Correlating Lidar range performance with atmospheric condition parameters

Correlating Lidar range performance with atmospheric condition parameters

Cristina Benzo, Ludovic Thobois, Maxime Hervo

Lidars have been increasingly integrated in various applications for its unique ability to measure continuously along multiple range gates for several kilometers. Lidars provide a significant advantage in meteorological networks as it extends beyond the measurement range of an anemometer to continually acquire wind speeds all throughout the boundary layer. Other applications such as wind field measurements for wind farm preventative energy management or airport shear detection highly depend on a lidar’s range of measurement to accurately visualize and forecast critical changes in wind speed. The ability of a lidar to measure several kilometers provides a unique advantage, yet the precise distance of its range capability greatly depends on atmospheric conditions. As a remote sensor, the performance of the return signal depends on the state of the atmosphere, which is based on extinction, backscatter, temperature, and more. Though a lidar range performance can be physically defined based on its signal processing qualities, the precise relationship of lidar range performance and atmospheric conditions is still unknown. Thus, this study aims to clarify and quantify this atmospheric relationship.

This study aims to rank the most impactful environmental variables on lidar range performance based on different regression and predictor techniques, as well as provide a first attempt at modelling range performance using machine learning and these environmental parameters. The variables considered in this analysis are lidar range, visibility, backscatter, PM2.5, relative humidity, AOD, temperature, and hour of the day.

The methods used in this analysis to determine correlation included a binned regression technique, to f-test statistical testing, and a random forest machine modeling technique. The binned regression technique was conducted by binning both the range and environmental variables to eliminate noisy conditions, visualize averaged patterns of range performance and environmental conditions, and conduct a regression fit to define the pattern between the two. The F-test provides a level of significance to each variable’s influence on range. Finally, a random forest model was trained and tested to predict lidar range performance based on several atmospheric measurements. Though lidar performance is not solely based on atmospheric conditions, the results showed promise and potential to combine with a physics-based model to account for overfitting and improve the loss function.

The results of the study yielded greater insights into the relationship between range and several atmospheric variables, as well as potential into how lidar range could be modelled based on environmental variable measurements. Backscatter and extinction demonstrated the highest impact on range performance across all methods applied, which follows the physical properties of the lidar’s carrier-to-noise ratio. With further development, predicting a lidar’s range performance could help users to plan and predict their measurement acquisition more efficiently.