Data-driven methods are gaining popularity for predicting complex turbulent flow systems. However, fundamental questions about the effectiveness of these methods under practical situations remain unanswered. Currently, low-order simplified models (such as Mann model for wind field prediction) are used in industry. These models are quick even under challenging conditions and work with limited measurements. However, their accuracy is generally low. Can we exploit the increase in computational resources and use data-driven methods, such as classical Bayesian inference-based data assimilation techniques or recently developed deep learning networks? The candidate will have state-of-the-art GPU computational facilities and will work with other group members to tackle the most challenging problems in time-series forecasting with wide-ranging applications from wind energy to drag reduction and financial markets.