Variable selection strategies for nearest neighbor imputation methods used in remote sensing based forest inventory

2012 ◽  
Vol 38 (5) ◽  
pp. 557-569 ◽  
Author(s):  
Petteri Packalén ◽  
Hailemariam Temesgen ◽  
Matti Maltamo
Keyword(s):  
Remote Sensing ◽  
Variable Selection ◽  
Forest Inventory ◽  
Nearest Neighbor ◽  
Imputation Methods ◽  
Selection Strategies ◽  
Nearest Neighbor Imputation

scholarly journals Landscape-scale parameterization of a tree-level forest growth model: a k-nearest neighbor imputation approach incorporating LiDAR data

2010 ◽  
Vol 40 (2) ◽  
pp. 184-199 ◽  
Author(s):  
Michael J. Falkowski ◽  
Andrew T. Hudak ◽  
Nicholas L. Crookston ◽  
Paul E. Gessler ◽  
Edward H. Uebler ◽  
...  
Keyword(s):  
Forest Inventory ◽  
Nearest Neighbor ◽  
Forest Growth ◽  
Tree Level ◽  
Lidar Data ◽  
Inventory Data ◽  
K Nearest Neighbor ◽  
Forest Inventory Data ◽  
Forest Growth Model ◽  
Nearest Neighbor Imputation

Sustainable forest management requires timely, detailed forest inventory data across large areas, which is difficult to obtain via traditional forest inventory techniques. This study evaluated k-nearest neighbor imputation models incorporating LiDAR data to predict tree-level inventory data (individual tree height, diameter at breast height, and species) across a 12 100 ha study area in northeastern Oregon, USA. The primary objective was to provide spatially explicit data to parameterize the Forest Vegetation Simulator, a tree-level forest growth model. The final imputation model utilized LiDAR-derived height measurements and topographic variables to spatially predict tree-level forest inventory data. When compared with an independent data set, the accuracy of forest inventory metrics was high; the root mean square difference of imputed basal area and stem volume estimates were 5 m2·ha–1 and 16 m3·ha–1, respectively. However, the error of imputed forest inventory metrics incorporating small trees (e.g., quadratic mean diameter, tree density) was considerably higher. Forest Vegetation Simulator growth projections based upon imputed forest inventory data follow trends similar to growth projections based upon independent inventory data. This study represents a significant improvement in our capabilities to predict detailed, tree-level forest inventory data across large areas, which could ultimately lead to more informed forest management practices and policies.

Estimating cavity tree and snag abundance using negative binomial regression models and nearest neighbor imputation methods

2009 ◽  
Vol 39 (9) ◽  
pp. 1749-1765 ◽  
Author(s):  
Bianca N.I. Eskelson ◽  
Hailemariam Temesgen ◽  
Tara M. Barrett
Keyword(s):  
Regression Models ◽  
Forest Cover ◽  
Nearest Neighbor ◽  
Negative Binomial ◽  
Negative Binomial Regression ◽  
Remotely Sensed Data ◽  
Imputation Methods ◽  
Nearest Neighbor Imputation ◽  
Us Forest Service ◽  
Cavity Tree

Cavity tree and snag abundance data are highly variable and contain many zero observations. We predict cavity tree and snag abundance from variables that are readily available from forest cover maps or remotely sensed data using negative binomial (NB), zero-inflated NB, and zero-altered NB (ZANB) regression models as well as nearest neighbor (NN) imputation methods. The models were developed and fit to data collected by the Forest Inventory and Analysis program of the US Forest Service in Washington, Oregon, and California. For predicting cavity tree and snag abundance per stand, all three NB regression models performed better in terms of mean square prediction error than the NN imputation methods. The most similar neighbor imputation, however, outperformed the NB regression models in predicting overall cavity tree and snag abundance.

K Nearest Neighbor Method for Forest Inventory Using Remote Sensing Data

2007 ◽  
Vol 44 (2) ◽  
pp. 149-165 ◽  
Author(s):  
Qingmin Meng ◽  
Chris J. Cieszewski ◽  
Marguerite Madden ◽  
Bruce E. Borders
Keyword(s):  
Remote Sensing ◽  
Forest Inventory ◽  
Nearest Neighbor ◽  
Remote Sensing Data ◽  
K Nearest Neighbor ◽  
Sensing Data

scholarly journals Tree Crowns Cause Border Effects in Area-Based Biomass Estimations from Remote Sensing

