LIDAR Height Measures in Tropical and Coniferous Forests Dar A. Roberts 1, Matthew L. Clark 2, Phil E. Dennison 3, Kerry Q. Halligan 1, Bothaina Natour.

Slides:



Advertisements
Similar presentations
Original Figures for "Molecular Classification of Cancer: Class Discovery and Class Prediction by Gene Expression Monitoring"
Advertisements

Mike Smalligan, Research Forester Global Observatory for Ecosystem Services Forest Department, Michigan State University August 2011 Above Ground Biomass.
The Effects of Site and Soil on Fertilizer Response of Coastal Douglas-fir K.M. Littke, R.B. Harrison, and D.G. Briggs University of Washington Coast Fertilization.
US Forest Disturbance Trends observed with Landsat Time Series Samuel N. Goward 1 (PI), Jeffrey Masek 2, Warren Cohen 3, Gretchen Moisen 4, Chengquan Huang.
Objectives (BPS chapter 24)
Radar, Lidar and Vegetation Structure. Greg Asner TED Talk.
FOR 474: Forest Inventory Plot Level Metrics from Lidar Heights Other Plot Measures Sources of Error Readings: See Website.
WFM 6202: Remote Sensing and GIS in Water Management © Dr. Akm Saiful IslamDr. Akm Saiful Islam WFM 6202: Remote Sensing and GIS in Water Management Akm.
Princeton University Global Evaluation of a MODIS based Evapotranspiration Product Eric Wood Hongbo Su Matthew McCabe.
The Introduction of a Knowledge-based Approach and Statistical Methods to Make GIS-Compatible Climate Map Goshi Fujimoto at MIG seminar on 16th, May Daly,
Questions How do different methods of calculating LAI compare? Does varying Leaf mass per area (LMA) with height affect LAI estimates? LAI can be calculated.
Comparison of LIDAR Derived Data to Traditional Photogrammetric Mapping David Veneziano Dr. Reginald Souleyrette Dr. Shauna Hallmark GIS-T 2002 August.
Overview of Biomass Mapping The Woods Hole Research Center Alessandro Baccini, Wayne Walker and Ned Horning November 8 – 12, Samarinda, Indonesia.
Lecture 17 – Forest remote sensing  Reading assignment:  Ch 4.7, 8.23,  Kane et al., Interpretation and topographic correction of conifer forest.
Relationships Among Variables
What is RADAR? What is RADAR? Active detecting and ranging sensor operating in the microwave portion of the EM spectrum Active detecting and ranging sensor.
Geo479/579: Geostatistics Ch13. Block Kriging. Block Estimate  Requirements An estimate of the average value of a variable within a prescribed local.
Introduction OBJECTIVES  To develop proxies for canopy cover and canopy closure based on discrete-return LiDAR data.  To determine whether there is a.
September In Chapter 14: 14.1 Data 14.2 Scatterplots 14.3 Correlation 14.4 Regression.
Embedded sensor network design for spatial snowcover Robert Rice 1, Noah Molotch 2, Roger C. Bales 1 1 Sierra Nevada Research Institute, University of.
Science Enabled by New Measurements of Vegetation Structure (ICESat-II, DESDynI, etc.) Some Ecological Considerations Jon Ranson & Hank Shugart Co-Chairs.
Mapping Forest Canopy Height with MISR We previously demonstrated a capability to obtain physically meaningful canopy structural parameters using data.
Quantitative Estimates of Biomass and Forest Structure in Coastal Temperate Rainforests Derived from Multi-return Airborne Lidar Marc G. Kramer 1 and Michael.
Intro. To GIS Lecture 9 Terrain Analysis April 24 th, 2013.
Getting Ready for the Future Woody Turner Earth Science Division NASA Headquarters May 7, 2014 Biodiversity and Ecological Forecasting Team Meeting Sheraton.
Validating Wykoff's Model, Take 2: Equivalence tests and spatial analysis in a design- unbiased analytical framework Robert Froese, Ph.D., R.P.F. School.
