Unit:
Κατεύθυνση Στατιστική και Επιχειρησιακή ΈρευναLibrary of the School of Science
Author:
Giannadaki Christina
Supervisors info:
Τρέβεζας Σάμης, Λέκτορας , Τμήμα Μαθηματικών, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Original Title:
Statistical techniques for improving prediction in crop progress stages with meteorological and satellite data
Translated title:
Statistical techniques for improving prediction in crop progress stages with meteorological and satellite data
Summary:
Crop Progress Reports (CPRs) of the USDA are listing the weekly progress made in the different phenological stages of selected crops and in particular of corn. In this thesis, our goal was to predict the CPRs of a full year by taking into account available data from related features in a way that we can beat the predictions based on empirical means from historical data. For this reason, we used two features, the mean Normalized Difference Vegetation Index (NDVI) and the Accumulated Growing Degree Days (AGDDs). In order to achieve our target we implemented several modeling approaches, including Independent Mixture Models and Hidden Markov Models HMMs and we compared different type of estimators and predictors by taking into account both features or treating them separately, or making data transformations, such as differences. The results showed that the aforementioned models cannot predict better than the historical data. Finally, we managed to obtain better predictions by using Simple Linear Regression. This study can be extended in several directions for future work.
Main subject category:
Science
Keywords:
Data analysis, corn, meteorological data, satellite data
