Land surface temperature : a comparison of products from polar orbiting and geostationary satellites
thesisposted on 2014-05-16, 14:50 authored by Edward Matthew Comyn-Platt
This thesis has investigated land surface temperature (LST) products based on data from the Advanced Along-Track Scanning Radiometer (AATSR) and the Spinning Enhanced Visible and Infrared Imager (SEVIRI). Particular attention was paid to the long-term goal of an LST retrieval which utilises the benefits of both polar orbiting and geostationary instruments, represented by AATSR and SEVIRI, respectively. This study included an assessment of several current methods of LST retrieval, all of which were variations of the generalised split-window (GSW) algorithm. Satellite retrieved LST were compared with in situ observations at three validation sites in Europe and Africa. The results of the validation exercise suggested that for homogenous sites, Gobabeb and the “RMZ” farm in Namibia, the GSW algorithms met validation accuracies ±1-2.5 K. The accuracy of the GSW algorithms was significantly poorer at the heterogeneous site in Evora, Portugal where daytime matchups only met validation accuracies of ±5 K. Cross-platform comparisons at and around the validation sites suggested that the products were strongly correlated for night-time matchups at all sites. The daytime matchups showed greater levels of discrepancy between instruments given the combined effects of viewing geometry and sunlit/sunshade fraction observed by the satellite instruments. A novel optimal estimation (OE) method of LST retrieval has been proposed within the 1DVar framework. 1DVar is a stand-alone OE model which is based on the same architecture as 3D/4DVar systems used in a number of numerical weather prediction (NWP) and climatic models. A full diagnostic assessment of the 1DVar OE system has been presented with respect to the quality of the retrieved LST estimate based on AATSR and SEVIRI data. This demonstrated that the 1DVar OE system could reduce the LST error to 5-20% of the error associated with a priori LST estimate for both AATSR and SEVIRI. However, the implementation of the 1DVar scheme was hindered by the lack of a reliable emissivity resource; hence much of the gain in precision was compromised to the propagated emissivity error. Validation and verification of the 1DVar OE system using real data was presented in Chapter 6 of this thesis. The comparisons with in situ observations suggested only marginal improvements from the performance of the GSW products. However, the residual brightness temperatures were in much closer agreement with observations, hence the dataset was self-consistent which provided added confidence. Furthermore, this demonstration of a 1DVar scheme justifies the use of SEVIRI and AATSR data within a 3D/4DVar system which would facilitate combined platform LST retrieval. Cross-platform comparisons demonstrated that the largest sources of discrepancy were physical differences which were independent of algorithm. There was a strong seasonality, ~4K, in all the cross-platform comparisons with daytime data. Furthermore, there was a strong correlation between the cross-platform discrepancy and view-angle. This was highlighted as a major caveat of satellite LST products which are generally inferred as nadir observations by the end-user community. This is not the case and understanding the viewing geometry associated with satellite LST estimates is paramount to their various applications.
Supervisor(s)Remedios, John; Good, Elizabeth; Saunders, Roger
Date of award2014-05-01
Author affiliationDepartment of Physics and Astronomy
Awarding institutionUniversity of Leicester