Spatio-temporal analysis of the relationship between LST from MODIS and air temperature in New Zealand

The ambient air temperature (T-a) is an important environmental parameter which can be estimated from satellite observations of the land surface temperature (LST) using a linear regression model. This paper attempts to answer the question of whether the series of a single pixel or a spatially averaged series over several pixels should be used for modelling T-a from remotely sensed LST data. Sensitivity of LST-T-a relationship to the moderate resolution imaging spectroradiometer (MODIS) window size, which determines the number of pixels contributed in the correlations, over a number of test sites in New Zealand was analysed. LST series of a single pixel over a period of 10 years gave a correlation coefficient 0.80 with T-a measurements. Bootstrapping by random resampling from seasonal subsets of both time-series was applied to determine seasonal and inter-annual variability of LST-T-a relationship. A fast Fourier filtering was applied for noise reduction and detection of dominant spectra in LST series. Spatially averaged time-series from larger windows, which included more pixels, showed slightly higher agreement with T-a measurements. We considered the effects of wind speed (WS) and wind direction (WD) on the LST-T-a relationship. Highest correlation between T-a and LST time-series was achieved using a 25 x 25 window at 2 a parts per thousand currency sign WS < 8 ms(-1). No significant effect due to WD was found in the results. MODIS-Terra nighttime (10:30 PM) observations showed the highestwhile MODIS-Aqua nighttime (1:30 AM) observations showed the lowest agreement with T-a measurements. These results indicate that the best approach for modelling T-a based on LST observations from MODIS in the long-term is to use a spatially averaged LST series over a window of 5 x 5 to 25 x 25 pixels, with a consideration of WS effects and observation times.