15arspc Submission 84

8/8/2019 15arspc Submission 84

1/17

1

SAMPLING CHANNEL FORM AND RIPARIAN VEGETATIONACROSS THE MURRAY-DARLING BASIN USING FULL

WAVEFORM LIDARDavid Moore

Terranean Mapping Technologies Pty. Ltd.PO Box 729, Fortitude Valley, QLD, 4006

Phone number (07) 3257 1011, Fax number (07) 3257 1037Email address [email protected]

Alys WallMurray-Darling Basin Authority

GPO Box 1801, Canberra, ACT, 2601


Thomas WilsonTerranean Mapping Technologies Pty. Ltd.PO Box 729, Fortitude Valley, QLD, 4006


Abstract

The Murray-Darling Basin has immense social, economic and environmentalsignificance. Thirteen of the Basins 23 valleys were categorised by the firstSustainable Rivers Audit (SRA) as being in very poor health and a further seven are in poor health (Davies et al 2008). The Murray-Darling BasinAuthority (MDBA) is responsible for collecting information on the health of theBasins rivers to assist the development of the Basin Plan.

In this project, running from December 2009 through September 2010, remotesensing methods were developed and applied to measure two of the fivethemes that form the basi s of the MDBAs Sustainable Rivers Audit (SRA).These themes are: 1) Physical Form (of the river channels), and 2) Vegetation

(the distribution of riparian foliage in three dimensions). The methodologydescribed here is based on dense (> 4 points per square metre) full waveformLiDAR and high resolution multispectral airborne imagery. A key requisite of theproject was the application of statistically rigorous sampling and the use ofgeospatial analyses to obtain objective and repeatable measurements.

Random stratified sampling was used to select 70 river sites for each of the 23major valleys that make up the Basin. The sites are stratified against fourelevation zones and vegetation types. A total of 1610 sites, each 2000 metresby 700 metres in size, were selected and surveyed with full waveform LiDARand multi-spectral Vexcel imagery. Rigorous field survey conducted across theBasin was used to verify the spatial accuracy of the data. Automated workflowsdeveloped in the TNTmips Spatial Modelling Langua ge were used to


2/17

2

generate a consistent set of primary datasets from classified LiDAR and VexcelImagery. Wetted areas within the river channels were mapped by interpretationof the LiDAR and Vexcel imagery. A semi-automated tool was developed tomeasure a range of river channel and foliage attributes by objective and

repeatable means. More than 50 measurements are generated for each of the1610 one kilometre river sections.


3/17

3

IntroductionMonitoring the health of the rivers of the Murray-Darling Basin poses a range ofchallenges, including the Australian capacity of expert resources required for

collecting information (Davies et al 2010). To address this challenge, the SRAadopted an innovative methodology for monitoring Physical Form andVegetation, based on full waveform LiDAR supplemented by multi-spectralairborne imagery (Johansen et al 2009). The raw data were collected within astatistically rigorous sampling design and the final data were derived throughthe implementation of objective and repeatable processes.

In summary, the project involved four stages, each generating progressivelymore refined information. These stages are:

1. Data collection; aerial and field survey

2. Primary processing of the LiDAR and imagery

3. Secondary processing to extract channel features (centreline, top banks,bottom banks, transect profiles and riparian buffer zones).

4. Data measurements of the physical dimensions of channel profiles andcentrelines, also vegetation measurements based on the verticaldistribution of the LiDAR point cloud.

Highly automated, innovative processes and algorithms were developed toprovide efficiencies and produce data matrices for statistical analysis byobjective and repeatable means.

Data CollectionA stratified sampling design was employed, generating 70 sites for each of the23 valleys in the Basin. Within each valley the sites were stratified againstelevation zone and broad vegetation type. A total of 1610 sampling sites wereselected. An additional 109 check sites were also selected to measure thespatial accuracies achieved in different types of vegetation and topography.

Each sampling site was defined as a 2000 metre by 700 metre rectanglealigned to the primary river channel. Within the rectangular site, a one kilometrestretch of river channel and adjacent riparian zone were analysed.

Figure 1. Each 2000 m x 700 m site is covered by two 577 m wide LiDAR swathes with35% overlap. A 1km section of river within each site is analysed.

2km x 700m Site Rectangle

1km river section

LiDAR SwathLiDAR Swaths


4/17

4

Figure 2. Channel sites (red) and field check sites (yellow triangles) overlaid on the 23major river valleys that make up the Murray-Darling Basin.

