 “Best” permutation for each model type determined by ranking RMSE, variance ratio, bias, and R 2 (Table 5).  Arizona favored less complex variable permutations.

Slides:



Advertisements
Similar presentations
Beyond Spectral and Spatial data: Exploring other domains of information GEOG3010 Remote Sensing and Image Processing Lewis RSU.
Advertisements

Predicting and mapping biomass using remote sensing and GIS techniques; a case of sugarcane in Mumias Kenya Odhiambo J.O, Wayumba G, Inima A, Omuto C.T,
Linking In situ Measurements, Remote Sensing, and Models to Validate MODIS Products Related to the Terrestrial Carbon Cycle Peter B. Reich, University.
A framework for selecting and blending retrievals from bio-optical algorithms in lakes and coastal waters based on remote sensing reflectance for water.
Characterization of radiance uncertainties for SeaWiFS and Modis-Aqua Introduction The spectral remote sensing reflectance is arguably the most important.
Spatial monitoring of older forest for the Northwest Forest Plan Janet Ohmann 1, Matt Gregory 2, Heather Roberts 2, Robert Kennedy 2, Warren Cohen 1, Zhiqiang.
Spatial monitoring of late-successional forest habitat over large regions with nearest-neighbor imputation Janet Ohmann 1, Matt Gregory 2, Heather Roberts.
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.
Comparison of biomass allometric approaches for regional scale carbon mapping Scott Powell – Montana State University Robert Kennedy – Boston University.
Mapping change in live and dead forest biomass with Landsat time-series, remeasured plots, and nearest-neighbor imputation Janet Ohmann 1, Matt Gregory.
Basic geostatistics Austin Troy.
Gradients or hierarchies? Which assumptions make a better map? Emilie B. Grossmann Janet L. Ohmann Matthew J. Gregory Heather K. May.
All for one or One for All? Mapping many species individually vs. simultaneously with random forest. Emilie Henderson, Janet Ohmann, Matthew Gregory, Heather.
Estimating forest structure in wetlands using multitemporal SAR by Philip A. Townsend Neal Simpson ES 5053 Final Project.
Gradient Nearest Neighbor (GNN) Method for Local-Scale Basal Area Mapping: FIA 2005 Symposium Interpolation Contest Kenneth B. Pierce Jr., Matthew J. Gregory*
ASTER image – one of the fastest changing places in the U.S. Where??
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
Curve-Fitting Regression
An Historically Consistent and Broadly Applicable MRV System Based on Lidar Sampling and Landsat Time-series Warren B. Cohen 1, Hans-Erik Andersen 1, Sean.
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.
CHAPTER 3 Community Sampling and Measurements From: McCune, B. & J. B. Grace Analysis of Ecological Communities. MjM Software Design, Gleneden Beach,
CHANGE DETECTION METHODS IN THE BOUNDARY WATERS CANOE AREA Thomas Juntunen.
Center for Satellite Applications and Research (STAR) Review 09 – 11 March 2010 Correction of Vegetation Time Series for Long Term Monitoring Marco Vargas¹.
Slides 13b: Time-Series Models; Measuring Forecast Error
Ensemble Learning (2), Tree and Forest
METHOD A sample design for Landsat-based estimation of national trends in forest disturbance and regrowth Presented by Kennedy at the Joint Workshop on.
Compton Tucker, GSFC Sassan Satchi, JPL Jeff Masek, GSFC Rama Nemani, ARC Diane Wickland, HQ Terrestrial Biomass Pilot Product: Estimating Biomass and.
These early results were obtained using one year’s set of FIA field data, DISTANCE: EUCLIDEAN WEIGHTING FUNCTION: NO WEIGHTS. NUMBER OF PLOTS: 696 NUMBER.
Accuracy Assessment. 2 Because it is not practical to test every pixel in the classification image, a representative sample of reference points in the.
Update of the National Biomass and Carbon Dataset 2000 using ALOS PALSAR L-band data Josef Kellndorfer, Wayne Walker, Oliver Cartus The Woods Hole Research.
