Air temperature estimation with MSG-SEVIRI data: Calibration and validation of the TVX algorithm for the Iberian Peninsula

Air temperature can be estimated from remote sensing by combining information in thermal infrared and optical wavelengths. The empirical TVX algorithm is based on an estimated linear relationship between observed Land Surface Temperature (LST) and a Spectral Vegetation Index (NDVI). Air temperature is assumed to be equal to the LST corresponding to the effective full vegetation cover, and is found by extrapolating the line to a maximum value of NDVI_max. The algorithm has been tested and reported in the literature previously. However, the effect of vegetation types and climates and the potential variation in NDVI of the effective full cover has not been subject for investigation. The present study proposes a novel methodology to estimate NDVI_max that uses observed air temperature to calibrate the NDVI_max for each vegetation type. To assess the validity of this methodology, we have compared the accuracy of estimates using the new NDVI_max and the previous NDVI_max that have been proposed in literature with MSG-SEVIRI images in Spain during the year 2005. In addition, a spatio-temporal assessment of residuals has been performed to evaluate the accuracy of retrievals in terms of daily and seasonal variation, land cover, landscape heterogeneity and topography. Results showed that the new calibrated NDVI_max perform well, with a Mean Absolute Error ranging between 2.8 °C and 4 °C. In addition, vegetation-specific NDVI_max improve the accuracy compared with a unique NDVI_max.     

Land surface emissivity retrieval from combined mid-infrared and thermal infrared data of MSG-SEVIRI

This work addressed the retrieval of Land Surface Emissivity (LSE) from combined mid-infrared and thermal infrared data of Spinning Enhanced Visible and Infra-Red Imager (SEVIRI) onboard the geostationary satellite—Meteosat Second Generation (MSG). To correct for the atmospheric effects in satellite measurements, a new atmospheric correction scheme was developed for both Middle Infra-Red (MIR) and Thermal Infra-Red (TIR) channels. For the MIR channel, because it is less sensitive to the change of water vapor content, the clear-sky and time-nearest European Centre for Median-range Weather Forecast (ECMWF) atmospheric data were used for the images where no atmospheric data are available. For TIR channels, a modified model of Diurnal Temperature Cycle (DTC) used by Göttsche and Olesen [Göttsche, F. M., and Olesen, F. S. (2001). Modeling of diurnal cycles of brightness temperature extracted from METEOSAT data. Remote Sensing of Environment, 76, 337-348.] and Schädlich et al. [Schädlich, S., Göttsche, F. M., and Olesen, F. S. (2001). Influence of land surface parameters and atmosphere on METEOSAT brightness Temperatures and generation of land surface temperature maps by temporally and spatially interpolating atmospheric correction. Remote Sensing of Environment, 75, 39-46.] was adopted. The separation of Land Surface Temperature (LST) and LSE is based on the concept of the Temperature Independent Spectral Indices (TISI) [Becker, F., and Li, Z. L. (1990a). Temperature independent spectral indices in thermal infrared bands. Remote Sensing of Environment, 32, 17-33.] constructed with one channel in MIR and one channel in TIR. The results of two different combinations (combination of channels 4 and 9 and of channels 4 and 10) and two successive days at six specific locations over North Africa show that the retrievals are consistent. The range of emissivity in MSG-SEVIRI channel 4 goes from 0.5 for bare areas to 0.96 for densely vegetated areas, whereas the emissivities in MSG-SEVIRI channels 9 and 10 are usually from 0.9 to 0.95 for bare areas and from 0.95 to 1.0 for vegetated areas. For densely vegetated areas, the emissivities in MSG-SEVIRI channel 9 are larger than the ones in channel 10, whereas the opposite is observed over bare areas. The rms differences between two combinations over the whole studied region are 0.017 for emissivity in channel 4, 0.008 for emissivity in channel 9 and 0.007 for emissivity in channel 10.     

Accuracy assessment of land surface temperature retrievals from MSG2-SEVIRI data

The accuracy of the Land Surface Temperature (LST) product generated operationally by the EUMETSAT Land Surface Analysis Satellite Applications Facility (LSA SAF) from the data registered by the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on board the geostationary METEOSAT Second Generation 2 (MSG2, Meteosat 9) satellite was assessed on two test sites in Eastern Spain: a homogeneous, fully vegetated rice field and a high-plain, homogeneous area of shrubland. The LSA SAF LSTs were compared with ground LST measurements in the conventional temperature-based (T-based) method. We also validated the LSA SAF LST product by using an alternative radiance-based (R-based) method, with ground LSTs calculated from MSG-SEVIRI channel 9 brightness temperatures (at 10.8 μm) through radiative transfer simulations using atmospheric temperature and water vapor profiles together with surface emissivity data. Two lakes were also used for validation with the R-based method. Although the LSA SAF LST algorithm works mostly within the uncertainty expectation of ± 2 K, both validation methods showed significant biases for the LSA SAF LST product, up to 1.5 K in some cases. These biases, with the LSA SAF LST product overestimating reference values, were also observed in previous studies. Nevertheless, the present work points out that the biases are related to the land surface emissivities used in the operational generation of the product. The use of more appropriate emissivity values for the test sites in the LSA SAF LST algorithm led to better results by decreasing the biases by 0.7 K for the shrubland validation site. Furthermore, we proposed and checked an alternative algorithm: a quadratic split-window equation, based on a physical split-window model that has been widely proved for other sensors, with angular-dependent coefficients suitable for the MSG coverage area. The T-based validation results for this algorithm showed LST uncertainties (robust root-mean-squared-errors) from 0.2 K to 0.5 K lower than for the LSA SAF LST algorithm after the emissivity replacement. Nevertheless, the proposed algorithm accuracies were significantly better than those obtained for the current LSA SAF LST product, with an average accuracy difference of 0.6 K.