Several thermal models are introduced, based on both a steady-state and a non-steady-state heat balance equation. These are used to predict the surface temperature of dry bare desert background elements. Sensitivity analysis is performed for steady-state models, which indicates that for large measurement error the semiempirical models yield higher accuracy than the physical model. The non-steady-state models take account of variations of the meteorological parameters, and exhibit approximately 47% better performance for rapidly altering weather conditions. Also, the coefficients in these models may be used for a larger period of time to predict the surface temperature. Experimental tests are performed to examine the higher precision of the new models for predicting both the temperature and the thermal image of desert backgrounds. The mean prediction error of the new models is typically more than 25% better. Also, the contrast (variance) of the predicted thermal images is much closer to that of the real images during the daytime.