Aristotle University Of Thessaloniki (AUTH)
Προσδιορισμός του συντελεστή διαβρωτικότητας της βροχόπτωσης στα πλαίσια της επιστήμης των δεδομένων με μεθόδους μηχανικής μάθησης και γεωστατιστικής
Abstract
dc:descriptionThe subject of the thesis concerns the determination of rainfall erosivity factor R of the Universal Soil Loss Equation for the totality of the Hellenic territory. The methodology of the research falls within the framework of Data Science and concerns the following: 1) The import into a tidy database of recording and non-recording rain gauge time series data and a terrain model, after cleaning from errors. 2) The location within recording gauge time series of erosion-producing stroms. and the evaluation of their erosivity. 3) The transformation of data, the computation of new variables and of the requisite statistical quantities. 4)The use of exploratory data analysis for the analysis of the data sets to summarize the main characteristics of R and EI with visual methods. 5) The evaluation of algorithms using methods of cross-validation and Inductive Statistics for the imputation of EI values and for the estimation of R in new locations. 6) The creation of maps for the estimation of R and of the corresponding standard error.For the processing and analysis, the following data were utilized: 1) The recording and non-recording rain gauge data series of the meteorological measuring stations of the Ministry of Environment, Energy and Climatic Change as archived in the Hydroscope Database. 2) The yearly average rainfall measurement values of the National Meteorological Service, of the research project GEOCLIMA and of the Dutch Royal Meteorological Institute. 3) The world digital terrain model SRTM of NASA.The analysis of the recording rain gauge time series revealed a significant percentage of missing data. There it was deemed necessary to proceed to a systematic imputation of the erosivity EI values, so that R would not be underestimated, given that the latter is computed on the basis of the yearly EI sums for every station. The imputation was done using non-recording rain gauge time series of the same stations and it was evaluated by a set of Machine Learning and Numerical Optimization algorithms. It turned out that best imputation result was provided by a Neural Network algorithm, the optimal results of which were selected by means of a Genetic algorithm.In order to estimate R from the point location measurements of the meteorological stations to the totality of the Hellenic territory, use was made of the imputed EI values. On the basis of this processed data a set of Machine Learning, Numerical Optimization and Geostatistics algorithms was applied and evaluated. Finally, the best result was produced by a Gaussian Geostatistics model for the spatial estimation of erosivity. This model included a nonlinear trend of the mean yearly rainfall.Lastly, the improvement of the estimation of the R factor was evaluated, using an algorithm based on Machine Learning methods of simultaneous imputation of the EI values and spatial estimation of R, against the use of empirical equations. Through this evaluation, it has been show the superiority of Machine Learning methods and that the imputation of the EI values is essential in order to estimate R values under incomplete rainfall data.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Vantas, Konstantinos
- Βάντας, Κωνσταντίνος
Subjects
dc:subject × 24- Συντελεστής διαβρωτικότητας βροχής
- Μηχανική μάθηση
- Νευρωνικά δίκτυα
- Γεωστατιστική
- Επιστήμη δεδομένων
- Αριθμητική βελτιστοποίηση
- Rainfall erosivity factor
- Machine learning
- Neural networks
- Geostatistics
- Data science
- Numerical optimization
- Περιβαλλοντική Μηχανική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη και Τεχνολογία υδάτων
- Μαθηματικά
- Φυσικές Επιστήμες
- Μοντελοποίηση και Προσομοίωση
- Environmental Engineering
- Engineering and Technology
- Water Science and Technology
- Mathematics
- Natural Sciences
- Modeling and Simulation
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/41373