A Rule Based Machine Learning Approach to Vegetation Classification using Remote Sensing Data

Abstract

The field of Vegetation Classification and Mapping is of paramount importance for monitoring vegetation levels, but has unfortunately resulted in trivial accuracy. Although Vegetation Mapping through the usage of Remote Sensing data has proven to be an effective tool, Vegetation Classification has struggled to be accurate as the traditional parametric Vegetation Classification methods are ineffective. This research proposes to utilize Machine Learning to classify vegetation by utilizing vegetation indices, Sentinel 2 and GEDI relative height data. A problem faced by Machine Learning models in Vegetation Classification is the inefficiency of manually collecting training data. However, this research proposes to incorporate rule based automated training data to feed to the machine learning model, rather than traditionally generating training samples manually. The spectral bands in Sentinel 2 data are utilized to accurately calculate vegetation indices, which are used as parts of the training data. Moreover, Sentinel 2 data can fill gaps in GEDI shots, generating additional training data. The high accuracy score of the model’s ability to calculate vegetation height and subsequently use it classify vegetation is demonstrative of the importance of Machine Learning with vegetation classification. Additionally, the model has the capability of reducing external weather conditions from cluttering it, making it extremely useful for future work. Ultimately, this research suggests for further investigation of Vegetation Classification using Machine Learning as a viable methodology.