Land surface temperatures derived from the advanced very high resolution radiometer and the along‐track scanning radiometer: 2. Experimental results and validation of AVHRR algorithms

This is the second paper in a series of three papers which discuss the problem of deriving accurate land surface temperatures (LSTs) from infrared radiometers on board satellites, such as the NOAA advanced very high resolution radiometer (AVHRR) and the ERS 1 along‐track scanning radiometer (ATSR). In the first part a detailed description of a theoretical model for deriving LSTs based on linearization of the radiative transfer equation was given. This paper focuses on an experimental investigation aimed at establishing the validity and accuracy of the so‐called split‐window technique applied to the land surface. The paper concentrates on the use of AVHRR data for LST estimation; the third part of this series will discuss ATSR data. Because of the lack of good ground‐truth data, an appropriate method has been devised to create an LST validation data set suitable for comparison with the AVHRR or ATSR infrared measurements. Multiple regression relations between satellite temperature measurements and areal averages of in situ temperatures are presented for several surface types from two field sites. Root‐mean‐square differences of about ±1.5°C are obtained for all of the surface types and the correlations are extremely good (r 2 ≥ 0.95). These results are considered to represent the limiting accuracies that can be expected from the split‐window formulation applied to high‐resolution AVHRR measurements. Results are also presented for dual‐channel and triple‐window algorithms. The effects of spatial variability and viewing geometry on surface temperature retrieval are discussed. It is found that for the vegetated surfaces studied, emissivity effects are minimized by cavity effects and angular effects are only important for structured vegetation (e.g., row crops) and certain viewing geometries. The spatial temperature variability is large at scales below 1 km 2 (>3°C), is small at a scale of 1–10 km 2 (<1°C), and increases only slowly with spatial scale beyond scales of 10 km 2 or so. Comparisons between the LST algorithms given in part 1, published algorithms by other authors and the in situ data indicate that for these when used with climatological atmospheres and laboratory measured emissivities.