Recipient Organization
UNIV OF CONNECTICUT
438 WHITNEY RD EXTENSION UNIT 1133
STORRS,CT 06269
Performing Department
Natural Resources & the Environment
Non Technical Summary
Accurate assessment and monitoring of forest biomass is pivotal for sustainable forest management and conservation and enhancement of aboveground carbon stocks (ACS). Acquisition of forest parameters through ground measurements is time-/-labor intense and cost prohibitive. Advanced remote sensing (RS) technologies, such as Light Detection and Ranging (LiDAR) data coupled with aerial imagery enable 3D modelling of forest structure in multitemporal fashion while resorting spatially-explicit information of ACS from local to regional scales. Connecticut enjoys terabytes of statewide, freely-available, LiDAR data and decimeter-scale aerial imagery with the unprecedented potentials for fine-grained ACS monitoring. However, such big data repositories are yet largely unexplored and derived forest science products are rare due to prevailing knowledge gaps and methodological challenges. We propose to develop a novel RS-data enabled Forest Carbon Estimation (FORCE) model to map statewide ACS stocks by coupling LiDAR data with ground-based forest inventory and analysis (FIA) data. The FORCE model will be spatially scalable and computationally extensible, allowing repeated ACS monitoring across time as new LiDAR/aerial imagery products become available. We will integrate FORCE model into an online geospatial gateway for ACS visualization and sharing algorithms, data layers, and map products. The outcomes of our concerted crossdisciplinary - forest ecology and RS - effort will permit a transformational increase in our capacity to generate spatially-explicit, fine-grained ACS estimates for the state of Connecticut while benefiting an array of stakeholders spanning from scientists, forest managers, and policy makers to public.
Animal Health Component
100%
Research Effort Categories
Basic
(N/A)
Applied
100%
Developmental
(N/A)
Goals / Objectives
The main goal of this project is to developa remote sensing data enabled forest carbon estimation model to map above-ground carbon stocksfor the state of Connecticut. We aim to address three specific objectives, which are to; 1) Develop an above-ground carbon stock monitoring system using LiDAR data and forest inventory and analyses data,2) Improve the LiDAR-based forest carbon modeling based on high-spatial resolution aerial imagery and3) Create a web-based interactive geospatial gateway to visualize and disseminate forest carbon model results, map layers, and open-source tools and algorithms.
Project Methods
The proposed Forest Carbon Estimation (FORCE framework) will consist of three key segments: processing, modelling, and visualization. Objective 1 and 2 entail processing of LiDAR and NAIP imagery to derive forest metrics and textural characteristics, respectively. Machine learning (ML) based modelling will then combine remote sensing derived metrics with FIA data. Objective 3 will create an online geospatial gateway to visualize and share the results.We will use LiDAR data from state-wide mission, which was flown in the early spring of 2016 during leaf-off condition with point cloud densities of 2+ points/m2. It has been shown that the LiDAR data set is effective for modeling a variety of forest characteristics including canopy height and canopy density Four-band (blue, green, red, and near- infrared) NAIP imagery from summer 2016 will be accessed via the data portal of USDA Geospatial Data Gateway (https://datagateway.nrcs.usda.gov/). We will develop Python-based high throughput pipelines for data processing and machine learning. Our algorithms and workflows will be optimized and parallelized to efficiently run on high performance computing (HPC) resources available at University of Connecticut (UConn). We will derive canopy height model (CHM) based on first return of LiDAR. Imagery-based masking operations will be tasked to exclude non-vegetated areas. The CHM production will build on object-based modeling approach equipped with adaptive segmentation and class modelling. Additional LiDAR derivative will include height percentiles, density metrics, canopy cover. Image. We will implement gray-level co-occurrence matrix (GLCM) (Haralick 1973) on the green channel of NAIP imagery to compute a suite of GLCM texture measures, such as mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment and correlation. The modelling phase will build on the Random Forest (RF) algorithm, which is an ensemble machine learning technique. It has been demonstrated that the choice of allometric model is important in biomass calculations. Thus, we will task two commonly found methods; Jenkins methodand CRM method to compare the biomass estimates and to understand impact of two approaches on final predictions. We will sample a set of 'FIA-like' field plots, which will be opportunistically located based on the forested pixels of the national land cover database (NLCD) maps. Primarily, we will develop two RF models to predict biomass. One model will only rely on LiDAR variables (Objective 1) and the other model will combine LiDAR and texture measures (objective 2). Carbon stock estimates will be compared to FIA data at plot level and county level. Predictions maps will be produced at 30 m x 30 m granularity. A public website will be created to disseminate the FORCE project findings and products to various stakeholders. The website will include an embedded ArcGIS online web map that provides interactive viewing of above ground carbon stocks along with other publicly available geospatial layers. Users will be able to perform custom analyses, query, and generate reports. The portal will allow FORCE model result downloads. Our gateway will provide information explaining the predictor variables, models as well as provide Python-based tools. Computer algorithms will be made available as open source materials via GitHub. The tools will also be incorporated into an ESRI format suitable for use in ArcGIS Desktop or ArcPro which are common software used by the stakeholders.