RMRS Forest Inventory and Analysis User Group Meeting 2010 April 13.
USDA Forest Service Remote Sensing Applications Center Forest Inventory and Analysis New Technology How is FIA integrating new technological developments.
An Object Oriented Algorithm for Extracting Geographic Information from Remotely Sensed Data Zachary J. Bortolot Assistant Professor of Geography Department.
Mapping Forest Canopy Height with MISR We previously demonstrated a capability to obtain physically meaningful canopy structural parameters using data.
Seto, K.C., Woodcock, C.E., Song, C. Huang, X., Lu, J. and Kaufmann, R.K. (2002). Monitoring Land-Use change in the Pearl River Delta using Landsat TM.
Review of Statistics and Linear Algebra Mean: Variance:
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.
1 Exploiting Multisensor Spectral Data to Improve Crop Residue Cover Estimates for Management of Agricultural Water Quality Magda S. Galloza 1, Melba M.
Biomass Mapping The set of field biomass training data and the MODIS observations were used to develop a regression tree model (Random Forest). Biomass.
Analysis of Forest Fire Disturbance in the Western US Using Landsat Time Series Images: Haley Wicklein Advisors: Jim Collatz and Jeff Masek,
Image Classification 영상분류
Translation to the New TCO Panel Beverly Law Prof. Global Change Forest Science Science Chair, AmeriFlux Network Oregon State University.
Development and evaluation of Passive Microwave SWE retrieval equations for mountainous area Naoki Mizukami.
Curve-Fitting Regression
Remote Sensing Realities | June 2008 Remote Sensing Realities.
Canada Centre for Remote Sensing Field measurements and remote sensing-derived maps of vegetation around two arctic communities in Nunavut F. Zhou, W.
Integration of biosphere and atmosphere observations Yingping Wang 1, Gabriel Abramowitz 1, Rachel Law 1, Bernard Pak 1, Cathy Trudinger 1, Ian Enting.
Challenges of monitoring natural disturbance processes using remotely sensed data in North Coast and Cascades Network: comparison of approaches Natalya.
BIOPHYS: A Physically-based Algorithm for Inferring Continuous Fields of Vegetative Biophysical and Structural Parameters Forrest Hall 1, Fred Huemmrich.
Variations in Continental Terrestrial Primary Production, Evapotranspiration and Disturbance Faith Ann Heinsch, Maosheng Zhao, Qiaozhen Mu, David Mildrexler,
Slide 1 NATO UNCLASSIFIEDMeeting title – Location - Date Satellite Inter-calibration of MODIS and VIIRS sensors Preliminary results A. Alvarez, G. Pennucci,
Figure 1. Map of study area. Heavy solid polygon defines “Cascade Mountains” for the purposes of this study. The thin solid line divides the Cascade Mountains.
Goal: to understand carbon dynamics in montane forest regions by developing new methods for estimating carbon exchange at local to regional scales. Activities:
Evapotranspiration Estimates over Canada based on Observed, GR2 and NARR forcings Korolevich, V., Fernandes, R., Wang, S., Simic, A., Gong, F. Natural.
LANDSCAPE SCALE MONITORING OF FOREST TREATMENTS AND DISTURBANCE S.E. Sesnie 1,2, A.D. Olsson 1, B.G. Dickson 1, A. Leonard 3, V. Stein Foster 3 and C.
Beyond Spectral and Spatial data: Exploring other domains of information: 3 GEOG3010 Remote Sensing and Image Processing Lewis RSU.
Variance of Similar Neighbors compared to Random Imputation Nearest Neighbor Conference August 28-30, 2006 Kenneth B. Pierce Jr and Janet L. Ohmann Forestry.
Development of an Ensemble Gridded Hydrometeorological Forcing Dataset over the Contiguous United States Andrew J. Newman 1, Martyn P. Clark 1, Jason Craig.
References: 1)Ganguly, S., Samanta, A., Schull, M. A., Shabanov, N. V., Milesi, C., Nemani, R. R., Knyazikhin, Y., and Myneni, R. B., Generating vegetation.