2021 ◽  
Vol 13 (8) ◽  
pp. 1592
Author(s):  
Nikolai Knapp ◽  
Andreas Huth ◽  
Rico Fischer
Keyword(s):  
Remote Sensing ◽  
Forest Inventory ◽  
Canopy Height ◽  
Biomass Estimation ◽  
Inventory Data ◽  
Border Effects ◽  
Tree Crowns ◽  
Forest Inventory Data ◽  
Estimation Uncertainty ◽  
Voxel Space

The estimation of forest biomass by remote sensing is constrained by different uncertainties. An important source of uncertainty is the border effect, as tree crowns are not constrained by plot borders. Lidar remote sensing systems record the canopy height within a certain area, while the ground-truth is commonly the aboveground biomass of inventory trees geolocated at their stem positions. Hence, tree crowns reaching out of or into the observed area are contributing to the uncertainty in canopy-height–based biomass estimation. In this study, forest inventory data and simulations of a tropical rainforest’s canopy were used to quantify the amount of incoming and outgoing canopy volume and surface at different plot sizes (10, 20, 50, and 100 m). This was performed with a bottom-up approach entirely based on forest inventory data and allometric relationships, from which idealized lidar canopy heights were simulated by representing the forest canopy as a 3D voxel space. In this voxel space, the position of each voxel is known, and it is also known to which tree each voxel belongs and where the stem of this tree is located. This knowledge was used to analyze the role of incoming and outgoing crowns. The contribution of the border effects to the biomass estimation uncertainty was quantified for the case of small-footprint lidar (a simulated canopy height model, CHM) and large-footprint lidar (simulated waveforms with footprint sizes of 23 and 65 m, corresponding to the GEDI and ICESat GLAS sensors). A strong effect of spatial scale was found: e.g., for 20-m plots, on average, 16% of the CHM surface belonged to trees located outside of the plots, while for 100-m plots this incoming CHM fraction was only 3%. The border effects accounted for 40% of the biomass estimation uncertainty at the 20-m scale, but had no contribution at the 100-m scale. For GEDI- and GLAS-based biomass estimates, the contributions of border effects were 23% and 6%, respectively. This study presents a novel approach for disentangling the sources of uncertainty in the remote sensing of forest structures using virtual canopy modeling.

scholarly journals High-resolution forest carbon flux mapping and monitoring in the Pacific Northwest with time since disturbance and disturbance legacies inferred from remote sensing and inventory data

2016 ◽  
Author(s):  
Huan Gu ◽  
Christopher A. Williams ◽  
Bardan Ghimire ◽  
Feng Zhao ◽  
Chengquan Huang
Keyword(s):  
Remote Sensing ◽  
Pacific Northwest ◽  
Forest Inventory ◽  
Stand Age ◽  
Net Ecosystem Productivity ◽  
Fine Scale ◽  
Ecosystem Productivity ◽  
Inventory Data ◽  
Forest Inventory Data ◽  
The Pacific

Abstract. Assessment of forest carbon storage and uptake is central to understanding the role forests play in the global carbon cycle and policy-making aimed at mitigating climate change. Current U.S. carbon stocks and fluxes are monitored and reported at fine-scale regionally, or coarse-scale nationally. We proposed a new methodology of quantifying carbon uptake and release across forested landscapes in the Pacific Northwest (PNW) at a fine scale (30 m) by combining remote-sensing based disturbance year, disturbance type, and aboveground biomass with forest inventory data in a carbon modelling framework. Time since disturbance is a key intermediate determinant that aided the assessment of disturbance-driven carbon emissions and removals legacies. When a recent disturbance was detected, time since disturbance can be directly determined by remote sensing-derived disturbance products; and if not, time since last stand-clearing was inferred from remote sensing-derived 30 m biomass map and field inventory-derived species-specific biomass regrowth curves. Net ecosystem productivity (NEP) was further mapped based on carbon stock and flux trajectories that described how NEP changes with time following harvest, fire, or bark beetle disturbances of varying severity. Uncertainties from biomass map and forest inventory data were propagated by probabilistic sampling to provide a probabilistic, statistical distribution of stand age and NEP for each forest pixel. We mapped mean, standard deviation and statistical distribution of stand age and NEP at 30 m in the PNW region. Our map indicated a net ecosystem productivity of 5.2 Tg C y−1 for forestlands circa 2010 in the study area, with net uptake in relatively mature (> 24 year old) forests (13.6 Tg C y−1) overwhelming net negative NEP from tracts that have seen recent harvest (−6.4 Tg C y−1), fires (−0.5 Tg C y−1), and bark beetle outbreaks (−1.4 Tg C y−1). The approach will be applied to forestlands in other regions of the conterminous U.S. to advance a more comprehensive monitoring, mapping and reporting the carbon consequences of forest change across the U.S.

Sign in / Sign up

Export Citation Format

Share Document