11 research outputs found
Charting the Realms of Mesoscale Cloud Organisation using Unsupervised Learning
Full text linkQuantifying the driving mechanisms and effect on Earth's energy budget, of
mesoscale shallow cloud organisation, remains difficult. Partly because
quantifying the atmosphere's organisational state through objective means
remains challenging. We present the first map of the full continuum of
convective organisation states by extracting the manifold within an
unsupervised neural networks's internal representation. On the manifold
distinct organisational regimes, defined in prior work, sit as waymarkers in
this continuum. Composition of reanalysis and observations onto the manifold,
shows wind-speed and water vapour concentration as key environmental
characteristics varying with organisation. We show, for the first time, that
mesoscale shallow cloud organisation produces variations in albedo
in addition to variations from cloud-fraction changes alone. We further
demonstrate how the manifold's continuum representation captures the temporal
evolution of organisation. By enabling study of states and transitions in
organisation (in simulations and observations) the presented technique paves
the way for better representation of shallow clouds in simulations of Earth's
future climate
Recommended from our members
Using high-resolution modelling to improve the parameterisation of convection in a climate model
Get PDFIn this work high-resolution numerical simulation (Large-Eddie Simulation, LES) has been used to study the characteristic factors causing and influencing the development of moist convective clouds. Through this work a 1D cloud-model was derived from first principles to represent the vertical profile of individual convective clouds. A microphysics framework was implemented to ensure identical behaviour in LES and cloud-model integration where the microphysical processes represented are numerically integrated using a novel adaptive step microphysics integration which uses the physical speed at which a process takes place to adjust the integration step size (in space and time). This work also introduces a simple representation of cloud-droplet formation which allows for super-saturation to exist in-cloud and through this provide more physical representation of the in-cloud state.
Together with high-resolution simulation of isolated individual and interacting multiple clouds in environmental conditions leading to shallow convection, the 1D cloud-model was used to infer that the principal influence on moist convective clouds is the entrainment of air from a cloud’s immediate environment which is significantly different from the environmental mean state. This suggests that convection parameterisations must represent the influence of moist convective downdrafts to properly predict the vertical structure of convective clouds so as to correctly predict the cloud-top height and vertical transport. Finally it was found that cloud-base radius is not in itself adequate as a means of classification for defining cloud-types as clouds with the same cloud-base radius showed large variation (≈ 600m) in cloud-top height. Based on simulations of individual convective clouds it was found that 3D simulations are necessary to capture the full dynamic behaviour of convective clouds (2D axisymmetric simulations have too little entrainment) and that agreement with the 1D cloud-model could only be found when entrainment was diagnosed from simulation instead of being parameterised by the traditional Morton-Turner model and only for 2D axisymmetric simulations, suggesting that the 1D cloud-model will require further extension or the diagnosis of entrainment improved
Can Recurrence Quantification Analysis Be Useful in the Interpretation of Airborne Turbulence Measurements?
Get PDFIn airborne data or model outputs, clouds are often defined using information about Liquid Water Content (LWC). Unfortunately LWC is not enough to retrieve information about the dynamical boundary of the cloud, that is, volume of turbulent air around the cloud. In this work, we propose an algorithmic approach to this problem based on a method used in time series analysis of dynamical systems, namely Recurrence Plot (RP) and Recurrence Quantification Analysis (RQA). We construct RPs using time series of turbulence kinetic energy, vertical velocity and temperature fluctuations as variables important for cloud dynamics. Then, by studying time series of laminarity (LAM), a variable which is calculated using RPs, we distinguish between turbulent and non-turbulent segments along a horizontal flight leg. By selecting a single threshold of this quantity, we are able to reduce the number of subjective variables and their thresholds used in the definition of the dynamical cloud boundary
Text entry performance of state of the art unconstrained handwriting recognition: a longitudinal user study
No full textWe report on a longitudinal study of unconstrained handwriting recognition performance. After 250 minutes of practice, participants had a mean text entry rate of 24.1 wpm. For the first four hours of usage, entry and error rates of handwriting recognition are about the same as for a baseline QWERTY software keyboard. Our results reveal that unconstrained handwriting is faster than what was previously assumed in the text entry literature. Author Keywords Handwriting, handwriting recognition, software keyboar
Continuous recognition and visualization of pen strokes and touch-screen gestures
No full textWe present a technique that enables continuous recognition and visualization of pen strokes and touch-screen gestures. We describe an incremental recognition algorithm that provides probability distributions over template classes as a function of users ’ partial or complete stroke articulations. We show that this algorithm can predict users ’ intended template classes with high accuracy on several different datasets. We use the algorithm to design two new visualizations that reveal various aspects of the recognition process to users. We then demonstrate how these visualizations can help users to understand how the recognition process interprets their input and how interactions between different template classes affect recognition outcomes