Forecasting. Model with indicator variables The choice of a forecasting technique depends on the components identified in the time series. The techniques.
Lecture 9 Forecasting. Introduction to Forecasting * * * * * * * * o o o o o o o o Model 1Model 2 Which model performs better? There are many forecasting.
Yueyang Jiang1, John B. Kim2, Christopher J
Temporal Classification and Change Detection
Alexander Loew1, Mike Schwank2
Western Mensurationists Meeting 2016
HIERARCHICAL CLASSIFICATION OF DIFFERENT CROPS USING
Operational Regional Carbon Assessment
Potential Landsat Contributions
Hyperspectral Image preprocessing
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.
Presentation transcript:

 “Best” permutation for each model type determined by ranking RMSE, variance ratio, bias, and R 2 (Table 5).  Arizona favored less complex variable permutations (except for certain regression tree methods).  Likely due to stronger empirical relationships between biomass and Landsat spectral data compared to Minnesota data.  Compared the “best” permutation across model types using same four criteria.  Regression tree methods yielded the lowest error for both AZ and MN.  GNN best preserved variance in Arizona.  RMA best preserved variance in Minnesota.  RMA and SGB were the least biased in Arizona.  GNN was the least biased in Minnesota.  Overall, RF was the best model type for optimizing all four criteria, and was superior to other methods at minimizing prediction error. Empirical Modeling  Data divided into model (2/3) and validation (1/3) sets.  Empirical biomass models developed for each of six modeling techniques and eight variable permutations.  Eight variable permutations based on four categories of predictor variables (Table 4). 1. RAW: Raw Landsat bands only (1) 2. RAW.SPECT: Raw Landsat bands + Derived spectral indices (1 + 2) 3. RAW.30: Raw Landsat bands + Derived spectral variables + Topographic variables ( ) 4. RAW.ALL: Raw Landsat + Derived spectral indices + Topographic variables + Climate Variables ( ) 5. TC: Tasseled Cap bands only (Brightness, Greenness, Wetness) 6. TC.SPECT: TC bands + Derived spectral indices (TC Bands + 2) 7. TC.30: TC bands + Derived spectral indices + Topographic variables (TC Bands ) 8. TC.ALL: TC bands + Derived spectral indices + Topographic variables + Climate Variables (TC Bands )  Biomass predictions for each model type/permutation validated in terms of RMSE, variance ratio, bias, and observed v. predicted R 2.  Performance of each permutation ranked to determine “best” permutation per model type and per sample scene.  RMA, GNN, and TC models selected for creation of 20 + year biomass trajectories that were smoothed with LanDTrenDR.  Biomass trajectories validated using biomass change data derived from successively remeasured FIA data. Introduction and Background  North American Carbon Program (NACP) and Forest Inventory and Analysis (FIA) goals:  Improve methods for quantification of forest disturbance and regrowth processes (Goward et al., 2008).  Characterize forest disturbance and regrowth dynamics by integrating FIA data and Landsat time-series data to model aboveground forest biomass dynamics.  Many approaches for empirical modeling of biomass using optical remote sensing data, but little consensus on best method.  Little is known about multi-temporal prediction of biomass.  Three study objectives: 1. Evaluate a suite of six statistical techniques for modeling biomass.  Reduced Major Axis (RMA) regression  Orthogonal regression technique.  Maintains data variance structure.  Generalized Additive Models (GAMs)  Generalization of multiple regression.  Useful for detecting non-linear relationships.  Gradient Nearest Neighbor (GNN) imputation  Assigns plot attributes to each unmapped pixel based upon a multivariate distance in gradient space.  Maintains co-variance structure among response variables.  Cubist  Regression tree technique.  Fits a multivariate linear model to each rule.  Stochastic Gradient Boosting (SGB)  Refinement of traditional regression tree analysis.  