The LiDAR was initially recorded using a Trimble Harrier 56 LiDAR instrument.A Harrier 68 LiDAR system was later deployed to accelerate the aerial surveyafter initial delays due to flooding. Both systems were operated with thefollowing parameters:

Flying Height 500 m above ground.Flying Speed: 205 km/hrScanning Angle 60 degreesOverlap: 35%Swath Width: 577 metresScan Rate: 76 HzPulse Rate: 200 kHzPoint Spacing:

o Along Track 0.5 mo Across Track 0.5 m

Ratio: 1 : 1 (along track : across track)Capture Point Density: 4.05 per square metre within swathAverage Point Density: 6.89 per square metreSpot Footprint: 0.25 m


5/17

5

The 0.15 metres accuracy that can be achieved using PPP post processing(and CORS where available) is well within the specified absolute accuracy(relative to ausgeoid98) of 0.5 metres. For each sortie LiDAR was recordedover a boot control site containing at least four horizontal control points on

vertical structures, such as building eaves, and six vertical control points onhard bare flat surfaces, such as roads. Surveyed points, measured to anaccuracy of better than 10 centimetres, were used as a gross error check duringthe data processing phase. Independent field data will be statistically analysedsubsequently to determine the means and standard deviations of the LiDARspatial errors in different vegetation types.

The aerial survey also included the capture of Vexcel multi-spectral imagerythat was ortho-rectified against LiDAR terrain surfaces. It was delivered as4-band multi-spectral GeoTIFF images, with false colour infra-red and naturalcolour enhancements in JP2000 format.

The aerial surveys were constrained to ensure that data was not captured whenwater was overflowing the river channels and the LiDAR and Vexcel imagerywere captured within two weeks of each other. These constraints and the needto survey field check points ahead of the aerial survey introduced somelogistical complexities that were exacerbated by extensive flooding thatoccurred early in the project (December 2009 through March 2010).Primary Processing

The raw instrument data are converted from the temporal / angular domain tospatial coordinates in three dimensions by reference to differential GPS and

Inertial Motion Unit (IMU). This process, performed using the Riegl programRiANALYZE, also converts the waveform signal to points based on a numberof parameters, including a signal amplitude threshold which determines thesignal noise ratio of the point cloud and the proportion of the returned signalcontained in all the points generated from each pulse.

After performing a gross error check for each sortie against the boot controlpoints, the two overlapping strips for each site are levelled, combined andtrimmed to the 2000 m x 700 m site polygon.

TerraScan software was used to automatically classify the LIDAR to apreliminary ground/non-ground classification that was refined through manual

editing and quality checking. The final classification includes the classes: 1)Ground / Water, 2) Vegetation, 3) Structures, and 4) Unclassified.

An automated batch process to produce the primary datasets was developed inthe object oriented TNTmips Spatial Modelling Language (SML). TNTmips iswell suited to this task due to its ability to integrate LiDAR LAS files with GISdata (ESRI shape & Grid) and image formats (GeoTIFF, JP2000). For example,a single tool developed in TNTmips automatically sorts batches of LAS filesassociating each file spatially with a site and assigns a name based on theValley, Site and capture date. The same tool then generates all the rastersurfaces, foliage density layers, contours and other primary datasets, thenoutputs these to a variety of formats and re-imports them to TNTmips internalformat for verification. The TNTmips job queue manager proved highly effective


6/17

6

in managing and optimising the computationally intensive processing acrossmultiple networked computers to produce the large number of output files for all1610 sites. The single batch process produced the following primary datasetsfrom the classified LAS files:

Table 1. The primary data sets.

No. Name Description format 1a Raw LiDAR One file per swath, named according to

Valley, Site ID, capture date, sortie, andUTM zone

LAS

1b Classified LiDAR One file per site, trimmed to sitepolygon, named according to Valley,site, zone.

LAS

2 Ground Points Ground points without built structures ascii text 3 Ground + Building Points Ground points with built structures ascii text 4 Vegetation Points Vegetation Points ascii text 5 DEM Raster terrain surface without built

structures 1 metre grid spacing Arc Grid

5 DTM Raster terrain surface with builtstructures 1 metre grid spacing

Arc Grid

7 CEM Vegetation height above ground 1metre grid spacing

Arc Grid

8 PLR Percent LiDAR returns by strata. Thepercentage of LiDAR returns within 17height ranges above the ground:

0.0 0.1 m 0.1 0.5 m 0.5 1.0 m 1 2 m 2 5 m 5 10 m 5 12 m 10 20 m 20 35 m > 2 m > 3 m > 5 m

>10 m > 12 m > 20 m > 35 m

Arc Grid

9 Contour 0.25 m interval contour ESRIShape

10 AOI Site rectangle polygon extracted fromGIS layer of all site rectangles

ESRIShape

The TNTmips SML script also generated ANZLIC compliant metadata, obtaininginformation, such as capture date and extent of coverage, from the input data

and other GIS layers.


7/17

7

An example of the file naming and directory structure can be seen below. Thisrefers to the classified LAS file (product 1b) in Valley 16, Campaspe, Site 7446.The data is in projection MGA94 zone 55.

