From Earth to Sky and Beyond: Sampling for Scientific and Industrial Innovation

Accurate modeling and forecasting are essential for decision making in agriculture and atmospheric sciences. First, a stochastic weather generator trained on more than 60 years of hourly data produces synthetic climate scenarios that drive a deterministic wheat growth model to produce probabilistic distributions of average and extreme crop outcomes. A web application allows users to upload regional weather data and obtain customized forecasts for risk management and optimization. Second, size distributions of atmospheric particles are reconstructed from scattering and backscattering lidar measurements at three wavelengths through a Bayesian log-normal mixture framework. Reversible jump MCMC and sequential Monte Carlo methods jointly infer the posterior spectrum of size distributions, quantifying the uncertainty in this ill-posed inverse problem. Together, these studies demonstrate the value of probabilistic scenario generation and Bayesian inference as robust tools for managing environmental variability in both crop productivity and aerosol characterization.