REGENERATION IMPUTATION MODELS FOR INTERIOR CEDAR HEMLOCK STANDS Badre Tameme Hassani, M.Sc., Peter Marshall PhD., Valerie LeMay, PhD., Temesgen Hailemariam,
Development and evaluation of Passive Microwave SWE retrieval equations for mountainous area Naoki Mizukami.
A single basin-wide estimate for basic density of wood (0.69) has been employed to map above-ground biomass and carbon stocks across Amazonia (Fearnside.
__________. Introduction Importance – Wildlife Habitat – Nutrient Cycling – Long-Term Carbon Storage – Key Indicator for Biodiversity Minimum Stocking.
MODSCAG fractional snow covered area (fSCA )for central and southern Sierra Nevada Spatial distribution of snow water equivalent across the central and.
Spatial distribution of snow water equivalent across the central and southern Sierra Nevada Roger Bales, Robert Rice, Xiande Meng Sierra Nevada Research.
Thinning mixed-species stands of Douglas-fir and western hemlock in the presence of Swiss needle cast Junhui Zhao, Douglas A. Maguire, Douglas B. Mainwaring,
DN Ordinate Length DN Difference Estimating forest structure in tropical forested sites.
FSU Jena – Department of Earth Observation CREATION OF LARGE AREA FOREST BIOMASS MAPS FOR NE CHINA USING ERS-1/2 TANDEM COHERENCE Oliver Cartus (1), Christiane.
11/23/2015Slide 1 Using a combination of tables and plots from SPSS plus spreadsheets from Excel, we will show the linkage between correlation and linear.
Scaling Up Above Ground Live Biomass From Plot Data to Amazon Landscape Sassan S. Saatchi NASA/Jet Propulsion Laboratory California Institute of Technology.
LiDAR Remote Sensing of Forest Vegetation Ryan Anderson, Bruce Cook, and Paul Bolstad University of Minnesota.
Spectral and Structural Differences Between Coniferous and Broadleaf Forest derived from LIDAR and AVIRIS Dar A. Roberts 1, Keely L. Roth 2, Eliza Bradley.
Using Lidar to Identify and Measure Forest Gaps on the William B. Bankhead National Forest, Alabama Jeffrey Stephens 1, Dr. Luben Dimov 1, Dr. Wubishet.
Spectral Discrimination of Plant Functional Types and Species across diverse North American Ecosystems Dar A. Roberts 1, Keely L. Roth 2, Philip E. Dennison.
Remote Sensing of Forest Structure Van R. Kane College of Forest Resources.
One float case study The Argo float ( ) floating in the middle region of Indian Ocean was chosen for this study. In Figure 5, the MLD (red line),
Arizona Space Grant Consortium Statewide Symposium Arizona Space Grant Consortium Statewide Symposium Light Detection and Ranging (LiDAR) Survey of a Sky.
2011 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) Aihua Li Yanchen Bo
The Effects of Spatial Patterns on Canopy Cover Estimated by FVS (Forest Vegetation Simulator) A Thesis Defense by Treg Christopher Committee Members:
The Effect of Fuel Treatments on the Invasion of Nonnative Plants Kyle E. Merriam 1, Jon E. Keeley 1, and Jan L. Beyers 2. [1] USGS Western Ecological.
Citation: Kato, A.., L. M. Moskal., P. Schiess, M. Swanson, D. Calhoun and W. Stuetzel, LiDAR based tree crown surface reconstruction. Factsheet.
Using LiDAR to Measure the Urban Forest in DeKalb, Illinois Dustin P. Bergman and Thomas J. Pingel Northern Illinois University Department.
Airborne LiDAR requires purchase, but offers a number of advantages; Airborne LiDAR requires purchase, but offers a number of advantages; Spatial resolution.
Integrating LiDAR Intensity and Elevation Data for Terrain Characterization in a Forested Area Cheng Wang and Nancy F. Glenn IEEE GEOSCIENCE AND REMOTE.
Module 2.8 Overview and status of evolving technologies REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 1 Module 2.8 Overview.
Counting the trees in the forest
PADMA ALEKHYA V V L, SURAJ REDDY R, RAJASHEKAR G & JHA C S
Factsheet # 19 Understanding multiscale dynamics of landscape change through the application of remote sensing & GIS Hyperspectral Remote Sensing of Urban.
