Interior West Forest Inventory & Analysis

Research/Studies

Overview of Research Problem Area

The Technique’s Team’s charge is to develop innovative methods for sampling and integrating inventory data with remotely sensed information to improve the quality and efficiency of inventory analysis and reporting products, and develop new products to meet changing customer needs.

Land managers, private industry, and the public in the Interior West are placing higher demands on forest inventory information. They need more timely estimates of forest population totals that are unbiased and precise and are calling for maps depicting the spatial distribution of forest attributes, as well as means to integrate inventory information with their numerous GIS layers. Maps of key forest inventory variables of known quality over large geographic areas would greatly enhance forest land managers ability to identify suitable wildlife habitat, assess resource loss to catastrophic events, evaluate management prescription alternatives, and many other analyses. In addition, there is a wealth of information hidden in the vast quantities of inventory and ancillary digital data sources. Improved forest resource estimates and automated and visual analysis tools would not only enhance FIA analysts’ ability to report on resource dynamics, but give customers the ability to explore relationships affecting the lands they manage.

Recent advances in the fields of remote sensing, statistics and computer science afford FIA the opportunity to both improve the efficiency of our annual inventory products, and expand our product line to include maps and dynamic analyses. Over the next 5 years, studies under this problem will result in the following:

• Improved sampling methods for resource inventory design and measurements
• More efficient means to quantify population totals within an annual data collection system
• New FIA map products of known quality and utility
• Automated data extraction, visualization, and analysis capabilities

Current Projects

NPIP: Nevada’s Photo-based Inventory Project

The complex nature of broad-scale, strategic-level inventory demands constant evolution and evaluation of methods to get the best information possible while continuously increasing efficiency. With the recent launch of the annual forest inventory in Nevada, FIA has the opportunity to improve precision in estimates of forest parameters, reduce field data collection costs, address the potential of strategic-level inventory on lands not traditional sampled by FIA such as rangelands and riparian areas, and refine definitions of forestland. This project involves acquisition, processing, and  interpretation of large-scale real time GPS controlled aerial photography (LSP) throughout the State of Nevada over two field seasons. The LSP data will be integrated with FIA field plots and ancillary, moderate resolution satellite imagery through flexible estimation procedures. The over-arching goal is to enhance the existing annual system by exceeding information requirements, stepping-up inventory timelines, and reducing inventory costs. Specific objectives include:

1. Estimate field-derived forest attributes with known confidence after just two years of data collection.
2. Estimate photo-derived forest attributes with known confidence after just two years of data collection using only the 2 photo phases.
3. Reduce field data collection costs by eliminating need to field visit nonforest plots.
4. Estimate photo-derived non-forest attributes with known confidence after just two years of data collection.
5. Refine definitions of forestland.

LSP Data were collected in two parts, during the 2004 and 2005 field seasons, respectively. Part 1 involved collection and processing of LSP over pre-defined Timberland and Woodland strata in the state of Nevada. Part 2 involved the collection and processing of LSP over pre-defined non-forest stratum in the state. Photo-interpretation is underway.

Interior West “Core Maps”

Spatial representation of many forest attributes across the Interior West Region have recently been produced, using a hierarchical modeling procedure. These forest attributes included: forest/nonforest mask, basal area, basal area by ponderosa pine species, biomass, crown cover, forest type/group, growth, quadratic mean diameter, stand age, stand density index, trees per acre, volume, and weighted stand height. The modeling procedure was based on a nonparametric approach that attempts to extract patterns between FIA plot data and several layers of geospatial predictor variables - principally from MODIS imagery. Investigations were conducted to 1) analyze different candidate models to determine a suitable regional forest/nonforest mask, 2) compare the performance of various forest attribute models, and 3) explore the utility of forest mapping products in common land management decision-making processes. Results from these analyses are presented in the form of accuracy assessments performed on independent test sets, as well as comparisons of map- and plot- based distribution functions. Curently, pdfs of these maps are available for download from this web site.

Poster: Mapping forest attributes in the Interior Western states using FIA data, MODIS imagery, and other geospatial layers (6.6 kb)

Delineating forest change across the United States

FIA has partnered with researchers from NASA, the University of Maryland, and other Forest Service units to identify disturbance patterns across the United States using FIA plot data and time series of Landsat satellite images. Spatially explicit predictions of biomass loss and gain from 1972 to 2002 will be produced in 2-year intervals using 25 Landsat scenes distributed throughout the country. The map-based analyses that will be made possible through this collaboration will complement FIA’s current ability to track disturbances at the county and state level. The map below shows an application of the kind of disturbance information that will result from this project. For a two-year period, estimated changes in forest cover were binned according to Washington forest practice guidelines to identify harvests that removed conditions necessary for spotted owl habitat (in black) and harvests that left structural elements critical for habitat (gray). Because changes were mapped at the Landsat pixel level, it was possible to map variation of harvest effects on habitat within individual harvest units. Other applications of the project’s change data may include mapping of the disturbance and subsequent recovery of forests affected by fires, storms, and insects (Healey et al. In prep.).

Investigation of high resolution imagery:

With the continuous advancement of remote sensing technology, high resolution imagery, such as Quickbird and photography from the National Agriculture Imagery Program (NAIP), is becoming more readily available for use in forestry applications. Quickbird imagery is currently the highest resolution imagery commercially available. It consists of 2.44 m (8 ft) resolution multi-spectral bands ranging from blue to near infrared and a panchromatic band acquired simultaneously at 0.61 m (2 ft) resolution. In the near future, NAIP will provide annually updated, orthorectified, natural color, aerial photography at 1 meter resolution across the continental U.S. We investigated two classification methods, an Individual Tree Crown delineation and classification procedure and a technique using Feature Analyst software, for classifying high resolution Quickbird and NAIP photography. Both methods were found to be effective for discriminating different vegetation types using Quickbird and NAIP photography, although the Quickbird imagery proved to be superior to the NAIP photography according to visual and numerical assessments (Frescino et al. In press.).

