Mapping air temperature using time series analysis of LST: the SINTESI approach

• Main Authors: S. M. Alfieri, F. De Lorenzi, M. Menenti
• Format: Article
• Language: English
• Published: Copernicus Publications 2013-07-01
• Series: Nonlinear Processes in Geophysics
• Online Access: http://www.nonlin-processes-geophys.net/20/513/2013/npg-20-513-2013.pdf

This paper presents a new procedure to map time series of air temperature (<i>T</i>_a) at fine spatial resolution using time series analysis of satellite-derived land surface temperature (LST) observations. The method assumes that air temperature is known at a single (reference) location such as in gridded climate data with grid size of the order of 35 km × 35 km. The LST spatial and temporal pattern within a grid cell has been modelled by the pixel-wise ratios <i>r (x,y,t)</i> of the LST at any location to the LST at a reference location. A preliminary analysis of these patterns over a decade has demonstrated that their intra-annual variability is not negligible, with significant seasonality, even if it is stable throughout the years. The intra-annual variability has been modeled using Fourier series. We have evaluated the intra-annual variability by theoretically calculating the yearly evolution of LST (<i>t</i>) for a range of cases as a function of terrain, land cover and hydrological conditions. These calculations are used to interpret the observed LST <i>(x,y,t)</i> and <i>r (x,y,t)</i>. The inter-annual variability has been evaluated by modeling each year of observations using Fourier series and evaluating the interannual variability of Fourier coefficients. Because of the negligible interannual variability of <i>r (x,y,t)</i>, LST <i>(x,y,t)</i> can be reconstructed in periods of time different from the ones when LST observations are available. Time series of <i>T</i>_a are generated using the ratio <i>r (x,y,t)</i> and a linear regression between LST and <i>T</i>_a. Such linear regression is applied in two ways: (a) to estimate LST at any time from observations or forecasts of <i>T</i>_a at the reference location; (b) to estimate <i>T</i>_a from LST at any location. The results presented in this paper are based on the analysis of daily MODIS LST observations over the period 2001–2010. The <i>T</i>_a at the reference location was gridded data at a node of a 35 km × 35 km grid. Only one node was close to our study area and was used for the work presented here. The regression of <i>T</i>_a on LST was determined using concurrent observations of <i>T</i>_a at the four available weather stations in the Valle Telesina (Italy), our study area. <br><br> The accuracy of our estimates is consistent with literature and with the combined accuracy of LST and <i>T</i>_a. We obtained comparable error statistics when applying our method to LST data during periods different but adjacent to the periods used to model of <i>r (x,y,t)</i>. The method has also been evaluated against <i>T</i>_a observations for earlier periods of time (1984–1988), although available data are rather sparse in space and time. Slightly larger deviation were obtained. In all cases five days of averages from estimated and observed <i>T</i>_a were compared, giving a better accuracy.