A generalized split-window algorithm for retrieving land-surface temperature from space

TLDR

The study proposes a generalized split‑window method for retrieving land‑surface temperature from AVHRR and MODIS data. The method derives angle‑dependent coefficients through regression on radiative‑transfer simulations and selects the appropriate coefficient set using atmospheric sounding data for lower boundary temperature and column water vapor. Simulations show that angle‑dependent coefficients achieve about 1 K accuracy across the full scan swath and that the algorithm improves LST retrieval accuracy while reducing sensitivity to emissivity uncertainty and instrument quantization error.

Abstract

Proposes a generalized split-window method for retrieving land-surface temperature (LST) from AVHRR and MODIS data. Accurate radiative transfer simulations show that the coefficients in the split-window algorithm for LST must vary with the viewing angle, if the authors are to achieve a LST accuracy of about 1 K for the whole scan swath range (/spl plusmn/55/spl deg/ from nadir) and for the ranges of surface temperature and atmospheric conditions over land, which are much wider than those over oceans. The authors obtain these coefficients from regression analysis of radiative transfer simulations, and they analyze sensitivity and error over wide ranges of surface temperature and emissivity and atmospheric water vapor abundance and temperature. Simulations show that when atmospheric water vapor increases and viewing angle is larger than 45/spl deg/, it is necessary to optimize the split-window method by separating the ranges of the atmospheric water vapor, lower boundary temperature, and the surface temperature into tractable subranges. The atmospheric lower boundary temperature and (vertical) column water vapor values retrieved from HIRS/2 or MODIS atmospheric sounding channels can be used to determine the range for the optimum coefficients of the split-window method. This new algorithm not only retrieves land-surface temperature more accurately, but is also less sensitive to uncertainty in emissivity and to instrument quantization error.