Predicting wildlife habitat:

Landscape-scale species habitat classification models (HCM) are essential to species monitoring and conservation.  However, both presence and absence points are necessary to create a statistically valid HCM.  At a broad scale, available species’ occurrence data is often lacking and usually only in the form of observed presences.  In order to replace the lack of absence data, pseudo-absence points can be generated.  Current techniques generate pseudo-absence points at random across the entire study region, but often fail to incorporate biological knowledge concerning the relationship of a species to its habitat.  To incorporate knowledge of the species-habitat relationship, ecologically-based pseudo-absence points were generated and combined with extant presence points in generalized linear models (GLM), to model northern goshawk (Accipiter gentilis atricapillus) nest attributes across forested regions of Zone 16 (central Utah highlands).  Ecologically-based pseudo-absence points were generated at random within constrained habitat variable envelopes, which are spatial representations of a species’ gross appropriate habitat.  The northern goshawk was selected to test this approach because it is a management indicator species and its habitat requirements have been well studied.  Nest site (nest tree to 0.10 ha surrounding the tree) was modeled at 30-m resolution while nest area (0.10 ha to 12 ha surrounding the nest tree) was modeled at 250-m resolution.  Predictor habitat variables were derived from Forest Inventory and Analysis (FIA) map products, USDA Landfire map products, and digital elevation models.  Null models using traditional pseudo-absence point generation techniques (i.e., no incorporation of the species-habitat relationship) were created at both resolutions to assess the method of ecologically-based pseudo-absence points.  GLM top models were translated into a likelihood of occurrence surface across Zone 16, serving as a guide for censuses, population monitoring, and landuse planning.  Model fit and predictive capability statistics suggest that when lacking true absence points, generating ecologically-based pseudo-absence points to accompany known presence points can be an economical and robust tool for species habitat modeling.  This technique can be applied to a variety of species, ecosystems, data resolutions, and spatial extents.  (Zarnetske, In press).

Rapid assessment of changes in U.S. forests from wildfire:

The extent and effects of large wildland fires has increased dramatically in recent years, and there is a need for estimates of forest resources affected by these wildfires.  These estimates need to be timely and available for national, regional, and local geographic extents.  In response to this need, efforts are underway to develop a burn severity monitoring system for the United States. This paper explores estimation options for integrating data from the Moderate Resolution Imaging Spectro-radiometer (MODIS), Burned Area Emergency Response (BAER) maps and nationwide forest inventory to produce estimates of forest resources affected by wildfires. Model-assisted strategies are being developed for national and regional applications. Small area estimation strategies are being developed for local geographical extents. Methods will be tied to a national-scale burn severity monitoring system.

Comparing statistical modelling techniques:

Many efforts are underway to produce broad-scale forest attribute maps by modelling forest class and structure variables collected in forest inventories as functions of satellite-based and biophysical information. Typically, variants of classification and regression trees implemented in Rulequest’s© See5 and Cubist (for binary and continuous responses, respectively) are the tools of choice in many of these applications. These tools are widely used in large remote sensing applications, but are not easily interpretable, do not have ties with survey estimation methods, and use proprietary unpublished algorithms. Consequently, three alternative modelling techniques were compared for mapping presence and basal area of 13 species located in the mountain ranges of Utah, USA. The modelling techniques compared included the widely-used See5/Cubist, generalized additive models (GAMs), and stochastic gradient boosting (SGB). Model performance was evaluated using independent test data sets. Evaluation criteria for mapping species presence included specificity, sensitivity, Kappa, and area under the curve (AUC). Evaluation criteria for the continuous basal area variables included correlation and  relative mean squared error. For predicting species presence (setting thresholds to maximize Kappa), SGB had higher values for the majority of the species for specificity and Kappa, while GAMs had higher values for the majority of the species for sensitivity. In evaluating resultant AUC values, GAM and/or SGB models had significantly better results than the See5 models where significant differences could be detected between models. For nine out of 13 species, basal area prediction results for all modelling techniques were poor (correlations less than .5 and relative mean squared errors greater than .8), but SGB provided the most stable predictions in these instances. SGB and Cubist performed equally well for modelling basal area for three species with moderate prediction success, while all three modelling tools produced comparably good predictions (correlation of .68 and relative mean squared error of .56) for one species. (Moisen et al., in press)

Evaluating kriging as a tool to improve maps of forest biomass:

The USDA Forest Service, Forest Inventory and Analysis program (FIA) recently produced a nationwide map of forest biomass by modeling biomass collected on forest inventory plots as nonparametric functions of moderate resolution satellite data and other environmental variables using Cubist software. Efforts are underway to develop methods to enhance this initial map. We explored the possibility of modeling spatial structure to make such improvements. Spatial structure in the field biomass data as well as in residuals from the map was investigated across 18 ecological zones in the Interior Western U.S. Exploratory tools included directional graphs of summary statistics, three dimensional maps, Moran’s I correlograms, and variograms. Where spatial pattern was present, field and residual biomass were kriged, and predictions made for an independent test set were evaluated for improvement over predictions in the initial biomass map. While kriging has some potential benefit when analyzing the field data and exploring spatial structure, kriging residuals resulted in little or no improvement in the initial biomass map developed using Cubist software. Stationarity assumptions, variogram behavior, and appropriate model fitting strategies are discussed (Freeman and Moisen, In press.).

USDA Forest Service - RMRS - Forest Inventory & Analysis
Last Modified: March 30, 2006  Monday, 17 September 2007 at 14:52:35 EDT