Using Machine learning method to estimate Air Temperature from MODIS over Berlin

Friday, 18 December 2015
Poster Hall (Moscone South)
Forough Marzban1, Reneh Preusker1, Sahar Sodoudi1, Hamid Taheri1 and Masoud Allahbakhshi2, (1)Free University of Berlin, Berlin, Germany, (2)GFZ, Potsadm, Germany
Land Surface Temperature (LST) is defined as the temperature of the interface between the Earth’s surface and its atmosphere and thus it is a critical variable to understand land-atmosphere interactions and a key parameter in meteorological and hydrological studies, which is involved in energy fluxes. Air temperature (Tair) is one of the most important input variables in different spatially distributed hydrological, ecological models.

The estimation of near surface air temperature is useful for a wide range of applications. Some applications from traffic or energy management, require Tair data in high spatial and temporal resolution at two meters height above the ground (T2m), sometimes in near-real-time. Thus, a parameterization based on boundary layer physical principles was developed that determines the air temperature from remote sensing data (MODIS). Tair is commonly obtained from synoptic measurements in weather stations. However, the derivation of near surface air temperature from the LST derived from satellite is far from straight forward. T2m is not driven directly by the sun, but indirectly by LST, thus T2m can be parameterized from the LST and other variables such as Albedo, NDVI, Water vapor and etc.

Most of the previous studies have focused on estimating T2m based on simple and advanced statistical approaches, Temperature-Vegetation index and energy-balance approaches but the main objective of this research is to explore the relationships between T2m and LST in Berlin by using Artificial intelligence method with the aim of studying key variables to allow us establishing suitable techniques to obtain Tair from satellite Products and ground data. Secondly, an attempt was explored to identify an individual mix of attributes that reveals a particular pattern to better understanding variation of T2m during day and nighttime over the different area of Berlin. For this reason, a three layer Feedforward neural networks is considered with LMA algorithm. Considering the different relationships between T2m and LST for different land types enable us to improve better parameterization for determination of the best non-linear relation between LST and T2m over Berlin during day and nighttime. The results of the study will be presented and discussed.