Π-ML: a dimensional analysis-based machine learning parameterization of optical turbulence in the atmospheric surface...

Area of Science:

• Physics
• Atmospheric Science
• Optical Engineering

Background:

• Optical turbulence, caused by atmospheric refractive index fluctuations, significantly distorts laser beams.
• Accurate modeling of optical turbulence strength (Cn2) is crucial for reliable free-space optical (FSO) communication systems.

Purpose of the Study:

• To propose a novel physics-informed machine learning (ML) methodology, named Π-ML, for estimating Cn2.
• To identify key atmospheric parameters influencing Cn2 through feature importance analysis.

Main Methods:

• Utilized dimensional analysis and gradient boosting for the ML model.
• Employed an ensemble of models for enhanced statistical robustness.
• Conducted systematic feature importance analysis to identify predictive drivers of Cn2.

Main Results:

• Identified normalized variance of potential temperature as the most dominant feature for predicting Cn2.
• Achieved high out-of-sample performance with an R² value of 0.958 ± 0.001.
• Demonstrated the effectiveness of the Π-ML approach in modeling optical turbulence.

Conclusions:

• The proposed Π-ML methodology provides an accurate and robust method for estimating optical turbulence strength.
• This advancement is vital for the successful development and deployment of future FSO communication links.
• Highlighting the importance of potential temperature variance offers new insights into atmospheric turbulence modeling.