\MDB\16-CMP\site_74446\1b_Classified_LAS\CMP_74446_z55.las

The ability to automate the generation of the primary datasets, producemetadata and perform internal checks within a single process contributedsignificantly to the efficiency of the project. The only labour intensive processeswere the levelling of the LiDAR strips, and the manual classification andchecking of the LIDAR point cloud.

The Vexcel imagery was captured, ortho-rectified, mosaicked, enhanced andwritten to standard files by separate processes implemented by Aerometrex PtyLtd.Secondary Processing

The primary datasets were used as the inputs to secondary processes thatextracted river channel features. Two steps were involved: 1) mapping of watersurfaces in the river channel, and 2) extraction of channel features and sitezones.

Water surfaces within the river channels were mapped by manual interpretationof the Vexcel imagery and LiDAR point cloud, using TNTmips geospatial editor.In most sites exposed water is clearly visible in the false colour Vexcel imagery.Where the water surfaces are obscured by overhanging vegetation, the LiDARground points can be used to map the edges of water surfaces. This is becausethe infra-red laser beam is reflected strongly by water. Where the water surface

is smooth the laser is reflected away from the LiDAR instrument and no signal isreturned from ground level. In turbulent water the laser beam may be scattered,returning a signal to the LiDAR instrument; however, even in this situation, acontrast in the density of ground points usually allows water to be distinguishedfrom ground.

A Variable Extraction Toolkit was developed in the TNTmips object orientedSpatial Modelling Language to map channel and riparian features (secondarydatasets) and also measure channel and riparian vegetation attributes(measurement datasets). The Variable Extraction Toolkit is integrated within asingle a graphical user interface, and contains a number of interactive andautomated feature extraction tools organised into a simple workflow.


8/17

8

Figure 3. The Variable Extraction Toolkit interface.

The first step of the feature extraction process is to map the channel centreline.

The approximate path of the river channel is digitised quickly over the reliefshaded DEM. The variable extraction tool then maps the channel centreline asa line of best fit through the lowest part of the channel. Where water bodieshave been mapped in the river channel, the channel centreline passesequidistantly though the centre of each water polygon. The channel centre lineis automatically trimmed to a length of one kilometre.

Figure 4. Mark a general channel route, then auto-generate a 1km channel centre line.


9/17

9

Flow direction is determined visually by inspection of the longitudinal profile ofthe channel centreline and if necessary by reference to a basin-wide drainagenetwork. The channel must be oriented correctly in order to assign the left andright banks relative to the flow direction.

Figure 5. Set flow direction.

Nineteen transects are semi-randomly generated at right angles to the channelcentreline.

Figure 6. Generate transects and top-bank bottom-bank points.


10/17

10

On each transect, points representing the tentative location of the left and righttop bank and left and right bottom bank are generated. The top bank points areinitially located at the first Riley Bench Index maxima. The Riley Bench Index iscalculated as the change in channel width over the change in channel slope.

Maxima in the Riley Index occur on the inside edge of horizontal surfaces(Pickup, 1976).

Figure 7. Riley Index plotted with bank profile.

The aim is to find the top and bottom of the active channel, which must bedistinguished from minor channels within the active channel and wider incisedchannels. It was necessary to provide sufficient flexibility to enable the operatorto guide the program to choose the correct channel, while minimising theamount of subjectivity that could be introduced. This was achieved by: 1)choosing between Riley Bench Index and Line of Best Fit (of the bank heightabove the channel centreline) as the criteria for selecting the bank points on theprofile, 2) allowing the user to specify a minimum bank height for all thetransects on a site, 3) by rejecting transects or sites that are unsuitable for theanalysis.


11/17

11

Figure 8. The decision process for mapping active channel banks.

The operators can view the channel in 3D using anaglyph glasses, switchbetween two different visualisations of the DEM surface and the false colourand natural colour Vexcel images of the site. Thus information, such as thepresence of exposed sediments and vegetation, can be taken into account. Asthe operator discards transects as appropriate (e.g. at confluences) andswitches bank points between the Riley Index and Line of Best Fit, the Line ofBest Fit generally converges with the Riley Index as viewed in longitudinalprofile.

Figure 9. Longitudinal profile of a channel site, showing the Riley Index (yellow), The

Line of Best Fit (green) and the channel centreline (red).

Is the systemconfined or flat

Is the system incised?

Yes

Has an inset channel, with depositional

features, formed within the incised trench?