Section 11.1 Day 2.
Factsheet # 21 Understanding multiscale dynamics of landscape change through the application of remote sensing & GIS Quantifying Vertical and Horizontal.
Lidar Image Processing
Funding: National Park Service, U.S.G.S.
Figure 1. Spatial distribution of pinyon-juniper and ponderosa pine forests is shown for the southwestern United States. Red dots indicate location of.
Using Remote Sensing to Monitor Plant Phenology Response to Rain Events in the Santa Catalina Mountains Katheryn Landau Arizona Remote Sensing Center Mentors:
National Forest Inventory for Great Britain
Using dynamic aerosol optical properties from a chemical transport model (CTM) to retrieve aerosol optical depths from MODIS reflectances over land Fall.
No notecard for this quiz!!
By: Paul A. Pellissier, Scott V. Ollinger, Lucie C. Lepine
Spatial interpolation
Objectives (IPS Chapter 2.3)
A Comparison of Forest Biodiversity Metrics Using Field Measurements and Aircraft Remote Sensing Kaitlyn Baillargeon Scott Ollinger,
Sources of Variability in Canopy Spectra and the Convergent Properties of Plants Funding From: S.V. Ollinger, L. Lepine, H. Wicklein, F. Sullivan, M. Day.
CHAPTER 3 Describing Relationships
Presentation transcript:

LIDAR Height Measures in Tropical and Coniferous Forests Dar A. Roberts 1, Matthew L. Clark 2, Phil E. Dennison 3, Kerry Q. Halligan 1, Bothaina Natour 3, and Geoffrey G. Parker 4 1. Dept. of Geography, Univ. California, Santa Barbara CA U.S.A, 2. Dept. of Geography and Global Studies, Sonoma State Univ. Rohnert Park CA, USA 3. Dept of Geography, Univ. Utah, Salt Lake City, UH, USA, 4. Smithsonian Environmental Research Center, Edgewater, MD , USA Abstract Combined LIDAR and hyperspectral measures have the potential of improving our ability to estimate carbon stocks, through biomass-height relations and carbon fluxes, through improved maps of forest species and physiology. While strong relationships have been observed between LIDAR-derived heights and above ground biomass, significant questions remain, such as whether these relationships are global or site specific depending on management, disturbance and climate. Furthermore, it is unclear the extent to which hyperspectral data complement LIDAR by providing measures of health, physiology and species. To address these questions, we have begun a program evaluating LIDAR and hyperspectral data at six highly variable sites ranging from tropical rainforest (La Selva), western coniferous forest (Wind River and Sierra Nevada), central United States (Yellowstone) to east-coast broadleaf deciduous forest (Harvard Forest, SERC). In this poster we report upon initial analysis of LIDAR data at La Selva, Wind River and Yellowstone. Two approaches were employed to generate a Digital Terrain Model (DTM) and calculate a Digital Canopy Model (DCM) from LIDAR. Both relied upon the concept of identifying bare earth between crowns, but relied on different means for determining appropriate window sizes and interpolating a DTM. LIDAR heights were evaluated for individual trees at all three sites and plots for two sites. In general, all sites showed a high correlation between LIDAR heights and measured heights that improved at plot scales. LIDAR tended to underestimate tree height, with errors increasing for more dense stands and for shorter trees. The most accurate tree heights were estimated at La Selva for pasture trees and for open stands at Yellowstone. Common error sources included rugged terrain, the inability to estimate an accurate DTM for dense stands and sparse LIDAR point density. Biomass showed a near-linear relationship to LIDAR-height at La Selva, but appears non-linear at Wind River. Study Areas Summary References Results Although preliminary, results from these three sites are encouraging including: LIDAR estimated tree height and measured height were highly correlated across a range of conditions from rugged Tropical Rainforest and Old growth coniferous forest to pasture trees, open stands and second growth forest Relationships between LIDAR tree height and field heights improved at plot scales All three sites showed a similar bias with a LIDAR under-estimates of tree height. Bias increased for shorter trees except at La Selva Error sources included a failure to accurately develop a DTM in dense