Software and datas for "Turbulent coherent structures in the atmospheric surface layer revealed by texture analysis and supervised classification of Doppler lidar observations"

Abstract:

Coherent turbulent structures such as streaks are essential components of the surface layer dynamics, impacting turbulent fluxes and pollutants' dispersion.  These structures can be observed using horizontal scans from a Doppler lidar, after extracting the turbulent component of the radial wind. A 13-month campaign took place in Dunkerque between 2021 and 2022 (northern France, on the North Sea coast) during which 40,000 quasi-horizontal scans were recorded. Two types of coherent structures were identified by eye: organized and disorganized streaks (the absence of structures falling into a third category called "others"). An automated classification method was developed, as manually classifying such a large number of lidar images would be impossible. For each image, a set of third-order numerical parameters, or features, was generated in three steps: first co-occurrence matrices, then texture parameters, and finally 'curve parameters'.  A training set consisting of 400 scans was built and used to train four supervised learning classification algorithms. The Quadratic Discriminant Analysis provided the lowest classification error, estimated at 5.2%, using the 10-fold validation method. A few physical parameters such as time of the day and wind direction were added to the features but were not selected by any algorithm as a most informative feature. The algorithm detected coherent structures on 61.0% of the whole set of lidar images and successfully distinguished organized and disorganized streaks. The preliminary analysis of the classification results showed a predominance of organized streaks during the daytime and disorganized streaks during the nighttime.

Funding and acknowledgments:

This project is co-funded by the CaPPA Labex and by the Région Hauts-de-France. CaPPA is funded by the French National Research Agency (ANR) through the PIA and under the contract ANR-11-LABX-0005-01, and by the Région Hauts-de-France and the European Regional Development Fund (ERDF). The lidar instrument was funded through the CPER IRenE, by the Région Hauts-de-France and the ERDF. The authors thank the Région Hauts-de-France, the Ministère de l’Enseignement Supérieur et de la Recherche and the European Fund for Regional Economic Development for their financial support to the CPER ECRIN program.
The authors also thank the Halle aux Sucres Learning Center for hosting the lidar instrument. Experiments presented in this paper were carried out using the CALCULCO computing platform, supported by SCoSI ULCO (Service COmmun du Système d’Information de l’Université du Littoral Côte d’Opale). This work is supported by the Moscow Center of Fundamental and Applied Mathematics at INM RAS in the framework of the agreement with the Ministry of Education and Science of the Russian Federation No. 075-15-2022-286.