|John Coulston||Overview of timber products monitoring: recent changes and applications||Tuesday||3b||Sequoyah||Advances in timber products monitoring||The timber products monitoring component of the FIA program is shifting to an annual effort. Here we provide an overview of the statistical design and provide relevant examples of how annual information can enhance reporting efforts in the Western United States.||The Advances in timber products monitoring session aims to highlight key technical changes to how the estimates of roundwood consumption and production are constructed. The annual timber products design is based on an innovative stratified simple random sample approach that approximates probability proportional to size design. We will review this design to provide context for subsequent presentations in the session. Given the programmatic thrust to shift to an annual sample based design we will also provide examples of some assessment applications that can be enhanced with annual timber products estimates. Our examples will focus on application in the Western United States.|
|James Westfall||Estimating change in annual timber products output using a stratified sampling with certainty design||Tuesday||3b||Sequoyah||Advances in timber products monitoring||Estimation of change is a key output of the TPO monitoring program in the U.S. Approaches to estimating the covariance between successive samples were evaluated, where often only a single sample unit occurred in both samples within a stratum. While the covariance estimation methods performed poorly, treating the samples as being independent provided results that were consistent with the Monte Carlo simulation variance. This outcome was partially due to some strata being sampled with certainty.||The national timber products output (TPO) monitoring program in the U.S. is adopting a stratified sampling approach to be conducted annually. Estimation of change from year-to-year is necessary, but is complicated due to shifts in the population as well as changing strata over time. In this study, various approaches to estimating the covariance between successive samples were evaluated. A primary challenge was that often only a single sample unit occurred in both samples within a given stratum. The result that none of the covariance estimation approaches performed adequately was largely overshadowed by the outcome that treating the samples as being independent provided an overall variance estimate that was very consistent with the Monte Carlo variance obtained via simulation. It is proposed that this outcome was a derivative of the sampling design, which included some strata that were sampled with certainty. Due to the complexities introduced through changes in populations and strata over time, being able to treat the samples as independent is very beneficial because it avoids the need to introduce complex covariance calculations into the estimation process.|
|Christopher Edgar||Alternative measures of size and sample-with-certainty thresholds in monitoring of timber production in the Lake States||Tuesday||3b||Sequoyah||Advances in timber products monitoring||A new sample design for annual monitoring of timber production is currently being implemented in the Lake States. We examine two key areas of the new sample design: the selection of an effective measure of size used in constructing strata and the identification of a threshold value for allocating mills into sample-with-certainty strata. We discuss the efficiency of the new design, implementation considerations, and potential areas for further research.||The Forest Inventory and Analysis program is implementing a new sample design for annual monitoring of timber production in the Lake States and other regions of the United States. We examine two key areas of the new design: the selection of an effective measure of size used in constructing strata and the identification of a threshold value for allocating mills into sample-with-certainty strata. Precision of estimates can be increased by using measures of size that are more highly correlated with the variable of interest. We review the availability of mill profile information in Minnesota, Michigan, and Wisconsin and discuss the strength of relationships of candidate measures of size and timber production. When sampling skewed populations, a few large units may account for a large portion of the total. We examine different approaches to allocating mills to sample-with-certainty strata and the impacts on precision of the estimates. We conclude the presentation with general discussion of the efficiency of the new design, implementation considerations, and potential areas for further research.|