stands or rugged terrain and a failure to get a return from tree tops at low LIDAR density. These errors varied with land-cover class and stand age Biomass and LIDAR height were linearly correlated at La Selva, but non-linear at Wind River. Non-linearity at Wind River may be partly due to damaged tree tops Future analysis will focus on integrating LIDAR structural analysis with hyperspectral analysis. Hyperspectral analysis will focus on estimating foliage physiological state and tree species Brown, S., 1997, Estimating biomass and biomass change of tropical forests: a primer Clark, M.L., Clark, D.B., and Roberts, D.A., 2004, Small-footprint lidar estimation of sub-canopy elevation and tree height in a tropical rain forest landscape, Remote Sens. Environ. 91, Natour, B., P.E. Dennison, and D.A. Roberts, Estimation of tree height using small-footprint lidar measurements in the Wind River Experimental Forest. Eleventh Biennial USDA Forest Service Remote Sensing Applications Conference, Apr 24-28, 2006, Salt Lake City, UT Methods 1. Tree Heights at La Selva 4. Error Sources Acknowledgements: This research was funded in part by “Multisite Integration of LIDAR and Hyperspectral Data for Improved Estimation of Carbon Stocks and Exchanges”, NASA Carbon Cycle Science grant NNG05GE56G Figure 1: Study areas including (a) La Selva Costa Rica and (b) Yellowstone. Wind River is not shown. at left). Three study areas are included in this poster (Figure 1), including Tropical Rainforest (La Selva), western hemlock/Douglas-fir (Wind River) and mixed conifer- broadleaf forest (Yellowstone). LIDAR data included FLIMAP (La Selva), Aeroscan (Wind River) and Optech ALTM 1233 (Yellowstone). Hyperspectral data exist at all three sites, sampled by AVIRIS, HYDICE or HYMAP but are not reported on here. Field data included tree height for individual trees (80 La Selva, ~ 200 Wind River, 1124 Yellowstone) and 32 plots at La Selva and 20 plots at Wind River. Tree height was generated as a Digital Canopy Model (DCM) calculated as the difference between the Digital Surface Model (DSM) and a Digital Terrain Model (DTM (Figure 2). Analysis focused on comparison LIDAR estimated heights to field measured trees and plots. Several approaches were employed for developing the DTM including: Inverse Distance Weighting or Ordinary Kriging of minima from DSM with variable window sizes (Figure 3). This was the approach used at La Selva and Wind River Segmentation using LIDAR intensity to identify crown objects. Use of cumulative distribution function or semiavariograms to identify window size for creating the DTM for each LIDAR-intensity class. Evaluation of optimal window size to minimize local slope errors (Figure 4). This was the approach used at Yellowstone. Figure 2: General procedure for generating a DCM 2. Tree Heights at Wind River 3. Tree Heights at Yellowstone  Tree heights were evaluated for individual trees and plots for two dominant land-cover classes. An example DCM is shown in Figure 5  Accuracies of individual tree heights varied with land- cover class, and were generally better for pasture trees than closed forest but were underestimated in both cases (Figure 6)  At plot scales tree heights were more accurate, producing a lower RMS, higher r 2 and lower underestimate (Figure 7) Figure 5: Example DCM from La Selva Figure 8, right: Shows a plot of LIDAR estimated tree height (x) compared to measured tree height (y). A best fit equation produced a slope slightly below 1, an intercept of 5.4 m and r 2 of A 1:1 line is shown as dashed. Figure 15, below: Relationship between biomass, determined using allometry from DBH and plot averaged heights at Wind River. Red corresponds to the youngest stands, deep blue older stands and green, old growth. A comparison of regression relationships for all sites demonstrates several important patterns including highest correlations for the most open stands (Pasture trees, Yellowstone), poorest correlations for dense stands (La Selva Old, Wind River second growth) and an overall underestimation of tree height (Table 1) Plot scale relationships showed higher correlations and lower error at both sites compared (smaller RMSE, smaller intercept: Table 1, in grey) Biomass/LIDAR height relationships were near linear at La Selva (Figure 14) but non-linear at Wind River (Figure 15), potentially due to damaged tree tops. In both cases, pooling age classes improved relationships. Biomass at Wind River far exceeded biomass at La Selva. 