s.hillard, r.deo, j.corcoran, d.kepler – mn dnr andrew...
TRANSCRIPT
S.Hillard, R.Deo, J.Corcoran, D.Kepler – MN DNRAndrew Brenner & Miles Strain – QSI
• Objective of study• Single Photon
LiDAR (SPL)• Project Area and
Collect Specifications
• MN-DNR LCCMR Forest Inventory Project
• Where SPL plays in the market
Presentation Outline
The Scale of the State’s land portfolio
• The MN-DNR administers 5.4 million acres, 3.7 is forested, and 2.7 is considered in our commercial management pool.
• We classify 47 cover types, 26 are forested cover types, of those, 19 are considered commercially viable, 12 are deciduous, and 7 are conifers.
MN DNR Lands
MN Counties >10% forested
Costly, yet vital – current costs are between $10 & $12 per acre.Develop a New Methodology for Forest Inventory
Stand based data – photo interpreted with multiple plots per stand.Results in an older inventory of 20+ years.Need to address multiple other natural resource values/benefits
Objectives
Three year Pilot Project
July 2017 – June 2019
Acquire high density Single
Photon LiDAR (SPL) Acquired during peak fall color 2017 728,000 acres (~1,135 square miles) >12 pulses per square meter Simultaneous, 4 band 30cm
photography Additional AOI flown with help from
the Superior National forest.
Cass - 628K acres
Lake - 100K acres
Development of Innovative Cost-Saving Methodology for Forest Inventory
Cass County, 432 gridded plots Lake County, 108 Stratified Random plots
1/10 acre fixed radius plots, full stem map, measured >5 in DBH:
Study Area and Field Sampling Design
• Lower-power laser and highly sensitive detection to achieve accurate range measurement
• 100 optical channels respond to each laser shot, achieving 6.0 MHz effective pulse rate
• Originally developed for earth-to-satellite rangefinding
• Became a technology accessible to Leica Geosystems through the acquisition of Sigma Space in 2016
• Lowers cost per acquired data point
Single Photon LiDAR
• Large-area mapping driven by cost per data point
• Dictates high effective pulse rates
• Ability to fly high enough to cover wide swaths, while maintaining high pulse rates
• Current technology has allowed a doubling of pulse rate roughly every two years
• SPL100 breaks pulse rate barriers a 6.0 MHz effective pulse rate
Single Photon LiDAR
Pulse Frequency
Points per second
Number of returns
60 kHz 6 million –100 optical channels
Up to 3 returns per laser shot
• Flew with a Leica RCD 30 – medium format camera– Mn DNR’s program based on peak color
imagery – deciduous species identification– 4 band, 6” gsd
• Flying imagery and LiDAR together– One mobilization– Spectral and structural data together– Need imagery for hydro flattening – Imagery limits LiDAR collection
opportunities– Large format cameras better over large
areas – small weather windows
Imagery System
• Things that went well– High point density– Good ground penetration
• Where there were challenges– Clouds limited acquisition– Hydro flattening a problem– Large amount of data– Use of RCD 30 created processing
challenges – many small frames• Control
– In rural areas – expensive – remote areas– High cost to collect NVA and VVA
accuracy assessment points required for 3DEP compliance
Operational Items
• Grid metrics were produced using FUSION software from the SPL point cloud.
• Grid metrics were used as predictor variables in multiple regression models for estimating a number of response variables measured within our 1/10 acre plots.
Working Towards the inventory
• 356 plots were located within the forested part of our AOI, and 333 plots were used for model fitting.
• Inventory models were predicted across the study area on 20m grid cells.
• Metrics Include:– Standing Volume – Above Ground Biomass – Basal area weighted height– Basal area– Trees per acre– Site index– Max Height
Working Towards the inventory
Working Towards the inventory
Working Towards the inventory
SW+HW
SW HW
FIA-plot: Single photon LiDAR
FIA field observations All trees <5 inch DBH
FIA-plot: Linear LiDAR
An FIA plot in the overlap area of SPL and linear LIDAR acquisitions in the Lake County, MN
FIA subplot-2 No. of trees(> 5”): 4 Maximum Ht.: 13.7-mMax DBH: 16.2-cm
FIA subplot-2No. of trees(> 5”): 4 Maximum Ht.: 13.7-mMax DBH: 16.2-cm
SPL Linear
Linear LiDAR
FIA observed no. of trees: 13; Max ht.: 14.3-m
Linear LiDAR
Working Towards the inventory
• We can use our current stand boundaries to extract the information by domain (stand).
• We are currently working on how to give associated error estimates for each attribute (more on that later)
Working Towards the inventory
Working Towards the inventory
Attribute SRS-SE % MA-SE % Design Effect
Volume 4.16 2.24 0.26
AGB 4.08 2.15 0.26
Basal Area 3.7 2.4 0.37
TPA 3.4 3.0 0.68
• The forested area of the AOI was estimated at 310,000 acres.
• We made estimates of the different attributes across the study area to compare the model assisted (MA) estimate to the design based estimate (SRS).
• The model assisted estimate had improved precision over the design based estimate, the MA estimate of TPA showed the least improvement over the SRS.
Future work and improvements• Comparison of post stratification vs pre-
stratification sampling
• Investigating sample size impacts on model parameter estimation.
• Small area variance estimates ( synthetic domain estimators)
• Mixed effects models with plot level random effects to improve precision ( hardwood/ softwood, topographic, etc).
• Developing models to predict age and classifying cover-type (to some degree)
Attribute SRS MatchMA Error
MA MatchSRS Error
Volume 1213 91
AGB 1247 89
Basal Area 885 125
TPA 488 227
• University of Minnesota investigated data driven stand delineation using object based segmentation.
• Northland Community Technical College also collaborated with us on collecting UAV data across a number of plots within the study area.
Additional Collaborators
• UMN-Remote Sensing lab conducted object based segmentation to produce “stand” objects.
• Stands were based on a large array of inputs.
• Has been the subject of much debate.
UMN-Object based segmentation(OBS)
Chippewa Superior
Band 1: Time ABand 2: Time BBand 3: Time B UMN
RSGAL
Canopy Height Change: 2012 - 2017
Wildfire
Timber Harvests
Wind Blowdown Wind Blowdown
Timber Harvests Wildfire
Ortho-imagery Multi-date CHM
Ortho-imagery Multi-date CHMOrtho-imagery Multi-date CHM
UMN RSGAL
Canopy Height Change: 2012 - 2017
Wind Blowdown
Timber Harvests
Wind Blowdown
Pike Bay
UMN RSGAL
Canopy Height Change: 2012 - 2017
• Red = can only be satisfied by linear-mode systems
• Blue = can be satisfied single-photon technology
• Purple = can be satisfied by either system, but…… may not be the best use of capital for SPL technology
A way to think about how touse different technologies…
10,000 square miles
Acknowledgements
UMN RSGAL