Silvilaser 2019 - Poster Presentations »
The Use of Deep Learning and Three-Dimensional Convolutional Neural Networks to Interpret LiDAR Data for Live Standing Carbon Estimation in the Context of Carbon Offsets.
As light detection and ranging (LiDAR) technology is increasingly available, it has become common to use these datasets to generate remotely sensed forest inventories across landscapes. Traditional methods for generating these inventories employ the use of height and proportion metrics to measure LiDAR returns and relate these measurements back to field data using predictive models. Unfortunately, these metrics fail to quantify horizontal canopy complexity, and can be subject to change between LiDAR acquisitions of varying parameters. This can make the transference of models between LiDAR data sets difficult, and limits the ability to apply predictive models to landscapes and LiDAR of varying origin.
Here we employ a three-dimensional convolutional neural network (CNN) to directly analyze the LiDAR point cloud and make estimates of forest carbon stock. CNNs are a deep learning technique that operate by scanning and transforming the LiDAR using a series of moving windows. We will discuss the philosophy, operation, and practicality of these models. We will also compare them to traditional approaches that make use of height metrics.
Finally, we will discuss ways in which this technology can be used to improve upon existing carbon offset inventory protocols. This includes producing accurate live standing carbon estimates, projecting future tree growth, and making use of high-resolution stereo satellite imagery to accomplish similar goals.