Potentially less sensitive to outliers or unbalanced data and more resistant to over-fitting.  Random Forests (RF)  Ensemble regression tree approach.  Constructs numerous small regression trees from which predictions are averaged. 2. Evaluate the effect of inclusion of ancillary predictor variables.  Spectral variables, derived spectral indices, topographically-derived variables, and climate variables. 3. Assess how the choices of model/predictor variables affect predictions of biomass change.  Limitations to empirical modeling of biomass using optical remote sensing data are well known.  Solution hinges upon leveraging the temporal information contained within the Landsat time-series (Kennedy et al., 2007).  Trends across 20 + year trajectories of biomass are noteworthy, especially in instances of forest disturbance or regrowth.  Smoothing biomass trajectories with a curve-fitting algorithm enables a more accurate evaluation of biomass dynamics (Figure 1).  The algorithm, Landsat Detection of Trends in Disturbance and Recovery (LanDTrenDR) uses segmentation rules to summarize the temporal progression of any spectral or derived spectral index (e.g. biomass) of Landsat imagery (Kennedy et al, in prep.).  Captures both abrupt and slow disturbance as well as diverse sequences of disturbance and regrowth. Comparison of methods to model aboveground biomass for derivation of 20 + year trajectories Powell, S.L 1, Kennedy, R.E. 2, Healey, S.P. 3, Pierce, K.B. 4, Cohen, W.B. 1, Moisen, G.G. 3, Ohmann, J.L. 1 1 U.S.D.A. Forest Service, Pacific Northwest Research Station, Corvallis, OR Dept. of Forest Science, Oregon State University, Corvallis, OR U.S.D.A. Forest Service, Rocky Mountain Research Station, Ogden, UT WA Dept. of Fish and Wildlife, Olympia, WA Landsat data development  Biennial Landsat time-series stacks acquired for two sample scenes (Table 1).  Arizona and Minnesota (Figure 2).  Geometrically corrected and radiometrically calibrated to surface reflectance with the LEDAPS algorithm (Masek et al., 2006).  Relative radiometric normalization to a common reference image (noted by * in table) using the Multivariate Alteration Detection (MAD) method (Canty et al., 2004). Stat.AZMN n Mean4764 Stdev5946 Min00 Max ArizonaMinnesota 06/21/198506/28/ /26/198608/21/ /02/198909/14/ /19/199007/31/ /22/199108/19/ /27/199308/24/ /14/199408/14/ /22/199709/04/ /25/199808/06/ /14/2000*07/24/ /12/200207/05/ /09/200309/05/2003* 08/31/200508/06/ /03/200609/13/2006 Model TypeAZMN RMATCTC.SPECT GAMSTC.SPECTTC.ALL GNNTCRAW.ALL CUBISTRAW.ALLTC.ALL SGBTCTC.ALL RFTC.ALLRAW.ALL StateMeas. Yrs.Inventory TypeUse# Plots AZ1995Periodic Cycle 2Remeasure Validation Annual Cycle 3Model136 MN Annual Cycle 12Model Annual Cycle 13Remeasure Validation 285 Table 3. Biomass summary statistics for AZ Cycle 3 and MN Cycle 12 inventories. Table 2. Summary of field measurement years, inventory type, how the data were used, and number of plots for each of the four FIA inventories used in this study. Figure 2. Arizona (Landsat path/row 37/35) and Minnesota (Landsat path/row 27/27) study scenes. Table 1. Landsat time- series stacks for two sample scenes, with normalization reference images noted by *. FIA data development  Plot-level estimates of live aboveground tree biomass developed for four inventories (Table 2).  Only used homogenous “single condition plots”.  Calculated biomass change for a subset of plots that were remeasured between successive inventories.  Summary biomass statistics for the model building inventories shown in Table 3. Table 4. Four groups of predictor variables used to construct the eight predictor variable permutations for empirical modeling. Climate variables were 18-year mean annual values derived from DAYMET (Thornton et al., 1997). Disturbance Index from Healey et al., Potential Relative Radiation from Pierce et al., Raw Landsat Bands (28.5 m) 2. Derived Spectral Indices (28.5 m) 3. Topographic Variables (28.5 m) 4. Climate Variables (1 km) Band 1NDVIElevationTemperature Band 2Tasseled-cap BrightnessSlopeGrowing Degree Days Band 3Tasseled-cap GreennessPotential Relative RadiationWater Vapor Pressure Band 4Tasseled-cap WetnessPrecipitation Band 5Tasseled-cap Angle (TCG/TCB) Shortwave Radiation Band 7Tasseled-cap Distance (sqrt(TCB2+TCG2)) Disturbance Index Results and Discussion Methods  Accuracy of biomass predictions varied by model type and variable permutation.  