Estimation Of All-weather Land Surface Temperature From Multi-source Satellite Remote Sensing Observations

Land surface temperature(LST)is an important parameter at the earth-atmospthere boundary.It is not only a sentitive indicator of climate change but also a direct input for numerous land-atmosphere models.Therefore,LST has been widely applied in many fields.Compared to site-based measurements,satellite remote sensing has outstanding advantages including high-density coverage?repeated observation and low cost in measuring LST.With the rapid development of society,an all-weather remotely sensed LST,especially the all-weather remotely sensed LST at moderate/high spatial resolution(e.g.1 km)has been in urgent need in associated studies and applications.However,the most widely-used thermal infrared(TIR)remotely sensing fails to acquire valid LST due to cloud coverage under non-clear sky conditions.Therefore,single-source TIR cannot meet the requirement of acquisition of the all-weather LST.Integrating the satellite TIR and passive microwave(PMW)LST and integrating the TIR LST and reanalysis LST are two feasible approaches for the estimation of the allweather LST.However,there are bottlenecks regarding integration of these sources.As for the integration of TIR and PMW observations,the current main problems are(1)TIR LST and PMW LST are physically inconsistent;(2)TIR LST and PMW LST are inconstistent in spatial resolutions and(3)TIR-PMW integrated LST are not really “allweather” available due to PMW data gap.As for the integration of TIR LST and reanalysis LST for estimation of all-weather LST,there are bare of associated studies.Under this background,this study provided targeted solutions of the above problems,developed three methods to estimate remotely sensed LST at moderate/high spatial resolutions.By forming a set of methodology on estimation of all-weather LST from multi-source remote sensing observations,this study provides the fundamental theory and support the studies and applications in terms of all-weather remotely sensed LST.In this study,following four kinds of work were conducted:(1)A Thermal Sampling Depth Correction(TSDC)method was proposed for estimation of PMW LST.The thermal sampling depth(TSD)makes PMW LST contains the thermal radiation from subsurface and thus,the PMW is physically inconsistent with TIR LST.Therefore,there are systematical bias between the two LST.However,previous methods estimate PMW LST without physically addressing the TSD issue and thus,leading to low accuracies over areas,such as barren land with large TSD.In this context,the proposed TSDC was applied to barren land and physically correct the PMW LST to the TIR LST by quantificationally addressing the TSD effect when estimating PMW LST.Results show that the PMW LST corrected by TSD has much lower bias with TIR LST than PMW LST from previous methods.Validation against ground measurements shows that the accuracies of TSDC PMW LST are 10.14?12.84 K higher than PMW LST from previous methods.Therefore,TSDC method can solve the first problem in the integration of TIR and PMW observations over the barren land area.(2)A Temporal Component Decomposition(TCD)method was proposed to integrate TIR and PMW observations for estimation of 1-km all-weather LST.PMW LST has much lower spatial resolution than TIR LST and thus,cannot obtain the details at fine scales.Therefore,it is necessary to downscale the PMW LST before integration of TIR and PMW LST.However,previous methods downscale the PMW LST without addressing the LST variations in the temporal dimension and therefore easily make the TIR-PMW integrated LST's accuracy and applicability limited.In this context,the proposed TCD method integrates the TIR LST and PMW LST in both spatial and temporal dimensions Results show that the TCD can overcome the limitation of applicability and block effect issues due to the insufficient downscaling in previous studies.Validation against ground measurements shows that the TCD LST accuracies of 1.29?2.71 K,which is 1.39?3.17 K higher than the PMW-TIR integrated LST from previous methods.Therefore,TCD method can efficiently solve the second problem in the integration of TIR and PMW observations.(3)A Reconstruction of Brightness Temperature(RBT)method was proposed to integrate TIR and PMW observations for estimation of 1-km all-weather LST.Due to the lack of PMW brightness temperature(BT)in the temporal and spatial gaps of polar-orbited PMW sensors,current TIR-PMW integrated LST has missing values and therefore is not really “all-weather” available.In this context,the proposed RBT method at first reconstructed the spatial seamless PMW BT and then integrate the reconstructed BT and TIR LST to estimate all-weather LST based on a machine-learning model.Results show that the reconstructed BT has satisfactory spatiotemporal continuity and accuracies of 0.89?2.61 K.Besides,the RBT LST is a real spatial-seamless and “all-weather” LST and shows accuracies of 1.45?3.36 K in the validation against ground measurements.Therefore,TCD method can efficiently solve the third problem in the integration of TIR and PMW observations.(4)A Reanalysis and Thermal Infrared Remote Sensing Merging(RTM)method was proposed to integrate TIR LST and reanalysis data for estimation of 1-km all-weather LST.Studies on estimation of an all-weather LST at moderate/high spatial resolutions are very rare.In this context,based on the enhanced model of Temporal Component Decomposition of LST from the second study,the proposed RTM method integrates the TIR LST and reanalysis data in both spatial and temporal dimensions.Validation against ground measurements shows that the RTM LST has satisfactory accuracy of 2.03?3.98 K under all-weather conditions.Besides,benefiting from the high spatial integrity,RTM LST is spatial-seamless.Compared with a previous method for estimating LST from TIR LST and reanalysis data,it is found that the RTM LST has higher spatial integrity with 10%?17% valid values and the accuracy of RTM LST is not limited by the numbers of available samples of TIR LST.Therefore,RTM has satisfactory generalization ability and applicability and is efficient in integrating TIR LST and reanalysis data for estimation of an all-weather LST.