[Paper Review] New achievements in optical turbulence forecast systems in operational mode

Elena Masciadri, Alessio Turchi+1 more|arXiv (Cornell University)|Nov 7, 2019

Remote Sensing in Agriculture4 citations

TL;DR

This paper presents an advanced operational forecast system, ALTA, for optical turbulence (OT) at the Large Binocular Telescope (LBT), using an auto-regression method that fuses real-time in-situ measurements with numerical model outputs (Astro-Meso-Nh) to achieve unprecedented forecast accuracy on 1–2 hour timescales. It further clarifies fundamental misconceptions about mesoscale vs. general circulation models (GCMs), demonstrating that mesoscale models outperform GCMs despite longer computation times, and refutes claims about GCMs being inherently more efficient or universally applicable without calibration.

ABSTRACT

In this contribution, we present the most recent progresses we obtained in the context of a long-term program we undertook since a few years towards the implementation of operational forecast systems (a) on top-class ground-based telescopes assisted by AO systems to support the flexible scheduling of observational scientific programs in night as well in day time and (b) on ground-stations to support free space optical communication. Two topics have been treated and presented in the Conference AO4ELT6: 1. ALTA is an operational forecast system for the OT and all the critical atmospheric parameters affecting the astronomical ground-based observations conceived for the LBT. It operates since 2016 and it is in continuous evolution to match with necessities/requirements of instruments assisted by AO of the LBT (SOUL, SHARK-NIR, SHARK-VIS, LINC-NIRVANA,...). In this contribution, we present a new implemented version of ALTA that, thanks to an auto-regression method making use of numerical forecasts and real-time OT measurements taken in situ, can obtain model performances (for forecasts of atmospherical and astroclimatic parameters) never achieved before on time scales of the order of a few hours. 2. We will go through the main differences between optical turbulence forecast performed with mesoscale and general circulation models (GCM) by clarifying some fundamental concepts and by correcting some erroneous information circulating recently in the literature.

Motivation & Objective

• To enhance short-term (1–2 hour) forecasting of optical turbulence and key astroclimatic parameters (seeing, isoplanatic angle, coherence time) at the LBT for adaptive optics scheduling.
• To address the lack of accurate, timely forecasts for flexible scheduling in both night and day operations at high-precision ground-based telescopes.
• To clarify and correct widespread misconceptions in the astronomical community regarding the comparative performance and applicability of mesoscale models versus general circulation models (GCMs) in optical turbulence forecasting.
• To validate the superiority of mesoscale models (e.g., Astro-Meso-Nh) over GCMs in short-term forecast accuracy despite longer computational times.
• To emphasize that model calibration is site-specific and that universal calibration remains unfeasible due to insufficient global observational data.

Proposed method

• The ALTA system integrates real-time in-situ measurements from a DIMM and a generalized SCIDAR with numerical forecasts from the Astro-Meso-Nh mesoscale model.
• An auto-regression technique is applied to correct and refine model outputs using current observational data, improving short-term forecast performance.
• The method leverages time-series filtering principles akin to Kalman filtering and machine learning, using a two-minute temporal resolution for raw data and re-interpolating to 20-minute intervals for visualization and analysis.
• Model performance is validated against long-term measurements (2005–2008) from the VATT telescope and nightly DIMM data at the LBT site.
• Comparisons between mesoscale (Astro-Meso-Nh) and GCM (OS18) forecasts are conducted across three distinct local time windows (14:00, 21:00, 03:00 LT), reflecting different forecast availability times.
• The study uses statistical metrics such as RMSE (0.54” for GCMs) and scattering plots to evaluate forecast accuracy and decorrelation levels.

Experimental results

Research questions

• RQ1 Can an auto-regression method combining real-time in-situ measurements with numerical model forecasts significantly improve short-term (1–2 hour) predictions of optical turbulence and related astroclimatic parameters at the LBT?
• RQ2 Why do mesoscale models like Astro-Meso-Nh outperform GCMs in short-term optical turbulence forecasting despite longer computation times?
• RQ3 What are the fundamental differences in model physics and application between mesoscale and GCM-based optical turbulence forecasts, and how do these affect forecast reliability?
• RQ4 Is it accurate to claim that GCMs are inherently more efficient or universally applicable than mesoscale models, especially in the context of astronomical site testing and operations?
• RQ5 Can a universal calibration for optical turbulence models be achieved, or is site-specific calibration still the only viable approach with current observational data?

Key findings

• The auto-regression method applied to ALTA significantly improves forecast performance for optical turbulence and key astroclimatic parameters (seeing, isoplanatic angle, coherence time) on 1–2 hour timescales, achieving previously unmatched accuracy.
• Despite longer computation times, mesoscale models (Astro-Meso-Nh) produce more accurate short-term forecasts than GCMs (OS18), as evidenced by lower RMSE and reduced decorrelation in scatter plots.
• The GCM forecast shown in the study is a composite of outputs available at different times (14:00, 21:00, and 03:00 LT), which invalidates fair comparisons with models that have consistent forecast availability.
• The claim that GCMs are preferable due to shorter computation times is fundamentally misleading, as the study shows that model accuracy is not determined by calculation speed.
• The belief that mesoscale models require site-specific calibration while GCMs do not is incorrect—both require validation, and model reliability depends on performance, not spatial extent.
• Universal calibration of optical turbulence models remains unfeasible due to insufficient global, heterogeneous observational data, and all current methods rely on site-specific calibration.