In terms of RMSE (Figure 4) the differences between model types were generally smaller than the differences between variable permutations within a model type.  Regression tree methods (Cubist, SGB, and RF) generally favored more complex variable permutations to achieve the lowest error among model types (Figure 4).  Other methods (RMA, GNN, and GAMS) tended towards less complex variable permutations and had slightly higher error rates.  In Arizona:  All model types and permutations fared poorly with respect to preservation of variance (Figure 5).  Nearly all model types and permutations were biased towards under prediction (Figure 6).  In Minnesota:  All model types and permutations deflated the prediction variance with the exception of RMA (Figure 5).  Virtually no differences between model types and permutations with respect to bias (Figure 6). RMA GNN RF AZ – Pre-FitAZ – Post-Fit MN – Pre-FitMN – Post-Fit Table 5. “Best” variable permutation for each model type and sample scene. Figure 4. RMSE by sample scene, model type, and permutation. Figure 5. Variance ratio by sample scene, model type, and permutation. Figure 6. Bias by sample scene, model type, and permutation. Figure 7. Observed biomass change (from remeasured FIA inventories) vs. biomass change predictions pre- and post-fit by LanDTrenDR curve-fitting. Figure 1. Sample biomass trajectories demonstrating the effect of curve-fitting (solid lines) on raw biomass predictions (dashed lines) for two recently disturbed FIA plots (plot-level biomass observations shown as stars). Figure 8. Effect of LanDTrenDR on biomass change RMSE for AZ (left) and MN (right) model types. AZMN  For all model types, the error in predicted biomass change was greatly reduced by curve-fitting with LanDTrenDR (Figure 7).  GNN and RMA exhibited the greatest improvements in Arizona and Minnesota respectively (Figure 8).  RF exhibited the least improvement, but was the most accurate model for predicting biomass change.  All model types were similarly affected by curve-fitting, with an overall reduction in the range of biomass predictions. Funded provided by NASA’s Carbon Cycle Science Program References Canty, M.J., A.A. Nielson, and M. Schmidt Automatic radiometric normalization of multitemporal satellite imagery. Remote Sensing of Environment, 91(3-4): Goward, S.N., J.G. Masek, W. Cohen, G. Moisen, G.J. Collatz, S. Healey, R.A. Houghton, C. Huang, R. Kennedy, B. Law, S. Powell, D. Turner, and M.A. Wulder Forest disturbance and North American carbon flux. EOS, 89(11): Healey, S.P., W.B. Cohen, Y. Zhiqiang, and O. Krankina Comparison of Tasseled Cap-based Landsat data structures for forest disturbance detection. Remote Sensing of Environment, 97: Kennedy, R.E., W.B. Cohen, and T.A. Schroeder Trajectory-based change detection for automated characterization of forest disturbance dynamics. Remote Sensing of Environment, 110: Kennedy, R.E., W.B. Cohen, Y. Zhiqiang, M. Fiorella, E. Pfaff, and M. Melinda. In Prep. Characterizing trends in disturbance and recovery in forests using the Landsat archive. Masek, J.G., E.F. Vermote, N. Saleous, R. Wolfe, F.G. Hall, F. Huemmrich, F. Gao, J. Kutler, and T.K. Lim Landsat surface reflectance data set for North America, Geoscience and Remote Sensing Letter, 3: Pierce, K.B., T. Lookingbill, and D.L. Urban A simple method for estimating potential relative radiation (PRR) for landscape-scale vegetation analysis. Landscape Ecology, 20: Thornton, P.E., S.W. Running, and M.A. White Generating surfaces of daily meteorological variables over large regions of complex terrain. Journal of Hydrology, 190: