Files
Abstract
Saltmarshes are capable of providing plants and animals with the biologically desirable environment while sequestering heavy metals. Saltmarshes, which have the highest value of net primary productivity among terrestrial and wetland ecosystems, are adversely impacted and threatened by sea-level rise and anthropogenic activity. Any serious investigation of the prevention or restoration practice of saltmarshes would be strengthened by well-established baselines characterizing vegetation, soil, and water in these coastal areas. In this study, field and laboratory tests were conducted for soil and porewater samples at saltmarsh sites in coastal Georgia (USA). A random forest (RF) model identified correlation among saltmarsh predominant vegetation types, redox potential, and salinity. This model confirmed that moisture content and sand content are two main drivers for the bulk density of saltmarsh soils, which directly affect plant growth and likely root development. Remote sensing has been widely used as a reliable technique to characterize soils from field to space and has been shown to be an effective approach to model soil properties at a broad scale. This dissertation presents machine learning algorithms such as RF and extreme gradient boosting machines (XGBoost) as well as super vector machines (SVM) and their application to model soil bulk density at saltmarsh sites along Georgia’s Atlantic coast, USA. For this study, Landsat-7 Enhanced Thematic Mapper Plus (ETM+) bands (excluding band6) were used as independent variables, and band 1 and band 4 were ranked as the most important attribute for bulk density prediction by XGBoost and RF, respectively. The results reveal that XGBoost algorithm has a higher accuracy (=0.88) than RF and SVM. Further, this study concludes that XGBoost requires less processing time in comparison with SVM in terms of the number of the tuning hyperparameters. SVM is insensitive to enlarging the dataset but XGBoost shows an increase in validation score as the dataset becomes larger.