|Erik Berg||Western region TPO annual sampling- first year experience||Tuesday||3b||Sequoyah||Advances in timber products monitoring||To guide TPO annual sampling plans, University of Montana staff simulated sampling of active mills in 11 western states. The national TPO R program was used to select mills to sample; outputs were post-processed for nonresponse. Differences in FIA TPO censused (the true volumes) and sample-predicted timber volumes varied from 0.44 to 4.73 percent and standard errors varied from 0.10 to 4.77 percent by state. Accounting for nonresponse was the most critical component of this work.||Annual mill sampling has been added to the suite of FIA TPO services, such as the periodic censuses of facilities, to provide stakeholders timely estimates of received roundwood and mill residue volumes. To help guide future sampling plans University of Montana (UM) staff developed non-replicated simulations of sampling protocols for 11 western states. Our goal was to identify protocols which minimized differences in state-level received timber volumes between the most recent UM censuses (assumed to represent true volumes, including nonresponse) and annual sampling estimates, and which also produced standard errors of the mean of less than five percent. The national TPO R program was used to select active mills to sample; we tailored program sampling percent, certainty volumes and product type mixes for each state. Program outputs were post-processed to adjust for nonresponse. Differences in censused and program-predicted state-level roundwood volumes varied from 0.44 to 4.73 percent and standard errors varied from 0.10 to 4.77 percent. Accounting for anticipated nonresponse was the most critical component of this work. These efforts have helped prepare UM staff to conduct informative annual samples.|
|Marcus Taylor||Improving residential fuelwood estimates for TPO||Tuesday||3b||Sequoyah||Advances in timber products monitoring||TPO is implementing a new methodology to estimate residential fuelwood use. In the past the program simply reported data from the Energy Information Administration (EIA) as firewood is typically obtained through a path not captured by TPOs surveys of primary wood processors. However, EIA estimates are only available for the four Census Regions while TPO strives to provide county-level data. This new model estimates residential fuelwood by county using EIA, Census, climate, and fuel price data.||Timber Products Output (TPO) studies report estimates of forest removals by county. The majority of data for the TPO program are collected by surveying primary wood processors, such as sawmills. This survey methodology largely omits fuelwood for residential use as many users of residential fuelwood obtain the product themselves or through a path not captured by the industrial surveys. Previously, data from the U.S. Energy Information Administration (EIA) was used for the residential fuelwood component of TPO reporting, however EIA data are not available on the same spatial scale as other TPO estimates nor do they provide geographic sourcing information. FIA is developing, testing, and implementing a new methodology to estimate residential fuelwood use that will better align with other TPO estimates. Using data from the EIAs Residential Energy Consumption Survey, Census American Community Survey, the National Climatic Data Center, and EIAs home heating fuel prices, volumetric data for annual residential fuelwood usage in TPO will be modeled to the county level as well as species distribution and county-of-origin information using ancillary TPO and FIA data.|
|Brett Butler||Timber Products Output Field Data Collection Methods: Past, Present, and Future||Tuesday||3b||Sequoyah||Advances in timber products monitoring||The past and current methods used to collect TPO data from mills will be reviewed. Potential future data collection protocols, based on the current science of survey design and implementation, will then be discussed. Specific components to be considered include questionnaire design, the use of incentives, and data collection modes.||The FIA Timber Products Output program has been collecting data from primary wood processing facilities across the U.S. for decades. The TPO program has recently switched to a sample-based data collection protocol and it is time to thoroughly review the field data collection methods. Traditional techniques often involved visits to mill locations and working with mill managers to complete the questionnaires. More recent effort have started to rely more on self-administered, mail-back questionnaires. This presentation will summarize the past and current methods used to collect TPO data. Potential future data collection protocols based on the current science of survey design and implementation will be discussed along with methods for testing these new approaches. Specific components to be considered include questionnaire design, the use of incentives, and data collection modes.|
|Rebekah Zehnder||Southern Timber Supply Analysis: Forest Inventory Data for All||Tuesday||3b||Sequoyah||Advances in timber products monitoring||The Southern Timber Supply Analysis web application summarizes USDA Forest Service Forest Inventory and Analysis (FIA) data for a user-defined supply area. The easy-to-use application produces estimates of the amount of timberland and standing timber, growth, and removals within a user-specified distance or trucking time of a site of interest in the U.S. South. Southern Timber Supply Analysis broadens delivery of FIA data, simplifying the process of examining forest inventory and sustainability.||The Southern Timber Supply Analysis web application summarizes USDA Forest Service Forest Inventory and Analysis (FIA) data for a user-defined supply area. The application produces estimates of the amount of timberland and standing timber, growth, and removals within a user-specified distance (50, 75, or 100 miles) or trucking time (1, 1.5, or 2 hours) of the users site of interest in the U.S. South. The analysis can be filtered by state and ownership, and timber quantities can be displayed by volume or by green weight. The results can be downloaded in a PDF report. The application also contains pre-made statewide reports available for download.
