Speaker; Francis Ofori

Title: The Impact of Weather Change on Nitrous Oxide Emission with 
Spatial Pattern Detection and Large Data Approximation

Abstract:

Among the ecosystems and fields affected by climate change associated with increasing greenhouse gas concentration, agriculture has been a focus due to its vital connection with climate and food supply. It could have substantial implications for the economy and agricultural management to study the detection of spatial pattern in regional climate change and the impact of weather change on greenhouse gases, specifically nitrous oxide (N_2 O ) emission of state crops.
To capture the spatial pattern of significant regional climate change, a Process-based Geographical Algorithm Machine (PGAM) procedure is proposed by viewing the spatio-temporal data sets as realizations of underlying random fields. Where the past and future climate scenarios of daily weather are simulated using multiple Global Circulation Models (GCM). The simulation differences and consistency of precipitation, minimum and maximum temperature in the state of Kansas produced by these climate models are also assessed using PGAM along with spatial Kolmogorov-Smirnov test. Then the climate change index described by a temporal distance metric resulted from PGAM is used to study the adverse effect on N_2 O emission connected with agricultural management practices based on linear mixed effect model. 
This project further delves into the effect of weather change on N_2 O emission with large data approximation. Encountering the computational challenge in statistical inference accounting for the autocorrelation that exhibits in both weather and N_2 O emission data, we intend to use some approximation methods to mitigate the computational burden. However, computational reduction must be well defined without sacrificing much of the accuracy of the result. With this general problem in mind, dealing with large or moderately large spatial data sets, we theoretically and numerically investigate the conditions under which the computational intensity and prediction accuracy can be balanced by adopting projection approach. An optimal rank selection method is proposed to achieve good accuracy in terms of Kullback-Leibler divergence and mean square prediction error while alleviating the computational cost. The accuracy and performance of the proposed method are evaluated via simulation and spatial regression analysis of N_2 O emission.