Terrestrial Vegetation Monitoring

“Terrestrial vegetation” is a key monitoring focus on all GULN parks. In addition to the general condition of park vegetation, various parks are strongly interested in several other vital signs that may be considered to be “related,” including Forest Health, Riparian (vegetation) Communities, Salt Marsh (vegetation) Communities, Fuel Load, and Non-Native Vegetation. See Chapter 3: Vital Signs for parks and vital signs.

GULN is developing, in collaboration with the USGS, a technology-based combined monitoring strategy that will simultaneously address all of these vital signs. Our combined vital signs approach is based on the recognition that these “vegetation” vital signs involve similar questions about aerial and spatial dimension, location, condition of plant stands or patches, and species-composition, except for Fuel Load, which does not involve a species component. These general questions may be quantitatively addressed by measuring the structure, dimensions and locations of vegetation patches or units as physical objects , without a priori consideration of what species one is looking at. As the questions may be answered, at least in large part, across vital signs and parks with like data (measurements), it appears reasonable that all of these vital signs may be effectively monitored using a common methodology and design . Accordingly, we are developing the GULN “Vegetation Structural Monitoring Protocol” (VSMP) as a combined-methods approach to address multiple vegetation-related vital signs across most or all parks in the network.

The initial and primary objectives of this VSMP is to provide reliable and consistent quantitative information on the physical structural characteristics of park vegetation resources at the larger-area and park/landscape level, together with limited taxonomic and typological biological description and conditional assessment of those sampled resources. Additional resource- and question-specific objectives will be developed as linked and derivative VS monitoring and SOPs (i.e., fuel loads, forest health, etc.) are added in following years. Initial development will focus on sampling “Terrestrial Vegetation.” After this primary protocol is completed, we will develop additional modules (essentially, a posteriori “secondary sampling designs” and associated analytical components) that will address the other vital signs in this group. See Chapter 5: Sampling Protocols for proposed modules and development dates.

The primary technological basis for the GULN VSMP is the use of airborne LiDAR coupled high-resolution Color-Infra-Red (CIR) digital photographic images to collect information about vegetation structure and composition. The use of LiDAR in vegetation assessment and monitoring is rapidly developing, both in technical refinement and in diversity and scope of potential application, as is exemplified in the growing international technical literature (i.e., Nayegandhi et al. 2006 (on a-posteriori evaluation of sampling footprint sizes); Thomas et al. 2006 (on photosynthetic rate and leaf area assessment in boreal mixed-hardwood forests); Anderson et al. 2006 (on canopy structure and biomass estimation in mixed forests); and Hinsley et al. 2006 (on assessment of woodland habitat for birds). The GULN VSMP will involve three major components: structural and dimensional measurement data collection using LiDAR; qualitative plant health and condition, and taxonomic identification data collection using CIR imaging; and “ground-truth” field surveys to verify and augment taxonomic, conditional, and distributional information derived from the LiDAR and CIR data. LiDAR and CIR images are “coupled,” or collected simultaneously in one survey flight; ensuring a high degree of sampling co-location and co-visitation with common revisit and panel design and sampling interval between these two methods. LiDAR and CIR will share a joint “two-level” development of sampling design, similar to that used for geomorphic monitoring. Ground truth sampling will be performed separately and on different scales and intervals with different revisit schedules.

LiDAR Sampling

The LiDAR sampling component will employ the same instrument technology as used for geomorphic sampling, and generate the same “ primary sampling design .” Sampling is performed in linear “flyway” belt-transects, creating a virtual systematic grid sample with sampling intensity and spacing of sampling points controlled by mechanisms described above. Primary sample frames are flyways and composites of flyways up to the whole-park area. Sample grids and points are georeferenced to fixed-location ground stations.

For the VSMP, LiDAR sampling will collect additional data that provide measures of object and layer surface elevation relative to mineral surface (landform, bald-earth). These data will include elevations of canopy top and one or more sub-canopy layers. These data are used in the analytical design to derive estimates of canopy and sub-canopy layer height and density, and measures of the patterns and heterogeneity that collectively provide 3-dimensional quantitative assessment of “object” (vegetation, canopy, dead timber on the ground) structure and size. These data are collected as a “ column wave-form signature ” on each sample point in the LiDAR survey; thus, all aerial, distributional and geospatial location aspects of the LiDAR sampling design apply equally to all layers within these vertical data columns.