Designed specifically for ease of use, Southern Timber Supply Analysis simplifies the process of examining forest inventory levels and sustainability within a custom area. It presents the results equivalent to running numerous EVALIDator queries in a matter of seconds, with very little effort required by the user. The information available through Southern Timber Supply Analysis will support economic development, conservation and sustainability efforts, and state forestry agencies and associations.
|Jack Triepke||ICE - Land Use, Land Cover and Agent of Change Data for NFS||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||This talk discusses a project utilizing historical scans of the NFS Region 3, Lincoln National Forest, implications for NFS, and practical methods used to support ICE measurements of the Forest Service scanning their historical imagery.||The Forest Service has invested in measuring land use, land cover and agent of change using the NAIP imagery source the Image Change Estimation project. The information provided from the ICE project is unique to supporting multiscale monitoring designs. The ICE project provides information to Forest Service R&D, S&P, and NFS Deputy Areas. Key support and reliance comes from the FIA program; using the FIA plot locations allows for design based estimators, LU/LC/AC data to be estimated on all lands, supports consistent/unbiased intensification across all lands, allows for monitoring change on permanent plots, and is nationally implemented. The NFS uses of ICE measures allows for temporal measures forward through time. As the Forest Service scans its historical aerial photos, the ability to measure land use, land cover and change agent estimates back through time supports tracking supporting a historical context through multiple decades. This talk discusses a project utilizing historical scans of the NFS Region 3, Lincoln National Forest, implications for NFS, and practical methods used to support ICE measurements of the Forest Service scanning their historical imagery.|
|Andrew Lister||Using biannual change detection maps for prestratification of photointerpreted survey samples for area change estimation||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||Map-based estimates often lack useful uncertainty metrics. For a forest inventory survey unit in Georgia, Landsat-based biannual change maps were used to prestratify plot data on cover change class derived from Landsat temporal signatures made with the TimeSync software. The goal of the project was to assess the accuracy of the change map, produce corrected area estimates of change, and assess the feasibility and costs of using this method for operationally producing biannual change estimates.||Monitoring land use and land cover change with ground plots or remote sensing is important to environmental managers and policymakers. With estimates directly from maps, traditional uncertainty indices are difficult to interpret through the lens of sampling theory. Estimates from design-based ground plot samples are interpretable using sampling theory, however they can be more expensive and are not wall-to-wall. The current study seeks to leverage the strengths of remote sensing- and plot-based area change estimation. For a forest inventory survey unit in Georgia, Landsat-based biannual change maps derived through time series analysis were used to prestratify plot data on cover change class derived from Landsat temporal signature graphics assembled using the TimeSync software. The goal of the project was to assess the accuracy of the biannual change product, produce corrected area estimates of change from that product, and assess the feasibility and costs of using this method for operationally producing biannual change estimates. Results will be presented and implications for the US Forest Services Forest Inventory and Analysis business practices will be discussed.|
|Joseph McCollum||Land use: ICE vs. P2||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||Image-based Change Estimation (ICE) is a relatively new photointerpretation project created by FIA. The ICE process involves collecting land use, land cover and disturbance information on FIA plots using two dates of aerial imagery. FIA already collects similar information through its normal business operations. Here ICE and FIA land use calls are compared across multiple states to evaluate potential for using the data sets to form blended estimates in areas with high nonresponse rates.