On-line prediction of astronomical seeing fluctuations with neural networks

The European Southern Observatory's Astronomical Site Monitor (ASM) for the Very Large Telescope at Cerro Paranal in Chile aims at delivering short-term predictions of the Seeing variability - from 10 up to 60 minutes ahead - to allow optimization of the observing strategy (observing block selection, scheduling of calibration tasks). Years of seeing monitoring at various astronomical sites have shown that the seeing is not stationary and motivated the definition of a variability measure, the FEFSC, i.e., the "Finite Exposure Fractional Seeing Change" (Racine 1996, Sarazin 1997), the impact of which on the performances of adaptive optics systems has been clearly illustrated (Rigaut & Sarazin 1999). Extensive data collected at Paranal have been used to appraise a forecasting methodology based on neural networks (NN). We show that NN achieve a success rate up to 70% in predicting the FEFSC trend. The forecasting tools which have been prototyped, are a contribution to a future telescope decision support system.