This project aims to develop more sophisticated algorithms for forest biomass retrieval, and applies them to the Democratic Republic of Congo (DRC). The region has the world’s second largest rainforest, the Congo Rainforest, which is a huge reservoir of carbon stocks on earth. The increasing deforestation activities in DRC may offset the carbon stocks and accelerate the process of anthropogenic climate change. In order to better estimate and monitor the deforestation and decreasing carbon stocks, we need to develop remote sensing-based biomass retrieval algorithms and establish a timely monitoring system.
Using tree heights measured by LiDAR, we can build a regression model between heights and bioss and use it to estimate biomass at a large scale. Yet most of the existing biomass retrieval algorithms use only one regression function for different types of forests, assuming the homogeneity of forest cover. However, many previous studies suggest that the regression results can vary greatly across different forest types. The variation across forest types causes a large uncertainty for the homogeneous retrieval algorithms. In order to reduce this uncertainty, this project try to combine classification and retrieval algorithms and develop multiple regressions for different forest types.
This project will achieve the following objectives:
- Combine multiple sources of remote sensing data, such as LandSat, areal photographs and LiDAR to classify forest types in DRC.
- Integrate multiple classifiers, such as maximum likelihood, neural network and support vector machine, by using ensemble learning technique, to improve the classification results.
- Use systematic sampling techniques to generate sample points for in situ measurements of biomass for different forest types.
- Build multiple regression models for various forest types from in situ observations and LiDAR data.