||Image-based Change Estimation (ICE) is a relatively new photointerpretation project created by the U.S. Forest Inventory and Analysis (FIA) program. The ICE process involves collecting land use, land cover and disturbance information on FIA plots by photo interpreting aerial images collected at two points in time. The FIA program already collects similar information through its standard process of checking the forest status of plots prior to field visitation (referred to as double sampling for post-stratification). This redundancy results in two similar, but different data sets which can lead to conflicting estimates of forest land area and other variables. In this presentation we compare ICE and FIA data from multiple states to evaluate the level of agreement for five general land use categories (forest, agriculture, developed, water, and other). Results show overall agreement ranged from 66% in Connecticut to 96% in the U.S. Virgin Islands. Although agreement varied significantly by state, the results for forest, agriculture and developed were generally high enough to indicate potential for using ICE and FIA interchangeably in areas where non-response rates are high.|
|Scott Pugh||TESTING FOR TEMPORAL CHANGE IN FOREST INVENTORY & ANALYSIS PROGRAM (FIA) DATA USING THE DIFFERENCE TESTER ON THE WEB||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||Online analytical tools from FIA provide current inventory and change estimates that guide management and policy decisions for our Nations forest resources. Temporal comparisons are made using these tools but the tools do not account for covariance. A new online tool called Difference Tester accounts for covariance when comparing estimates of a chosen attribute between two consecutive inventory cycles (z-score and p-value indicate likelihood of a statistical difference).||Online analytical tools from FIA provide current inventory and change estimates that guide management and policy decisions for our Nations forest resources. A broad range of temporal comparisons are of interest to analysts. For example, an increase in the number of standing dead trees can indicate a forest health issue. Often, these comparisons have covariance because the estimates are derived from observations on the same plots over time (not independent). Hypothesis testing for statistically significant differences should account for covariance but FIA tools to date have not offered this benefit. A new online tool called Difference Tester (|
|Karen Schleeweis||Proximate Causes of Forest Loss 2001-2011 in a Landscape Mosaic Context||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||We aim to evaluate gross and intact forest loss nationally and their causal processes. Here we use two dates of the new epoch of National Land Cover Database (NLCD) land cover change maps, the annual North American Forest Dynamics (NAFD) attribution maps and the Landscape Mosaic methodology. Results are given in terms of regional differences and Landscape context. An advantage of this neighborhood approach is that it alleviates many issues related to per-pixel comparisons of maps.||The areas of forest loss are of concern for many Earth System applications. Equally important are changes in the spatial arrangement of forest. To really understand these dynamics and their proximate causes and landscape context is invaluable. Here we use two dates of the new epoch of National Land Cover Database (NLCD) land cover change maps, 2001 and 2011, to analyze changes in gross and interior forest. We combine these metrics with signals, measured in 4.41-ha neighborhood, of proximate causal processes. NLCD maps provide land cover change classes and North American Forest Dynamics (NAFD) attribution maps provide classification of fire, harvest, and insect/stress for forest canopy cover loss events nationally (CONUS). Annual NAFD attribution data are subset to the same 2001-2011 period for analysis. An advantage of this neighborhood approach is that it alleviates many of the issues related to per-pixel map comparisons. We discuss variations in changing forest patterns and proximate causes across the Resources Planning Act (RPA) regions and Landscape Mosaic categories, which locates and measures dominant land cover and the degree of land cover heterogeneity at a specified scale.|
|Zhiqiang Yang||Classification of satellite time series parameters instead of individual images supports temporally coherent model-assisted estimation at annual time steps||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||We demonstrate the use of time series analysis to support design-based inference by estimating the area of tree cover over time for major regions of the United States, combing annual CCDC-based maps of tree cover with a national-scale random sample of tree cover history collected using the TimeSync tool.||There are mature methods for using maps to support design-based estimation of forest characteristics. An obstacle to applying these model-assisted methods every year is the lack of temporally coherent maps. When land cover models are applied to individual images from a series of years, image noise makes it likely that cover class will fluctuate implausibly through time. This cause big problems for model-assisted change estimates.
