<p> We report the observation of a net inward, up-gradient turbulent particle flux which occurs when a collisional drift waves generate a sufficiently strong radially sheared azimuthal zonal flow in a cylindrical magnetized plasma. At low magnetic fields (B&le;1.0 kG), particle transport is outward at all radii. As the magnetic field is further increased to 1200G, an up-gradient inward particle flux develops between the peak of the velocity shear and the maximum density gradient. The mean density gradient is also observed to steepen in response to this inward flux. Time-domain and bispectral Fourier domain analysis shows that at the peak of the velocity shear, where the particle flux is outward, the turbulent Reynolds stress acts to reinforce the shear flow. In contrast, in the region of the inward particle flux, the zonal flow drives the fluctuations, and a transient increase in the shearing rate is occurs prior to an increase in the magnitude of the inward flux. The results suggest a hypothesis in which the shear flow is responsible for the up-gradient particle flux and the corresponding steepening in the mean density gradient. However, a linear instability analyses using experimentally measured density and E&times;B flow profiles in a linear, modified Hasegawa-Wakatani theory model with the coupled potential and density fluctuations failed to reproduce the essential elements of our experimental observations, suggesting some other mechanism is responsible for the inward flux. We summarize recent new experimental results which point towards the possible role of finite ion temperature gradient effects, possibly combined with parallel flow shear, in driving up-gradient particle flux.</p>

The simulation of the atmospheric turbulence has wide applications in science and engineering. There are a variety of modelling techniques for synthesizing a wind speed time series available today. To decide which model to use one has to know the characteristics of the simulation technique. In this dissertation, alternative approaches for synthetically generating a wind speed time series are discussed. These approaches include: (1) the use of independent values from a specific probability distribution; (2) the use of an algorithm based on the statistical behavior of a one step Markov chain; (3) the use of an algorithm based on the behavior of a transition probability matrix that describes the next wind speed value statistically as a function of the current wind speed value and the previous wind speed value; (4) the use of Box-Jenkins model; (5) the use of the Shinozuka algorithm; (6) the use of an embedded Markov chain; and (7) the use of the fractal concept. The ability of each approach to capture the statistical properties of the desired wind speed time series is discussed. Wind speed collected at Windsor, Massachusetts and at Altamont, California are used as target wind speed values. Each model will be used to generate a synthetic wind speed for each site to compare with the real wind speed. The performance of each model will be decided on by the statistical similarity of the synthetic wind speed to the real wind speed. The criteria of the statistical similarity include the mean and the variance of the wind speed values, the probability distribution of the wind speed values, the power spectrum of the wind speed values and the autocorrelation function of the wind speed values. One of the applications of the wind speed simulation is to fill in a missing segment of a wind speed time series. In this context the missing segment of the wind speed at Cuttyhunk island is simulated and filled in for the study of a wind/diesel energy conversion system.