Temperature forecast and dome seeing minimization

Abstract

Abstract. Dome seeing may strongly deteriorate the final sharpness of a point source astronomical image, reducing its Image Quality. Both the telescope enclosure and the mirrors may contribute to the dome seeing, if air convection is induced by differences of temperature between them and external air. The prediction of the external air temperature with respect to a given time interval allows one to preset in advance the air conditioning temperature value in the telescope enclosure. With the aim to study the neural networks capabilities and limits to make short term temperature prediction, a few case studies have been carried out by using an autoregressive neural network model. The actual goal is to understand if and with which constraints a neural network model can actually be used in a NTT-like dome (i.e. telescope is in open air when observing and heat sources are highly controlled when close or inside the telescope's dome) for steering the daytime air conditioning system. We do not present any interface with an actual telescope: this paper presents a feasibility study about the forecasting methodological approach rather than its operational application to a specific telescope.The results show that on site output prediction of a neural network are competitive with respect to a linear prediction approach.

3D numerical simulations of optical turbulence at the Roque de Los Muchachos Observatory using the atmospherical model Meso-Nh

“…Also we study the ability of the MesoNh model to simulate the temperature near the ground around the telescope. The results show that the difference between the measured and simulated temperatures is ∆T ≤ 3 degrees, which is, comparable to the results obtained with other techniques, such as multiregression (Buffa & Porceddu 1997). This is a very promising result considering that it was proven (Racine et al 1991) that to prevent seeing caused by outside temperature T 2m and the primary mirror temperature T M differences, the T 2m should be known with a precision of about 1 degree.…”

Artificial Intelligence in Astronomical Optical Telescopes: Present Status and Future Perspectives

“…Additionally, they proved that in the case of the fuzzy coding of seeing (dividing the seeing degree into Good, Moderate, and Bad), the forecasting accuracy of dynamic recurrent neural networks outperforms that of KNN methods. Buffa and Porceddu [89] studied the problem of forecasting observatory site temperatures and proved that the nonlinear autoregressive neural network model is more competitive than the traditional linear filtering algorithm.…”