Remote Estimation Of Complete Urban Surface Temperature Based On Typical Urban Surfaces

Complete urban surface temperature(Tc)is a key variable for quantifying the urban surface-atmosphere interactions.The estimation of Tc usually requires detailed knowledge of surface geometry and facet temperatures,which are difficult to obtain by remote sensing.Therefore,simple but efficient strategies for the remote estimation of Tc are currently lacking.In addition,remote sensing cannot obtain enough directional radiometric temperatures(DRTs),kernel-driven models can simulate DRTs in all directions in the upper hemisphere space,but which the kernel-driven model that is most suitable for simulating urban surfaces is not yet clear.Facing these challenges,we proposed a comprehensive solution to estimate Tc remotely using only the DRTs without the assistance of ground-based information.First,in order to obtain a kernel-driven model that is most suitable for the simulation of DRTs of thermal anisotropy in urban surfaces,this paper conducts a comparative evaluation of 10 kernel-driven models from thermal anisotropy,DRTs and hot spot simulation accuracy.The results show that ?the simulation accuracy of 2-kernel models is better than that of 1-kernel models,and the difference of simulation accuracy(RMSE)between two types is close to 0.5 K.?Dual-kernel models based on physical deduction is superior to empirical-type dual-kernel models.That is,kernel function deduced with strict physical meaning has higher simulation accuracy.?The accuracy of optimal kernel-driven model is the highest(RMSE is above 0.5 K),the hot spot is the most accurate,which can simulate various urban surfaces.Second,by comparing Tc with the nadir temperature(To),DRTs within the entire hemisphere,and solid-angle integral temperature(SIT),we obtain the following results:?The optimal viewing azimuth and zenith angles at which the DRTs(TOVA)are the closest to Tc are ?s±90°(?s is the solar azimuth angle)and 45?60°,respectively.The MAE for the TOVA that most closely approximated Tc is above 1.08 K.?The estimation of Tc could be further improved once the DRTs from all-direction are available,and the MAE for the SIT that most closely approximated Tc is only 0.87 K.?The MAE of the SIT that best approximates Tc by combining only four DRTs and a kernel-driven model is only slightly higher than that of the specific SIT based on the original DRTs,indicating that the remote estimation of Tc using limited DRTs is plausible.Simultaneously,combined with the needs of actual field measurements,we give the suggested flight plans for the remote sensing of Tc without ground-based information corresponding to each scenario.In conclusion,these proposed strategies for estimating Tc are generally simple and effective,and can help facilitate the estimation of Tc and therefore help advance the thermal remote sensing of urban climate and environment.