Estimating the spatial distribution of daily air temperature by Time Series Analysis of MODIS Land Surface Temperature

The lack of information about air temperature (Ta) spatial distribution is a central problem in agro-environmentalstudies at all scales. Air temperature is in fact the most important variable, along with radiation, to influence cropdevelopment and its spatial distribution. It is used as input in a large variety of models ranging from prediction ofpotential evapotranspiration and crop yields to climatic impact studies.The aim of the Italian project Agroscenari is to determine future adaptation of crops to climate change by studyingrelatively small areas representative of specific production systems. This requires knowledge of climate at muchhigher spatial resolution than GCM-based scenarios. This applies particularly to areas, such as the Valle Telesina(Italy), one of our study areas, characterized by complex relief (Alfieri et al, 2010). We have down-scaled GCMscenariosin two steps: a) statistical downscaling by relating GCM-fields of air temperature to observations griddedat a 35 km x 35 km resolution; b) by using MODIS Land Surface Temperature to characterize sub-grid spatialvariability of time series of air temperature. We describe here the step (b).Relation between air temperature and surface temperature. Near surface air temperature and Land Surface Temperature(LST) observed from satellites are correlated. Daily land surface temperature (LST) of Terra ModerateResolution Imaging Spectrometer (MODIS) sensor was used for the analysis. We have established a regressionequation for all available stations relating daily observations of air temperature with LST observations for theperiod 2000 - 2006.Characterization of spatial and temporal patterns. We have calculated for each MODIS pixel time series of theratio of LST at each pixel-location to the LST at the location of a reference node of the 35 km x 35 km grid.Removal of cloud affected pixel values and gapfilling was performed by HANTS algorithm (Menenti et al, 1993)producing continuous time series of maximum and minimum LST. Fourier analysis of the time series of LST ratioswas performed for each year, showing three main periodical components (yearly, half yearly and seasonal). Wehave evaluated the interannual variability of amplitude and phase values to conclude that we could use their meanvalues to characterize the annual temporal profile of the pixel-wise ratio.Estimation of the spatial pattern of daily air temperature. Daily maximum air temperature at each pixel locationwas estimated by combining the spatial and temporal pattern of the ratio with the regression equation giving airtemperature as function of surface temperature. This gives maximum air temperature at any location as function ofmaximum air temperature at the reference node.Estimated maximum air temperature was compared with observation at four stations in Valle Telesina area givinga RMSE between 2.9 and 4.2 k.