Land surface temperature and emissivity retrieval from satellite measurements

Abstract

Land surface temperature (LST) is maintained by the incoming solar and longwave irradiation, the outgoing terrestrial infrared radiation, the sensible and latent heat flux, and the ground heat flux. Therefore, LST is a good indicator of the energy balance at the Earth's surface. Long-term and reliable estimates of LST are required for multiple purposes, e.g. as input to general circulation models, numerical weather prediction, climate change detection, vegetation health monitoring, change detection related to desertification processes etc. Only satellite-based radiance measurements provide the temporal coverage and spatial resolution required to run these models and analyze the processes. For measurements in the atmospheric windows, the top-of-the-atmosphere (TOA) radiance is dominated by surface-leaving radiance. Therefore, LST can be estimated from radiances measured by satellite sensors. The sensor-specific TOA measurements are influenced by surface properties (LST and emissivity) and atmospheric absorption and transmission. Besides the complications due to atmospheric attenuation, a direct separation of LST and emissivity from passive radiometric measurements alone is not feasible because the problem is underdetermined: for a sensor with N spectral channels, there are N measurements but N + 1 unknowns (i.e. N spectral emissivities and LST). In order to make this ill-posed problem solvable, different assumptions for regularization of the underdetermined condition are made. This work is a contribution to the field of spaceborne remote sensing for measurement of Earth's surface temperature and emissivity from passive radiometry. At the Institute of Meteorology and Climate Research, Forschungszentrum Karlsruhe/University of Karlsruhe, Germany, NOAA AVHRR data are archived at the Meteorological Satellite Applications (MSA) group since several years and MSG SEVIRI data are received from 2004 using a High Rate User Station. The aim of this study was to select an appropriate method for estimating LST, which necessitates emissivity estimation as a pre-requisite, and establish an operational set-up with the method adapted to NOAA AVHRR and MSG SEVIRI data. The temperature independent thermal infrared spectral indices (TISI) method, which is a physical method, was employed for emissivity estimation, and subsequently LST and emissivity were decoupled from surface radiances. A physical method was approved, rather than methods based on empirical relationships which could be easily implemented, in order to capture the emissivity dynamics for various land surface types. During daytime, reflected solar irradiance and surface emission at ∼3.8 μm are approximately equal. The reflectivity is derived using atmospherically corrected surface radiances and a combination of day-night radiance ratios (TISI) between two channels. In order to resolve the underdetermination, it was assumed that emissivity remains constant for day and night and land surface behaves like a Lambertian surface. The TISI method was adapted to NOAA 9-16 AVHRR channels and MSG-1 SEVIRI channels. A numerical analysis was performed with simulated surface radiances showing that the achievable accuracy is better than 1.5 K - 2 K for LST and about 0.005 for emissivity (AVHRR channel 5) independent of surface type. The major source of error could be improper atmospheric information, because the atmospheric corrections cannot be better than the supplied information. However, it was assumed that atmospheric information used in the present study was correct. The method was applied to AVHRR data for a large part of central Europe, and for different days in order to observe seasonal differences. Additionally, a simple normalized difference vegetation index (NDVI)-based method for emissivity estimation was tuned for the study area using TISI-based emissivity as an input and results from both the approaches were intercompared.