Remote sensing analysis of small scale dynamic phenomena in the Atmospheric Boundary Layer

The aim of this thesis project was to study the coherent turbulent structures (convective rolls & streaks) and more generally the medium to large fluctuations in the atmospheric boundary layer using the observations recorded by a single Doppler (wind) lidar during a 2-month campaign in Paris, France. An innovative method was developed in order to classify automatically the radial wind speed patterns visible on the quasi-horizontal lidar scans, based on texture analysis parameters and supervised machine learning algorithms. A 150-case training ensemble was built using ancillary data (satellite pictures and weather observations) to ascertain the manual classification into four types: rolls, thermals, streaks and “others”. The performance of the classification process was assessed on the training ensemble using the 10-fold cross-validation method. A very satisfying 9% error was obtained for the Quadratic Discriminant Analysis algorithm, using only 5 texture analysis parameters classifiers. This process was then applied to classify the whole dataset (4577 lidar scans) and the results showed that the classified structures respected a plausible diurnal cycle and were associated with the meteorological parameters as expected by the theoretical knowledge. The size of the coherent structures in the direction transverse to the mean wind were estimated from the wind spectrums on a four-day case study. They ranged from 400 to 800 m for the cases classified as streaks, and from 1.3 to 2.0 km for the cases classified as rolls. These results pave the way for future long-term studies providing statistical insight on the frequency of occurrence of the different structure types, their physical properties, and their impact on pollutants’ concentrations.