar, J., Ly, H., Li, Z., Chen, J., Pokrant, H., and Huang, F. 1997. Multitemporal, multichannel AVHRR data sets for land biosphere studies-artifacts and corrections. Remote Sensing of Environment, Vol. 60, pp. 35-57.

Coll, C., Caselles, V., and Schmugge, T. J. 1994a. Estimation of land surface emissivity differences in the split-window channels of AVHRR. Remote Sensing of Environment, Vol. 48, pp. 127-134.

Coll, C., Casselles, V., Sobrino, J. A., and Valor, E. 1994b. On the atmospheric dependence of the split-window equation for land surface temperature. International Journal of Remote Sensing , Vol. 15, pp. 105-122.

Fedosejevs, G., O'Neill, N. T., Royer, A., Teillet, P. M., Bokoye, A. I., and McArthur, B. 2001. Aerosol optical depth for atmospheric correction of AVHRR composite data. Canadian Journal of Remote Sensing, Vol. 26, No. 4, pp. 273-284.

Kerr, Y. H., Lagouarde, J. P., and Imbernon, J. 1992. Accurate land surface temperature retrieval from AVHRR data with use of an improved split window algorithm. Remote Sensing of Environment, Vol. 41, pp. 197-209.

Krasnopolsky, V. M., and Breaker, L. C. 1994. The problem of AVHRR image navigation revisited. International Journal of Remote Sensing, Vol. 15, pp. 979-1008.

Ouaidrari, H., Goward, S. N., Czajkowski, K. P., Sobrino, J. A., and Vermote, E. 2002. Land surface temperature estimation from AVHRR thermal infrared measurements: An assessment for the AVHRR Land pathfinder II data set. Remote Sensing of Environment, Vol. 81, pp. 114-128.

Prata, A. J., and Platt, C. M. R. 1991. Land surface temperature measurements from the AVHRR. Proceeding of The 5th AVHRR Data Users' Meeting, Tromso, Norway, (pp. 433-438), Edited by.

Price, J. C. 1984. Land surface temperature measurements from the split window channels of the NOAA 7 Advanced Very High Resolution Radiometer. Journal of Geophysical Research, Vol. 89, pp. 7231-7237.

Qin, Z., and Karnieli, A. 1999. Progress in the remote sensing of land surface temperature and ground emissivity using NOAA-AVHRR data. International Journal of Remote Sensing, Vol. 20, pp. 2367-2393.

Rahman, H., and Dedieu, G. 1994. SMAC: a simplified method for the atmospheric correction of satellite measurements in the solar spectrum. International Journal of Remote Sensing, Vol. 15, pp. 123-143.

Rao, C. R. N., and Chen, J. 1999. Revised post-launch calibration of the visible and near-infrared channels of the advanced very high resolution radiometer on the NOAA-14 spacecraft. NOAASIS.

Salisbury, J. W., and D'Aria, D. M. 1992. Emissivity of terrestrial materials in the 8-14 um atmospheric window. Remote Sensing of Environment, Vol. 42, pp. 83-106.

Sobrino, J. A., Coll, C., and Caselles, V. 1991. Atmospheric correction for land surface temperature using NOAA-11 AVHRR channels 4 and 5. Remote Sensing of Environment, Vol. 38, pp. 19-34.

Sobrino, J. A., and Raissouni, N. 2000. Toward remote sensing methods for land cover dynamic monitoring: application to Morocco. International Journal of Remote Sensing, Vol. 21, No. 2, pp. 353-366.

Sobrino, J. A., Raissouni, N., Simarro, J., Nerry, F., and Petitcolin, F. 1999. Atmospheric water vapor content over land surfaces derived from the AVHRR data: application to the Iberian Peninsula. IEEE Transactions on Geoscience and Remote Sensing, Vol. 37, No. 3, pp. 1425-1434.

SOPFEU (Société de protection des forêts contre le feu), 2003. Statistiques. URL : http://www.sopfeu.qc.ca/.

Teillet, P. M. 1992. An algorithm for the radiometric and atmospheric correction of AVHRR data in the solar reflective channels. Remote Sensing of Environment, Vol. 41, pp. 185-195.

Ulivieri, C., Castronouvo, M. M., Francioni, R., and Cardillo, A. 1994. A split-window algorithm for estimating land surface temperature from satellites. Advances in Space Research, Vol. 14, No. 3, pp. 59-65.

Van de Griend, A. A., and Owe, M. 1993. On the relationship between thermal emissivity and the normalized difference vegetation index for natural surfaces. International Journal of Remote Sensing, Vol. 14, pp. 1119-1131.