Figure 1 - uploaded by Tara M. Barrett
Content may be subject to copyright.
Contexts in source publication
Context 1
... 2 forest area ( km ); and (4) presence of invasive species ( km ). While this is only a selection of the important indicators for monitoring change in boreal forests, the selection spans the range of correlation between field and lidar data, providing a good understanding of tradeoffs among varying designs in achieving precise estimates of different indicators. To be able to provide consistent national estimates of both real-valued and classified attributes, we assumed that the basic plot layout, definitions, and core measurements would be the same in Alaska as for the rest of the U.S. While most of the country collects some forest health attributes on a 1/16 subsample of plots, we would use standardized simplified protocols on all plots for understory vegetation, downed woody material, and soils. In addition, the proposed Alaska protocol would collect a depth to frozen ground measurement, important to modeling permafrost, and some basic information on mosses. We tested all of the proposed protocols with field trials using crews of two or three people, and used a process of testing and revision to create a set of field measurements that could be consistently completed within 1 day by a three-person crew. An annualized inventory system, which would measure a panel of field plots from throughout Alaska each year, was dropped from consideration due to logistical and cost constraints. Instead, the proposed inventory would use five inventory units in the boreal region (fig. 1), each assessed within a time span of two to three years, with a 10 to 12 year interval between remeasurements. We tested the efficiency of various combinations of a double-sampling design with lidar and field data in the Tanana inventory unit (fig. 1) using two strata: the Legacy Stratum contained all the land included in a 1970s timber inventory, and the New Stratum contained more nonforest and black spruce but a very small expected proportion of white spruce and hardwoods. Any point within each stratum would have an equal probability of selection. Ideally, a preliminary sample would be used to estimate population variability of the four chosen indicators. Lacking resources to do this, we instead used available information to construct expected population variability. From recent inventories in south-central Alaska that used the identical plot layout, 340 field plots that contained boreal forest types (white spruce, black spruce, aspen, paper birch, cottonwood, and balsam poplar) were identified. We developed an expected proportion of forest in the Tanana inventory unit, using the NLCD 2001 classification of LandSat Thematic layers (Homer et al. 2007). A 1971-74 air photo classification provided data for estimating proportions of forest types in the Tanana inventory unit. The 340 field plots were reweighted by the expected probability of different forest types, and bootstrapped with 1000 repetitions to make an estimate of expected population variability for each of the four indicators for each stratum. Stratified sampling was used to estimate precision with varying numbers of field plots and double-sampling for regression was used to estimate precision with ...
Context 2
... 2 forest area ( km ); and (4) presence of invasive species ( km ). While this is only a selection of the important indicators for monitoring change in boreal forests, the selection spans the range of correlation between field and lidar data, providing a good understanding of tradeoffs among varying designs in achieving precise estimates of different indicators. To be able to provide consistent national estimates of both real-valued and classified attributes, we assumed that the basic plot layout, definitions, and core measurements would be the same in Alaska as for the rest of the U.S. While most of the country collects some forest health attributes on a 1/16 subsample of plots, we would use standardized simplified protocols on all plots for understory vegetation, downed woody material, and soils. In addition, the proposed Alaska protocol would collect a depth to frozen ground measurement, important to modeling permafrost, and some basic information on mosses. We tested all of the proposed protocols with field trials using crews of two or three people, and used a process of testing and revision to create a set of field measurements that could be consistently completed within 1 day by a three-person crew. An annualized inventory system, which would measure a panel of field plots from throughout Alaska each year, was dropped from consideration due to logistical and cost constraints. Instead, the proposed inventory would use five inventory units in the boreal region (fig. 1), each assessed within a time span of two to three years, with a 10 to 12 year interval between remeasurements. We tested the efficiency of various combinations of a double-sampling design with lidar and field data in the Tanana inventory unit (fig. 1) using two strata: the Legacy Stratum contained all the land included in a 1970s timber inventory, and the New Stratum contained more nonforest and black spruce but a very small expected proportion of white spruce and hardwoods. Any point within each stratum would have an equal probability of selection. Ideally, a preliminary sample would be used to estimate population variability of the four chosen indicators. Lacking resources to do this, we instead used available information to construct expected population variability. From recent inventories in south-central Alaska that used the identical plot layout, 340 field plots that contained boreal forest types (white spruce, black spruce, aspen, paper birch, cottonwood, and balsam poplar) were identified. We developed an expected proportion of forest in the Tanana inventory unit, using the NLCD 2001 classification of LandSat Thematic layers (Homer et al. 2007). A 1971-74 air photo classification provided data for estimating proportions of forest types in the Tanana inventory unit. The 340 field plots were reweighted by the expected probability of different forest types, and bootstrapped with 1000 repetitions to make an estimate of expected population variability for each of the four indicators for each stratum. Stratified sampling was used to estimate precision with varying numbers of field plots and double-sampling for regression was used to estimate precision with ...
Citations
... The results from this study indicate that a model-assisted approach can be used to generate defensible, design-unbiased estimates of total biomass, in a cost-effective manner, over large areas of relatively inaccessible forestland within Alaska. However, it should be noted that biomass is only one of many forest inventory attributes that are assessed in the FIA inventory program, and the optimal combination of plot and remote sensing data will vary significantly depending on the parameter of interest (Barrett et al., 2009). ...
Cost-effective monitoring of forest carbon resources is critical to the development of national policies and enforcement of international agreements aimed at reducing carbon emissions and mitigating the impacts of climate change. While carbon monitoring systems are often based on national forest inventories (NFI) utilizing a large sample of field plots, in remote regions the lack of transportation infrastructure often requires heavier reliance on remote sensing technologies, such as airborne lidar. The challenge motivating our research is that the efficacy of estimating carbon with lidar varies across the various carbon pools within forest ecosystems. Lidar measurements are typically highly correlated with aboveground tree carbon but are less strongly correlated with other carbon pools, such as down woody materials (DWM) and soil. Field measurements are essential to both (1) estimate soil and DWM carbon directly and (2) develop regression models to estimate tree carbon indirectly using lidar. With limited budgets and time, however, decision makers must find an optimal way to combine field measurements with lidar to minimize standard errors in carbon estimates for the various pools. We introduce a multi-objective binary programming formulation that quantifies the tradeoffs behind the competing objectives of minimizing standard errors for tree carbon, DWM carbon, and soil carbon. Using NFI and airborne lidar data from a remote boreal forest region of interior Alaska, we demonstrate the operational feasibility of the method and suggest that it is generalizable to other carbon sampling projects because of its generic mathematical structure.
