Proposal Description

Forecasting and spatial extrapolation are core statistical and machine learning goals for environmental data. Existence of high dimensionality and non-traditional spatiotemporal autocorrelation in the data make it crucial to develop models that can provide accurate prediction while being computationally feasible. At the same time, measures of uncertainty are necessary for decision makers. In this session, there will be a series of talks by speakers who address such challenges in agricultural and ecological applications. The session will showcase a range of methods with data applications in agriculture and ecology. There will be a contrast of methods from deep learning to dimension reduction. Specifically, speakers intend to showcase machine learning methods for spatial disease ecology, species distribution modeling, and camera trap image analysis. Both the disease ecology and species distribution modeling applications entail predicting a dynamic spatial process with non-stationarity. In agriculture, applications will highlight prediction of spatial soil moisture extremes and agriculture production. Furthermore, it will be shown that employing dimension reduction can lead to substantial improvement in the efficiency of the estimated parameters specially for cases with limited sample size. Speakers in this session are early career. This session is jointly organised by Dr Hossein Moradi Rekabdarkolaee and Dr. Toryn Schafer.