Gentle, James E.Lisic, Jonathan James2016-04-192016-04-192015https://hdl.handle.net/1920/10182The purpose of this research is to develop and study the methodology and the underlying theory for prediction of agricultural land cover for a set of commercial crops at the parcel level. The observational unit will be called a land cover unit (LCU). Each LCU will have an associated land cover sequence. A land cover sequence is an ordered set of known categories indexed by a set of fixed consecutive years. An LCU is the maximally contiguous section of land with respect to a single land cover sequence, not transected by public transportation arteries or permanent hydrographic boundaries. LCUs are not observed, instead they are estimated through an image segmentation algorithm known as mean shift, applied to high resolution imagery products. The predictors of land cover are constructed under the assumption of temporal stationarity; this assumption limits the length of the land cover sequence that can be used to aid in the prediction. Land cover sequences for the LCUs are estimated from classified pixel level data. Prediction of future land cover is performed through a Bayesian hierarchical multinomial probit model accounting for spatially correlated crop rotation preferences. Application to major commercial crops in La Porte County, Indiana, is provided using high resolution imagery and thematic maps from the United States Department of Agriculture. The theory and methods are applicable to prediction of agricultural crops in other areas with a relatively stable pattern of agricultural land cover.150 pagesenCopyright 2015 Jonathan James LisicStatisticsAgricultureComputer scienceCrop RotationImage SegmentationLand CoverMean ShiftMultinomial ProbitPredictionDissertation