We produced the first national time series analysis of the Landsat satellite record (1985-2018), using the CCDC algorithm. This analysis fits harmonic functions to all cloud-free acquisitions for each pixel (1000+ images), allowing us to see beyond image-specific noise. Parameters of these functions remain the same until a land cover change is detected. Applying models to time series parameter makes map calls both more accurate by accounting for the entire growing year and more stable between detected changes.
We demonstrate the use of time series analysis to support design-based inference by estimating the area of tree cover over time for major regions of the United States, combing CCDC-based maps of tree cover with a national-scale random sample of tree cover history collected using the TimeSync tool.
|Olaf Kuegler||Monitoring Change in the Pacific Islands: Non-traditional Estimates of Change||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||PNW-FIA has measured the forest in the U.S. affiliated Pacific Islands since 2001 and recently completed a full remeasurement of these surveys. Interest in traditional FIA components of changes (growth, removals and mortality) is very limited. However, monitoring changes in forest structure and biodiversity as well as any increases of invasive species are of paramount interest.||PNW-FIA has measured the forest in the U.S. affiliated Pacific Islands (American Samoa, Guam, Republic of Palau, Commonwealth of Northern Mariana Islands (CNMI), Federated States of Micronesia (FSM), and Republic of the Marshall Islands (RMI)) since 2001 and recently completed a full remeasurement of these surveys. Apart from changes in food crops (e.g., change in number Coconut trees), interest in traditional FIA components of changes (growth, removals and mortality) is very limited. However, monitoring changes in forest structure and biodiversity as well as any increases of invasive species are of paramount interest.
The Micronesia Challenge (www.miconesiachallenge.org) is a commitment by Palau, FSM, RMI, Guam and CNMI to conserve at least 30% of the near-shore marine resources and 20% of the terrestrial resources across Micronesia by 2020". In 2013, the Micronesia Conservation Trust started to fund the collection of additional plots in order to monitor the progress of conservation efforts.
While the FIA database structure is designed to allow the estimation of the traditional components of change, it can also be used to estimate change in forest structure, biodiversity as well as changes in invasive species.
|Mauricio Vega-Araya||Costa Rica's Land Use, Land Cover, and Ecosystems Monitoring System (SIMOCUTE).||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||This work presents the state of progress of SIMOCUTE in Costa Rica, emphasizing the use of tools such as Collect Earth Online (CEO), the image-based change estimation (ICE) tool and FIESTA (Forest Inventory ESTimation and Analysis), an R package developed by the US Forest Service.||Costa Rica implemented a National Forestry Inventory (NFI) in 2014 as part of its MRV processes. This inventory was implemented by the Costa Rica's forest office. Among others products, the NFI generated a systematic gird (SG) of points. Field plots were established on a subset of these. The SG has formed the basis of the countrys National Land Use, Land Cover, and Ecosystems Monitoring System (SIMOCUTE). SIMOCUTE aims to be the countrys official platform for coordination and institutional and sectoral integration, to facilitate the management and distribution of information and data related to the countrys land use, land cover, and ecosystems. In addition to the NFI, SIMOCUTE includes a mapping subsystem and a land use and land cover monitoring system based on photo-interpreting these attributes from the grid of points. The latter subsystem generates tabular estimates of the areas of the countrys different land uses and land covers, and changes in them through time.
