Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
High Resolution Inventory Services
DNR Workshop Operationalizing LiDAR based Forest Inventory
January 25, 2016 | tesera.com
The most detailed, accurate and reliable forest inventory data and maps available in the marketplace
HRIS Management & Production TeamTESERA
Bruce MacArthurbruce .m acarthur@tesera .com
President and CEO
Ian Mossian .m oss@tese ra .com
Chief Analytics Officer
Dwight Crousedwigh t.crouse @tese ra .com
Senior Data Analyst
Alex Josephale x.jose ph@te sera .com
Director, Susta inable Solu tions
Dwight Scott Wolfedwigh t.wolfe@te sera .com
Chief Com pliance Officer
Shannon Pattersonshannon .patte rson@te sera .com
Director, User Experienceand Com m unica tions
Inventory … Tool Chain? … What Attributes? … Serving Whom? … With What Kinds of Outputs? … What About Species? … How Do They Use This Stuff? … Highest Priority Improvements?
DNR Workshop AgendaTESERA
17staff
2Post Doctorate (PhD)Analytics, Statistics,
Climate Risk
Experienced entrepreneurs, integrators designers and collaborators
7Software, GIS, Data, Web Design Leaders
5+Project
Management Support
Overview of the Team & the ExperienceTESERA
500+Projects 5
Experienced foresters
HRIS OverviewTESERA
January-19-16
Tesera HRIS Data Compilation
ColourInfrared
Photography
ALS/LidarData*
CIR/LidarData
Fusion
MicrostandGridcell
Delineation
LiDAR/CIRIndice
Compilation
Cutblock / Linear Feature/ Disturbance
GroundPlotData
PhotoPlotData
ReferenceDatasets
Forest LayerLand Cover Class
TargetDataset
GridcellsMicrostands
Young StandCutblockInventoryAttributes
GroundPlot
Data CompilerGrowth Projections
GenerateTerrainIndices*
GenerateClimateWNA
Indices
Roads / PipelinesHydrographySeismic Lines
SampleDesign
January-19-16
Tesera HRIS Modeling & Production
ReferenceDatasets
Forest LayerLand Cover Class
TargetDataset
Gridcell Inventory
Young StandCutblockInventoryAttributes
CustomizedIndicator Variable
Selection **Discrimant Analysis
R Subselect Improve
Tree Cover I
Species CompositionLorey’s Mean Tree Height
Dominant Tree HeightSite Index
AgeCrown Competition Factor
Crown ClosureBasal Area
Trees Per HectareVolume
...Stand Structure Class
Cumulative Distribution Index
Tree Cover II
Tree ListsStand and Stock Tables
Regresssion**R glmnet
kNNCustomized
SciPy KDTree
Fuzzy C-MeansClassification
Model TestingImprovement
Take-One-Leave-Onek-Fold Validation
RMSPESRB
KHAT
Land Cover**
Vegetated-Treed
Vegetated-Non TreedShrubs Open
Shrubs ClosedHerbs
Grasses
Non VegetatedRockWater
River Bank
Blowdown
Regresssion**R glmnet
QCQuantilesMoments
CodesViewing
Microstand Inventory
Final HRIS
Aggregate SimilarMicrostands intoLarger Polygons
HRIS Tool Chain – Ground Plot Data /Inven tory Attribu te s (Python)TESERA
• Plot compiler o Stems Per Hectare o Basal Area o QMD o Lorey’s Mean Tree Height o Merch Volume o Gross Volume o Region o Crown Area o Cumulative Distribution Index (Moss) o GINI Coefficient (Lorenz Curve Area Difference) o Stand Variance Index (STVI; Staudhammer and LeMay 2005) o Species Composition (Basal Area; Crown Area)
• Stand & Stock Tables o TabSpDclLd o TabSpDcl o TabSpLd o TabDclLd o TabSp o TabLd o TabAll
• Tree Compiler (Python) o Species
Update codes o Dbh o [Status (Live or Dead)] o Height o Stems Per Hectare o Basal Area o Merch Volume o Gross Volume o Crown Area o Log number, volume, length, large & small end diameter
• Stand Structure Classification o Version 1 custom algorithm o Version 2 fuzzy c-Means (Bezdek 1981; Python) o Pre- and post classification classifiers
7
HRIS Photo Plot Data /Land Cove r Classifica tionTESERA
Grid Cell Land Cover Classification
• Vegetated o Forested (≥ 6% CC) o Non Forested
Shrub Open (≤ 50% cover) Shrub Closed Herb-Forb Grassland
• Non Vegetated
o Rock Bare Rock Sand & Gravel
o Water Flooding – Beaver pond Flooding - Other Plus River Bank
• Severe Windthrow
TESERA
HRIS Tool Chain
● Organize Data :- Postgre s- Quality Control Tools (Python)
● Users Update Config File - In te ractive in in te rpre te r
● Data Dictionary- Variab le Types- Crea ted Autom atica lly- Manual Updating- Option To Change Variab le Nam es
drive:E interActiveConfig:YES backupOldConfig:YES backupConfigFileName: backupFilePath:/Rwd/Python/Config/Backup/ dataInPath:/Rwd/ dataOutPath:/Rwd/ dataInFileName:IDFREGTREE.csv dataOutFileName:IDFREGPLOT.csv errorDiagnosticPath:/Rwd/Python/PyReadError/ loreysHeightVarName:HL cdiVarName:CDI giniVarName:GINI stviVarName:STVI qmdVarName:QMD spPropBphExt:_B spPropCareaExt:_CA spRankPrefix:SP spRankPctPrefix:PCT defaultRankSpName:NA maximumTreeDbh:140 dbhInterval:1 ccVarName:CC printDataInVarTypes:YES defaultFileExtension:.csv useForestTreeConfig:YES dataInUniquePlotIdVarName:MPLOTID dataInUniqueTreeIdVarName:TREEID treeSpeciesVarName:SPECIES dbhVarName:DBH sphVarName:TPH basalAreaPerHectareVarName:BPH regionVarName:REGION heightVarName:HT crownAreaVarName:CAREA merchVolumeVarName:MVPH grossVolumeVarName:GVPH
9
TESERA
HRIS Tool Chain – Basic Inputs (Indices)
