Ocean surface air temperature derived from multiple data sets and artificial neural networks

This paper presents a new method to derive monthly averaged surface air temperature, T-a, from multiple data sets. Sea Surface Temperature (SST) from the National Centers for Environmental Prediction (NCEP) and total precipitable water (W) from the SSM/I sensor are used as inputs to Artificial Neural Networks (ANN). Surface air temperature (T-a) measurements from the Surface Marine Data (SMD) are used to develop and evaluate the methodology. When globally evaluated with SMD data, the bias of the new method is small (0.050 degrees C +/- 0.26 degrees C), and the accuracy expressed as root-mean square (rms) differences has a small global mean (0.73 degrees C +/-. 0.37 degrees C). These biases and rms differences are smaller than those obtained using NCEP reanalyses and TIROS Operational Vertical Sounder (TOVS) data products. When evaluated with the TOGA-TAO array measurements over the tropical Pacific, the ANN mean bias and rms differences have similarly small values: 0.37 degrees C and 0.61 degrees C, respectively.