This work presents the state of progress of SIMOCUTE in Costa Rica, emphasizing the use of tools such as Collect Earth Online, the image-based change estimation tool and Forest Inventory ESTimation and Analysis, an R package developed by US Forest Service.
|Todd Schroeder||Developing county-level harvest and conversion rates for southeastern forests using Landsat time series and U.S. Forest Inventory and Analysis (FIA) plots||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||In the southeastern U.S. reliable estimates of harvest area and volume are needed at larger scales and smaller time-steps than what FIAs plot sample alone can reliably produce. In this study Landsat disturbance maps and time series metrics are combined with FIA plots to estimate harvest and conversion rates for 67 counties in Georgia. Results are used to discuss advantages of agent specific predictors, and the benefits of using model-assisted approaches to improve estimation of rare events.||Known as the nations wood basket, the southeastern United States contains some of the most biologically diverse and dynamic landscapes in the country. In order to help sustainably manage the effects of forest management reliable estimates of harvest area and volume are needed at spatial and temporal resolutions that are typically smaller than what FIAs plot sample alone can reliably address. Remote sensing techniques, particularly the use of Landsat time series algorithms, have surfaced as an effective way to map the location, extent, and timing of forest harvest activities and other types of disturbance. In this study Landsat disturbance maps and time series metrics are used to develop Fay-Herriot small-area estimation models to predict annual harvest and conversion (i.e. loss of forest to non-forest use) rates for 67 counties in Georgia from 1986 to 2010. Results are used to discuss the advantages of using agent specific predictor variables, as well as the benefits of combining remote observations with FIA plots to improve annual estimation of rare events. Potential for using county-level harvest estimates in timber products studies, FIA reporting and other applications are also discussed.|
|Paul Patterson||Photo-based or Pixel-based Change Estimation?||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||The precision of photo-based and mapped-based (pixel-based) estimates of change categories within a geographic region are compared. A statistically based method for increasing the precision of photo-based estimates of change are presented. Finally the statistical and technical properties of the photo-based and pixel based methods for change estimation are compared.||Aerial photography may be used to estimate change within a landscape. This involves establishing a base sample of photo-plots and interpreting the characteristics of a grid of points within each of the photo-plots. The points within the photo-plot give an estimate of the proportion of the change categories on photo-plot; which are combined for the overall estimate. In many situations (REDD+) when using the base sample the precision of the estimates of the change categories are not sufficient. Another method to estimate change is to create a disturbance map (e.g., from the Landsat stack), then using a probability sample of the pixels evaluate the pixel level accuracy of the map, and finally using the accuracy assessment output construct an estimate of the proportion of each change category. A sample size based method for increasing the precision of the photo-based estimates will be presented. This will be followed by a discussion of the statistical and technical properties of the photo-based and pixel based methods.|
|Nicholas Nagle||Developing FIA survey weights with high spatial and temporal resolution by calibration to Landsat time series||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||Estimates of FIA plot characteristics are desired at spatial and temporal scales not directly supported by the plot sample design. A method based on penalized maximum entropy is presented that allows the calibration of survey weights to large sets of auxiliary data. Here this method is used to produce survey weights that have been calibrated to Landsat time series for two FIA regions covering 67 counties in Georgia. Results are used to discuss the pros and cons of entropy-based design-weights.||FIAs plot sample is not designed for the spatial and temporal resolution of annual, wall-to-wall maps or small area estimates. While many auxiliary data are available for calibrating FIA survey weights to small domains, many of these auxiliary data are collinear or noisy, making it difficult to use standard survey weighting approaches. A penalized maximum entropy approach is presented for survey weight generation. Like LASSO, this approach allows the efficient use of collinear and noisy auxiliary variables. Compared to LASSO, the penalized maximum entropy has the desirable property of producing non-negative survey weights, but does not possess the same desirable sparseness properties as LASSO. We demonstrate the application of this methods to produce a time series of survey weights for 67 counties in Georgia that are calibrated to Landsat time series that have been categorized using a recently published non-linear, trajectory fitting algorithm. Although entropy-based design weights are slightly less precise than LASSO, they can be used with multiple forest attributes to produce estimates which are internally consistent such that sub-populations sum to larger regions, and tabular estimates match spatial estimates.|
