10:45 AM - 12:15 PM
[ACG39-P04] Forest above-ground biomass estimation using national forest inventory with airborne LIDAR, SAR data and optical satellite data in Taiwan.
Keywords:above-ground biomass, machine learning, remote sensing, Taiwan
Introduction.
Forests have an important impact on ecosystem processes including carbon and water cycling and energy fluxes, which in turn affect local and regional climate. Understanding the dynamics of the forest's above-ground biomass (AGB) is essential to grasp how managing the forest affects climate change. The use of remote sensing data can provide spatially explicit information of AGB from local to global scales. Currently, AGB estimation via remote sensing has been widely used due to its promising results. In previous studies, AGB has been estimated using various prediction methods and types of remote sensing data. However, due to the limitations related to either optical, SAR or Lidar data alone, combining those data types is indispensable. The method by combining the advantages of active and passive data sources will improve the accuracy of aboveground forest biomass estimation. In this study, we combined the multiple sources data (Lidar, SAR, Optical satellites, National Forest inventory data) to estimate AGB in Taiwan. We also constructed and compared the accuracies of four different prediction methods (support vector regression (SVR), multi-layer perceptron neural network (MLPNN), K-nearest neighbor (KNN), and random forest (RF)) for estimating AGB of forests in Taiwan. Finally, mapping Taiwan forest AGB using satellite remote sensing data and a good fit machine learning approach.
Materials and Methods.
- Study area: Study area are conducted in Taiwan, which is a mountainous island of 3,598,000ha off the southeast coast of Mainland China and north of the Philippines.
- Field reference data: In this study, total 3486 plots were collected with the plot size was set to 0.05ha, but if a plot was found to include three trees with DBH > 100cm, its plot size was expanded to 0.1ha. In those plots, there are 1564 plots were obtained from the 4th National Forest Resource Inventory in Taiwan (TNFRI4) between 2008 and 2012, plots were provided by the Taiwan Forest Bureau, and 1922 plots were obtained from results of MOST projects. In addition, the forest classification data obtained from Council of Agriculture Executive Yuan Taiwan in 2017. There are 4 main types of forest include: broadleaved forest, coniferous forest, bamboo forest and mix forest. Each of forest type were conducted the different model between plots data and remote sensing data.
- Remote sensing data: Lidar CHM for Taiwan at a spatial resolution of 20m was supplied by Department of Geomatics National Cheng Kung University. Optical data are used in form of spectral reflectance mosaic based on Landsat 5 and 7 ETM+ data. The SAR data were collected and processed by Japan Aerospace Exploration Agency (JAXA). Optical data and SAR data was collected in 2012 (same year with the TNFRI4).
- Method: There are five major steps were followed to build the AGB map in Taiwan: (1) predictors data extraction (LiDAR, SAR, Optical satellite...), (2) training and testing datasets processing, (3) using learning marching (SVR, MLPNN, KNN and RF) to build AGB estimation model (4) model calibration and assessment, and (5) Using the good fit model to mapping AGB in Taiwan.
Expected result.
- Techniques of AGB estimation are improved by using various prediction methods and combining multiple sources of remote sensing data.
- Comparative analysis AGB models from different remote sensing data sources.
- The estimated AGB maps showed similar goodness-of-ft statistics in Taiwan.
Forests have an important impact on ecosystem processes including carbon and water cycling and energy fluxes, which in turn affect local and regional climate. Understanding the dynamics of the forest's above-ground biomass (AGB) is essential to grasp how managing the forest affects climate change. The use of remote sensing data can provide spatially explicit information of AGB from local to global scales. Currently, AGB estimation via remote sensing has been widely used due to its promising results. In previous studies, AGB has been estimated using various prediction methods and types of remote sensing data. However, due to the limitations related to either optical, SAR or Lidar data alone, combining those data types is indispensable. The method by combining the advantages of active and passive data sources will improve the accuracy of aboveground forest biomass estimation. In this study, we combined the multiple sources data (Lidar, SAR, Optical satellites, National Forest inventory data) to estimate AGB in Taiwan. We also constructed and compared the accuracies of four different prediction methods (support vector regression (SVR), multi-layer perceptron neural network (MLPNN), K-nearest neighbor (KNN), and random forest (RF)) for estimating AGB of forests in Taiwan. Finally, mapping Taiwan forest AGB using satellite remote sensing data and a good fit machine learning approach.
Materials and Methods.
- Study area: Study area are conducted in Taiwan, which is a mountainous island of 3,598,000ha off the southeast coast of Mainland China and north of the Philippines.
- Field reference data: In this study, total 3486 plots were collected with the plot size was set to 0.05ha, but if a plot was found to include three trees with DBH > 100cm, its plot size was expanded to 0.1ha. In those plots, there are 1564 plots were obtained from the 4th National Forest Resource Inventory in Taiwan (TNFRI4) between 2008 and 2012, plots were provided by the Taiwan Forest Bureau, and 1922 plots were obtained from results of MOST projects. In addition, the forest classification data obtained from Council of Agriculture Executive Yuan Taiwan in 2017. There are 4 main types of forest include: broadleaved forest, coniferous forest, bamboo forest and mix forest. Each of forest type were conducted the different model between plots data and remote sensing data.
- Remote sensing data: Lidar CHM for Taiwan at a spatial resolution of 20m was supplied by Department of Geomatics National Cheng Kung University. Optical data are used in form of spectral reflectance mosaic based on Landsat 5 and 7 ETM+ data. The SAR data were collected and processed by Japan Aerospace Exploration Agency (JAXA). Optical data and SAR data was collected in 2012 (same year with the TNFRI4).
- Method: There are five major steps were followed to build the AGB map in Taiwan: (1) predictors data extraction (LiDAR, SAR, Optical satellite...), (2) training and testing datasets processing, (3) using learning marching (SVR, MLPNN, KNN and RF) to build AGB estimation model (4) model calibration and assessment, and (5) Using the good fit model to mapping AGB in Taiwan.
Expected result.
- Techniques of AGB estimation are improved by using various prediction methods and combining multiple sources of remote sensing data.
- Comparative analysis AGB models from different remote sensing data sources.
- The estimated AGB maps showed similar goodness-of-ft statistics in Taiwan.