SNAMP Pub #13: Predicting surface fuel models and fuel metrics using lidar and CIR imagery in a dense, mountainous forest
Article Title: Predicting surface fuel models and fuel metrics using lidar and CIR imagery in a dense, mountainous forest.
Authors: Marek K Jakubowski, Qinghua Guo, Brandon Collins, Scott Stephens, Maggi Kelly.
- Forest fire behavior models need a variety of spatial data layers in order to accurately predict forest fire behavior, including elevation, slope, aspect, canopy height, canopy cover, crown base height, crown bulk density, as well as a layer describing the types of fuel found in the forest (called the “fuel model”).
- These spatial data layers are not often developed using lidar (light detection and ranging) data for this purpose (fire ecologists typically use field-sampled data), and so we explored the use of lidar data to describe each of the forest-related variables.
- We found that many of the forest canopy structure metrics that are commonly used in fire behavior modeling (e.g. tree height, canopy base height, canopy cover) were estimated well with lidar data (up to 0.87 correlation coefficient).
- However we also saw that the descriptive strength of the lidar data generally decreased with increased lidar penetration into the forest canopy.
- Specifically in the Sierra Nevada mixed conifer forests, we found that general fuel types were mapped with lidar data at up to 76% accuracy, while specific surface fuel types were more difficult to assign.
Surface forest “fuel models” (these are descriptions of the type of fuel found in a forest, and can be general in description: mostly shrub, more timber, etc., or specific: “dwarf conifer with understory” or “low load compact conifer litter”) and canopy structure metrics (such as average height, basal area, canopy cover, etc.) are critical spatial inputs to fire behavior models such as FlamMap (http://www.firemodels.org/index.php/national-systems/flammap) or FARSITE (http://www.firelab.org/science-applications/science-synthesis/72-farsite). Typically, the assignment of these “fuel models” is subjective and done by fire ecologists based on their knowledge of the area, and canopy structure metrics are often estimated on a large spatial scale from sparsely sampled field data and visual interpretation of remotely sensed imagery. Because Lidar generates wall-to-wall coverage of vertical forest structure, it is a good candidate to estimate these inputs for fire behavior models.
In this paper, we present a comprehensive examination of forest fuel models and forest fuel metrics derived from lidar and color infrared (CIR) imagery (CIR is often used for mapping vegetation since plants reflect infrared light well) for use in fire behavior modeling. Specifically, we used high-density, discrete return airborne lidar data and National Agriculture Imagery Program (NAIP) 1-meter resolution imagery to find the optimal combination of data input (lidar, imagery, and their various combinations/transforms), and method (we used three types of methods: clustering, regression trees, or machine learning algorithms) in order to extract surface fuel models and canopy metrics from Sierra Nevada mixed conifer forests. All lidar-derived metrics were evaluated by comparing them to field data and deriving correlation coefficients.
Specific surface forest “fuel models” (these are detailed descriptions like “dwarf conifer with understory” or “low load compact conifer litter”) proved difficult to predict in this dense forest environment, although general fuel types (such as predominantly shrub, or mostly timber) were estimated with reasonable (up to 76% correct) accuracy because fewer of the light energy from the Lidar penetrated to the forest floor in denser forests, making accurate characterization of understory shrubs more difficult. The predictive power of canopy metrics increases as we describe metrics higher up in the canopy. The accuracy—in terms of Pearson’s correlation coefficient—ranged from 0.87 for estimating canopy height, through 0.62 for shrub cover, to 0.25 for canopy base height.
- Lidar data can be used to predict fuel types even in dense mixed-conifer forests; however, predicting specific surface fuel models (as defined by Scott & Burgan, 2005) is difficult.
- Many canopy structure metrics can be accurately estimated using lidar over the entire study area.
- Using lidar data alone was best for “simple” metrics (e.g. canopy height); more complex input (such as a combination of lidar and imagery) was required for more complicated metrics (e.g. shrub cover or crown base height).
- The predictive power of lidar to estimate canopy metrics generally decreases with increased lidar penetration towards the forest floor.
Scott, J. H. and R. E. Burgan, 2005. Standard fire behavior fuel models: a comprehensive set for use with Rothermel's surface fire spread model. General Technical Report. RMRS-GTR-153. USDA Forest Service.
Jakubowski, Marek K., Q. Guo, B. Collins, S. Stephens, M. Kelly. 2013. “Predicting surface fuel models and fuel metrics using lidar and CIR imagery in a dense, mountainous forest,” in Photogrammetric Engineering & Remote Sensing, 79(1): 37-49.
The full paper is available here.
For more information about the SNAMP project and the Spatial team, please see the: Spatial Team Website.
To learn more about lidar data, check out our lidar FAQs sheet. For more information on lidar data in SNAMP, see the spatial team website: http://snamp.cnr.berkeley.edu/teams/spatial, and our spatial team newsletters that focus on lidar: Vol. 2, No. 3, and Vol. 5, No. 1.