Quantifying uncertainties in land surface temperature due to atmospheric correction: Application to Landsat-7 data over a Mediterranean agricultural region

The impact of using non-coincident radiosoundings to remove atmosphere effect from thermal radiances is analyzed here. We considered 27 Landsat-7 ETM+ images acquired over a Mediterranean agricultural region, benefiting from nearby radiosoundings launched almost 2 hours later, and from the availability of a network of ground stations deployed over different types of ecosystems. We observed that, in the conditions of our images, surface temperature estimates slightly improved when considering one atmospheric profile interpolated to our particular date, time and location, in comparison with the use of non-coincident radiosoundings. However, it may imply an error up to ±2.5 K for brightness temperatures (in particular for very high temperatures and during summer when the atmosphere was warmer and the vapor pressure was higher), leading to important errors in the derivation of surface energy fluxes. The characterization of the lowest atmosphere layer appeared to be essential to improve the estimates of brightness temperatures.