Files
Abstract
This dissertation is composed of three studies that evaluated the use of Landsat time series data for the detection of forest disturbance and estimation of the growing stock volume of the forest stands in the state of Georgia, United States.
We explored the use of Landsat time series dataset to estimate the disturbance year of the forest and estimated the growing stock volume of forest stands using both Landsat time series and the disturbance year data created using the Landsat time series.
The first study is the preliminary research that is dedicated to creating a model that detects the year of forest disturbance. Landsat time series data and an algorithm were used to estimate the forest disturbance. We selected 7 counties along coastal Georgia as our study area. The main objective of this work was to address the current age-class forest structure in the study area where the pine plantation forest is intensively managed. As an outcome of the study, a disturbance detection map was created. We performed an accuracy assessment for the disturbance map and acquired 52% of overall accuracy, which is acceptable precision.
The second study enlarged preliminary research performed in the first study. The two primary goals of this study are (i) to establish and examine a reliable framework for Georgia's state-wide monitoring of forest disturbances; (ii) to consider and discuss the use and effect of information on forest disturbance maps. From the first study, we changed our area of interest to the entire state and used all available Landsat time series datasets. The overall accuracy of the data for the year of disturbance was more than 8%. This result is 35 percent higher than the overall accuracy of the first study.
In the third study, We have established a model of random forest regression that estimates the increasing volume of forest growing stock volume.
The goal of this work was to to test whether (i) using all available Landsat imagery and (ii) application of the bias correction approach would improve the accuracy of the estimation of the forest stand growing stock volume.
We used the Forest Inventory and Analysis dataset that is maintained by the US Forest Service as a field plot data. We used Random Forest as an estimation method. We obtained 65% of relative RMSE in the best model.
These studies demonstrate the importance in the application of Landsat time series dataset to spatially explicit forest inventory. The outcome created in these studies is expected to provide fundamental information of the forest resources in Georgia to allow better decisions for regional scale forest management and conservation.
We explored the use of Landsat time series dataset to estimate the disturbance year of the forest and estimated the growing stock volume of forest stands using both Landsat time series and the disturbance year data created using the Landsat time series.
The first study is the preliminary research that is dedicated to creating a model that detects the year of forest disturbance. Landsat time series data and an algorithm were used to estimate the forest disturbance. We selected 7 counties along coastal Georgia as our study area. The main objective of this work was to address the current age-class forest structure in the study area where the pine plantation forest is intensively managed. As an outcome of the study, a disturbance detection map was created. We performed an accuracy assessment for the disturbance map and acquired 52% of overall accuracy, which is acceptable precision.
The second study enlarged preliminary research performed in the first study. The two primary goals of this study are (i) to establish and examine a reliable framework for Georgia's state-wide monitoring of forest disturbances; (ii) to consider and discuss the use and effect of information on forest disturbance maps. From the first study, we changed our area of interest to the entire state and used all available Landsat time series datasets. The overall accuracy of the data for the year of disturbance was more than 8%. This result is 35 percent higher than the overall accuracy of the first study.
In the third study, We have established a model of random forest regression that estimates the increasing volume of forest growing stock volume.
The goal of this work was to to test whether (i) using all available Landsat imagery and (ii) application of the bias correction approach would improve the accuracy of the estimation of the forest stand growing stock volume.
We used the Forest Inventory and Analysis dataset that is maintained by the US Forest Service as a field plot data. We used Random Forest as an estimation method. We obtained 65% of relative RMSE in the best model.
These studies demonstrate the importance in the application of Landsat time series dataset to spatially explicit forest inventory. The outcome created in these studies is expected to provide fundamental information of the forest resources in Georgia to allow better decisions for regional scale forest management and conservation.