● Lidar – LAS Tools● LiDAR /CIR Products
- (Blom ASA; Petteri Packalen)- CIR/LiDAR data fusion (C)- Microstands (eCognition ?)- Grid cells
● ClimateWNA (aka PRISM in OR)● Terrain Indices (ktpi)
- R Raster Package*● Stage (2007) Terrain Indices*
*AWS Cloud Processing*pyRserve*Multiple instances
Microstands &Gridcells
Terrain Indices
10
TESERA
HRIS Tool Chain – Reference Data Analysis
● Select XY Variables – Standard csv file for user interaction● Classification – Fuzzy C-Means (Bezdek 1981): Python 2.7● Coarse Variable Selection - Discriminant Analysis: R Subselect● Coarse Variable Importance Assessment : Python● Handling Autocorrelation: R● Linear, Binomial Log Odds, Multinomial Log Odds: R glmnet, Pandas, sciKitLearn● kNN; Multiple Discriminant Analysis: R MASS● Evaluation: Take-One-Leave-One; k-fold validation – R packages● Evaluation: RMSPE, Bias, SRB, KHAT – Python; U -Error - R MASS● QC : Quantile Checker, Code Lists – Python ● QC: Orthogonal Regression: OrthogonalDistanceRegression (Python function)
11
TESERA
HRIS Tool Chain – Target Data Processing
● Large datasets – All Python● kNN assignments (Tree Lists + BA adj Tree Lists): KDTREE● Linear equation application● Species transformations: Unpack, Repack● Age as function of height & site index: iterative routine● Quantile + code range checker: referance vs. target● Discrete class generator● Compile unique combinations● Data dictionary compiler● Data transformation manager● Grid cell to microstand summary routine● PrognosisBC batch file production and summary routines● Stand structure classification
12
TESERA
Who
● Spray Lake Sawmills, SW Alberta, 330,000 ha – 2 Parcels, 2008 to present
● UBC Alex Fraser Research Forest – Knife Creek, ~ 3500 ha, complex stands, variable radius plots, outliers - 2014, 2015; Negotiations to extend to 1 million Ha in IDF
● WIRE Services (Manitoba Hydro), Costa Rica, biomass & carbon estimation – 2014
● Island Timber (in negotiation) – Vancouver Island – 250,000 ha – site productivity, unstable slopes, standard inventory
● Sechelt Community Forest (in negotiation) – coastal mainland – 10,000 ha standard inventory attributes
● BCMoF: Landscape Vegetation Inventory (LVI; Landsat + Photo Plots)
13
TESERA
Highlights
● Species recognition● Deriving unbiased compatible estimates for height, site index, and stand age● Terrain and Stage -Terrain indices● Advancing the system in a cloud and web -enabled environment● Complex stands: Linking the inventory to the growth projection system and using a
forest estate model for harvest scheduling and estimations of sustainable timber supplies (new for BC)
● Identification of outliers where additional samples are needed● Establishment of a simple guideline for use of variable radius plots● Vast improvements in quality control procedures● Extensive documentation to support existing products ● Routines can mostly be used by people familiar with computers but not expert
programmers and with minimal training in use of statistics ● Thorough review by clients and third parties (government agencies)
14
TESERA
Challenges & Opportunities for Improvement
● Standardized datasets and methods for reporting on reliability of inventory (best practices)
● Deploying additional analyses pathways as part of the process● Extending the process for use with large datasets ● Height, age, site index compatability and removing bias● Species proportions (vs. Photo Interpretation)● Species proportions with respect to change in diameter● Corresponding LiDAR + CIR metrics● Parametric methods as alternative to kNN (Holy Grail)● Dominant tree & Lorey’s mean tree height ● Crown closure (Gill et al. 2007; Betchold 2004)● Height -to -live crown● Post (grid cell) processing stand delineation● Tree lists: Integrated Inventory, Treatment Unit, Silv presc, GY Foresting, Harvest
Scheduling, Inv. Reconciliiation and Update Process● Operational Applications (Post Production Stand/Treatment Unit Delination)
16
TESERA
Tree List Discussion Topics (Emphasis on Complex Stands)• Why?
o Species Habitat Assessment o Fire (& other Disturbance)
Hazard/Risk Management o Course Woody Debris (Recruitment) o Log Supply Forecasting o Partial Cutting o Simulating Natural Disturbance o Watershed Dynamic Characteristics o Growth and Yield Forecasting
• What
o Species x Diameter x Status (L/D) o Additional attributes
… o Labeling
Stand structure classification o Site productivity (complex stands)
• How o kNN
Sample design, intensity, and distribution
o Parametric techniques Complex stands?
• Managing
o Silviculture prescriptions o Growth and yield forecasting o Growth and yield monitoring o Inventory update o Forest Estate Analysis
Prescriptions Timing of application (harvest delays) Growth and response curves
o Integration with forest operations Delineating polygons consistent
with prescription guidelines. Locating treatment units in the field Enabling adjustments to the inventory
tree lists / stand and stock tables …
ALS/LiDAR CIR … Other?