|Gretchen Moisen||Comparing and combining observation systems for land use and land cover change in Georgia||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||Using data collected in the north central section of Georgia, we compare four sources of land use and land cover change data, including that from FIA plots, ICE plots, Timesync plots, as well as Landsat-based maps. We explore ways in which these four sources can be combined through model-assisted estimation methods, as well as through logistically harmonized systems, to produce the best information with an eye toward cost.||FIA has access to a variety of observation systems for understanding land use and land cover change in the US. There is the network of ground plots, photo-based observations collected through Image-based Change Estimation (ICE) methodology, Landsat-based observations collected through Timesync, as well as a variety of change map products. Using data collected in the north central section of Georgia, we compare how four sources of land use and land cover change information differ in terms of the story they tell about land use and land cover dynamics. That is, for each of the observation systems we assess the ability to detect net and transitional change estimates through time that are statistically different than zero, as well as trends that are significantly increasing or decreasing. We also explore ways in which these four sources can be combined through model-assisted estimation methods, as well as through logistically harmonized systems, to produce the best information with an eye toward cost.|
|Jonathan Knott||Distributional shifts of regional forest communities in the eastern U.S.||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||We identified regional forest communities of the eastern U.S., and we assessed distributional shifts of these communities. We found 11 of 12 communities have shifted their distributions, either in their centroid or area. We assessed the impact of forest- and climate-related predictors on changes in forest community distribution but found that, although significant, the model performed poorly, indicating the lack of a strong community-level response to climate change.||The impact of climate change on forests is often measured by species-level changes, but much less is known about community-level responses. Here, we identified 12 regional forest communities of the eastern U.S. using data from the Forest Inventory and Analysis Program and the Latent Dirichlet Allocation topic model. In addition, we detected distributional shifts by evaluating movement in community centroid and changes in community area over the last three decades. We found 11 of 12 communities had significant shifts in centroid (eight in longitude, seven in latitude), and 5 of 12 communities had significant changes in area (three expansions, two contractions). Mixed-effects models revealed significant forest-related (fire frequency, basal area, and nitrogen deposition) and climate-related (temperature, precipitation, and precipitation change) predictors of changes in community distribution, but the model performed poorly (marginal R2 = 0.07). This may reflect resilience of forest communities to climate change, but may also indicate a lag between climate change and community-level responses. By identifying changes in forest communities, our results can provide useful information for managers of vulnerable forest communities.|
|Luca Morreale||Using non-forest conditions to identify forest fragments and quantify tree growth across a changing landscape||Wednesday||3b||Sequoyah||Understanding landscape change using FIA data||This research aims to quantify the effects of forest fragmentation on tree growth and mortality in temperate forests across the northeastern US. Using FIA measurements of forest and non-forest subplot conditions and the locations of individual trees, we are able to identify forest edges and quantify differences between trees at the edge and those in the forest interior. Forest fragmentation is a pervasive landcover regime that needs to be included in our understanding of forest ecosystems.||Forest fragmentation is a ubiquitous land cover change phenomenon with important implications for the health and function of forested ecosystems worldwide. Designing studies that capture both the variability of regional land use regimes, as well as accurately quantifying ecosystem responses on the level of individual trees presents a distinct challenge. The USFSs Forest Inventory and Analysis (FIA) database is uniquely suited to address this multiscale research challenge due to the density, spatial range and detailed, on-the-ground measurements. The purpose of this study is to quantify the magnitude and spatial variability of tree growth and mortality responses to forest fragmentation in the northeastern US. Our research leverages the spatially explicit FIA measurements of forest and non-forest subplot conditions and the locations of individual trees. The condition-boundary dataset allows us to compare trees near the forest edge to those in the interior to understand both organismal and community responses to edge stimuli, such as increased sunlight. This study characterizes ecosystem responses to fragmentation across a range of forest types and land cover adjacencies.|