Machine Learning methods for classification of Acid Sulfate soils in Virolahti
Estévez Nuño, Virginia (2020)
Estévez Nuño, Virginia
2020
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2020052915446
https://urn.fi/URN:NBN:fi:amk-2020052915446
Tiivistelmä
Acid Sulfate (AS) soils are among the most dangerous soils naturally occurring soils. This is due to the several ecological damages that they can generate. In Europe, the highest concentration of this type of soils is located in Finland. This represents one of the major environmental problems of the country. To solve this issue, it is essential the localization of the areas where these soils appear, and try to avoid their exposition to oxidizing conditions. Despite of the effort done during the last decade, there are hardly any AS soils mapping done in the country. The main reason is that the traditional methods used for AS soil mapping are very laborious and take a long time. Nowadays, thanks to new technologies a large amount of data is generated in soil science. As a result, several machine learning techniques can be used for the classification of AS soils. The use of these techniques will streamline the process and improve the accuracy of the AS soils mapping. The study of this master thesis has focused on the evaluation of different machine learning techniques for the AS soils classification in a small region of Finland. The goal is to find suitable methods for this. Random forest, Gradient boosting, Support vector machine and a Convolutional neural network have been analyzed in detail. The study area corresponds to Virolahti, which is located in the south-east of Finland. Two different datasets have been created, one for the case of deep learning and another for the rest of the methods. The results show that gradient boosting and random forest are very good methods for the classification of AS soils, whereas support vector machine is not very good. The convolutional neural network gives poor results, may be due to the small size of the dataset created.