898 trees used in initial evaluation, two windows selected, with 7 m for non-forest and 13 m for forest Analysis of individual trees in open stands produced the highest r 2 (0.90) and lowest RMS (2.3 m) of all sites (Figure 10) A positive intercept implies either the surface was overestimated or the tree top was missed Table 1: Regression between LIDAR height (x) and measured height (y) at the tree study locations including RMSE for each site. Plot statistics are shown in grey. Figure 11: Plots showing RMSE and Mean Squared Error as a function of slope class (a) and land- cover class (b). Height estimates were least accurate on steep slopes. Heights were underestimated in old growth and tall forest and overestimated in more open sites. From Clark et al., a) b) Lidar Height (m) Elevation (m) Vegetation Height (m) Iteration 1: 20 mIteration 2: 15 mIteration 3: 10 m 2. Calculate point-DTM elevation residuals remove x,y,z points (+) with residuals > 0.25 m 3. Interpolate DTM from filtered xyz points or 5-m IDW 1-m IDW 1-m OK 1. DSM grid overlay find 33-cm DSM minima cell in each coarse-scale grid cell (e.g., 20 x 20 m) Iteration Processing Steps Figure 3: Generating a DTM with variable window sizes Figures 2 & 3 adapted from Clark et al., 2004 Lidar intensityThreshold Choice Cumulative Distributions Intensity 9 m window 11 m window13 m window a)b) c) Local Slope d) Figure 4: Segmentation using LIDAR intensity (a) quantitative selection of a threshold from overlapping gaussian distributions (b); variable window size selection from cumulative distribution functions (c); Assessing DTM quality using local slope (d) Figure 6: S catterplot of LIDAR tree height (x) to measured tree height (y) for old growth (a) and pasture trees (b). Adapted from From Clark et al., Figure 7: Plot scale relationships between LIDAR tree height (x) and measured tree height (y). Relationships were better for tree stems only plotted against mean height (a) compared to plot averages that included all of the stems, such as palms (b). Adapted from Clark et al., Site Comparisons and Biomass Errors originate in the failure to locate bare ground (underestimation of heights), failure to capture variable terrain (under or over estimation) and failure to sample tree tops At La Selva, errors were highest on steeper slopes (Figure 11a) and varied with land-cover classe (Figure 11b). Heights were underestimated in old growth forest, but overestimated in pastures At Wind River, height estimate were low, with the largest error in dense, shorter stands. In this area, the DTM most likely over estimated true surface height At Yellowstone, height errors increased as LIDAR point density decreased (Figure 12). Highest height accuracies were achieved with point densities in excess of 3 (Figure 13) y=8.92x R 2 = 0.75 y=59.96x-1264 R 2 = 0.56 Figure 9, left: Shows a plot of LIDAR estimated tree height (x) compared to measured tree height (y) for 15 forest plots. Error bars show +/- 1 standard deviation. At plot scales the r 2 improved considerably and intercept decreased. Figure 10: Shows a plot of LIDAR estimated tree height (x) compared to measured tree height (y) for 898 trees in open stands in Yellowstone. A 1:1 line is shown as dashed. These sites had the highest correlation and lowest error, but still had a positive intercept (2.34 m) and slope less than one (0.941). a) b) Figure 12: Average LIDAR point density at Yellowstone (a) and dependence of error on point density (b). A majority of the tree crowns were sampled by less than 1 LIDAR point/m 2. Figure 13: Improved height estimates using only crowns with point densities of 3 or greater Figure 14, right: Relationship between biomass, determined using allometry from DBH and plot averaged heights at La Selva. Red corresponds to plantation trees, green old growth.  Accuracies of individual trees were comparable to La Selva, generally resulting in an underestimate for shorter trees in closed second growth stands (Figure 8)  Accuracies improved at plot scales, resulting in an intercept closer to zero (3.88 m), a slope near 1 (1.057) and high r 2 of 0.95 (Figure 9) b) a)b) a)b)