LiDAR datasets collected for the VSMP are readily assessable for diverse a posteriori use in “ secondary sample design development ”: data may be stratified using geospatial and elevation data collected within the sample, and / or with reference to prior information (e.g., park maps, research records, etc.), data may be regrouped to address questions at different scales, and data may be sub-sampled and clustered by any randomization or targeted sampling approach (e.g., to examine a selected recognized “patch” or vegetation unit on the landscape).

CIR Image Collection and Sampling

CIR images, high-resolution digital photographs, are simultaneously taken with LiDAR data as survey transects are flown. These images form a mosaic of photographic images that can be joined together to create a “seamless” digital image of the entire LiDAR primary sample frame being considered. The CIR images are then geo-rectified, or spatially coordinated, with the co-collected LiDAR datasets. CIR images contain pixilated color signatures of what was over-flown in the flyway. These color pixel data, combined in co-analysis with LiDAR sample point structural data, create a machine-interpretable color-with-structure signature of a LiDAR “foot print” (combination or grouping of adjacent LiDAR data points into a larger “averaged” unit for analysis) that will be used to monitor the physical structure, coarse taxonomic composition, and, potentially, some aspects of plant health, of the sampled vegetation.

CIR datasets are, like LiDAR datasets, assessable for diverse a posteriori use and analysis. When linked with LiDAR data, all sub-sampling , grouping and stratification applied to the LiDAR dataset will automatically correlate with the CIR data, and reciprocally so, when such manipulations are applied to the linked CIR data. In addition, CIR data may be independently used in a posteriori sample design with separate analyses to address non-structure-related questions, such as possible spread of a plant pest or pathogen that causes a detectable (and geospatially-explicit) color change within CIR datasets.

Ground Truth Sampling

Both LiDAR and CIR sampling methodologies generate potentially large, consistent, and reliable “machine-logic” datasets and analytical outcomes information. These are the primary monitoring outputs that will be interpreted by our program staff and reported to park managers. For effective utilization in monitoring, outcomes from both methodologies will need to be calibrated and qualitatively verified by on-the-ground or “ground-truth” sampling.

Ground-truth sampling is currently in early development with collaborating vegetation experts located at Louisiana State University . We anticipate that this sampling will be designed to provide some amount of data about species composition, aerial spacing of larger individuals within selected sample frames, and, potentially, data reflecting plant health or condition. Possible sampling methods will include conventional plant ecology practices such as transect and plot sampling, point-quarter, line-intercept, and nearest-neighbor methods. Plot and transect size will be determined by subject matter experts taking into consideration both the questions being asked and the scale of the vegetation being sampled.

Distribution of sampling plots will be accomplished by reference to the park LiDAR virtual grid ( a posteriori plot location selection) or to extant park vegetation maps; ground-truth is intended to provide a posteriori identification or description of a “feature” (an apparent “patch” with structural and/or color characteristics differing from surrounding areas) observed in the LiDAR – CIR combined dataset. Thus, ground-truth efforts will be distributed to target those features within the larger LiDAR primary sample frame. For ground-truth sampling, the sample space of inferential interest will be within the boundary of the “feature” selected on the LiDAR dataset or park map. Allocation of sample plots within that feature will utilize either grid-based systematic distribution (based on a sub-set of the LiDAR virtual grid, scaled to fit the intended feature) or a simple random distribution across the set of grid points (Figure 4.2). The number, shape and size of plots used within the feature sample space will be scaled to provide a sample size adequate to estimate the properties of interest for that patch.

Figure 4 . 2 LiDAR primary virtual systematic grid “whole-park-sample” (A) with super-imposed secondary, or a posteriori , sampling designs, including change-of-grid combined-point “foot prints,” observed vegetation features, and ground-truth sampling within observed vegetation features B, C, and D

Sampling Intervals, Revisit Schedules, and Panels used in the GULN VSMP

The GULN program has yet to finalize the sampling interval, frequency and date-scheduling of terrestrial vegetation monitoring. One possible model we are currently considering for LiDAR – CIR vegetation sampling is that GUIS and PAIS be treated as one panel to be sampled in conjunction with geomorphic monitoring, while other parks would be distributed among two or more panels which would be sampled on a “1 on, 3 off” revisit design (type 1-3]. We further anticipate that some parks may receive additional sampling runs following “acute storm events” or if a park requests assessment of a possible acute event, such as a pest outbreak. Sampling frequency and revisit schedules for ground-truth sampling will be developed on a per-park and as-needed basis in consultation with the parks and with CESU collaborating vegetation experts.

Link to Terrestrial Vegetation Intranet page (NPS only)