Assign Top Bank on theincised trench

Assign Top Bank oninset channel

No Yes No

Is the active channel well

defined? Do consistent depositional

features, riparian vegetation or catchment location allow top of

bank to be assigned

Omit from metrics analysis

Assign top of bank

No

Yes

No

Omit from metrics analysis

Yes

Yes

No


12/17

12

After the bottom bank and top bank points have been generated for each of the19 transects, top and bottom bank lines are interpolated between the pointsusing a path following algorithm that seeks to optimise the bank lines byapplying, in order of priority, the following criteria:

1. Minimise change of slope along the bank line

2. Minimise the length of the bank line

3. Follow the zone of maximum terrain surface inflexion.

Figure 10. Interpolate top and bottom bank lines.

The accuracy of the bank lines is less critical, for the purposes of this project,than the accurate placement of bank points on transects. This is because thephysical form measurements apply directly to the transects, while the bank linesare used only for segmenting the site into channel, banks and floodplains for thepurpose of measuring vegetation structure in these Site Zones. Therefore nointeractive control was provided for the mapping of bank lines.

After generating the bank lines, the site is segmented into the Site Zones listedin Table 2 and illustrated in Figure 11.


13/17

13

Table 2. Site zone polygons for vegetation measurements

Zone Description Left bed Between channel centreline and left bottom bank

Right bed Between channel centreline and right bottom bank Left bank Between left bottom bank and left top bank Right bank Between right bottom bank and right top bank 25LB 25 metre buffer from left top bank 25RB 25 metre buffer from right top bank

50LB Between 50 metre buffer from left top bank and 25 metre bufferfrom left top bank

50RB Between 50 metre buffer from right top bank and 25 metre bufferfrom right top bank

50LP From 50LP to site boundary 50RP From 50RP to site boundary

Figure 11. Site polygons segmenting channel and adjacent riparian zone.

The secondary data sets produced by these methods are listed in table 3. Eachdata set includes ANZLIC compliant metadata.


14/17

14

Table 3. Site zone polygons for vegetation measurements

No. Name Description Format

1 ChannelCL Computed channel centreline -lowestpath or mid path through water.

Shape + TNTmips

2 Transects Maximum 19 transects for a site Shape + TNTmips

3 BankPts Left and right top and bottom points forall transects on a site

Shape + TNTmips

4 BankLines Interpolated left and right, top andbottom bank lines for a site.

Shape + TNTmips

5 BankPolygons Polygons generated from bank linesand perpendicular end line.

Shape + TNTmips

6 SiteZones Site zone polygons described in table 2. Shape + TNTmips

Data Measurement

The final data generated for the project consist of two data matrices: 1) aVegetation measurement matrix, and 2) a Physical Form measurement matrix.These matrices are to be used as inputs for subsequent statistical analyses.

The spatial units for the vegetation measurements were produced byintersecting the Site Zone polygons with vegetation polygons from the bestavailable healthy vegetation cover mapping, to mimic a pre-1750 vegetationmap that had been compiled for the MDBA for this purpose. For each resultingpolygon, representing the intersection of a site zone polygon with a vegetation

polygon, the mean of the Percentage LiDAR Returns for each of the 17 stratalisted in Table 1 are calculated as polygon attributes. These data represent thevertical distribution and density of foliage by vegetation type across the zonesassociated with each river channel.

Intersecting the vegetation polygons with the Site Zone polygons, calculatingthe mean PLR for each stratum as polygon attributes and writing the results to adata matrix in CSV format is performed as a single process within the VariableExtraction Tool.

Similarly, a range of Physical Form measurements is generated from thetransect profiles, bank points and channel centreline. These attributes, listed inTable 4, are calculated as attributes of the GIS line features and are exported toa data matrix for each site as an automated process.


15/17

15

Table 4. Physical Form Measurements

Input Variable Description ChnlCent Channel centreline; length in metres

ValCent Valley centre; length in metresMaxElev Maximum elevation of channel centreline MinElev Minimum elevation of channel centreline ElevDiff Elevation Range of the channel centrelineTranMin Minimum elevation of transect TBnkAng(L&R) Left and right bank top angles TBnkHt(L&R) Left and right bank top height BBnkHt(L&R) Left and right bank bottom height TBnkArea Cross sectional area below top bank level AngLn(L&R) Length of profile from bottom bank to top bank left and right

SAngLnR Length of profile from spill height to bottom bank Convex(L&R) Cross sectional area of the bank profile above the bank angle line. Concave(L&R) Cross sectional area of the bank profile below bank angle line. Inflect(L&R) Number of times bank lines crosses bank angle line Length(L&R) Length of bank lines for the site left and right banks WetBed Does transect intersect water? Yes or No WetWdth Width of water at transect from wetted area layer BedWdth Distance between left and right bottom banks ChnlWdth Distance between left and right top banks ChnlDpth Channel Depth; height of spill level above bottom bank

AreaChan Channel area between left and right top banks AreaBnk(L&R) Area between top and bottom banks for the site left and rightStrmPow Stream Power; (max elev minus min elev) / valley centreline length ManMade Evidence of manmade channel or features, or water control


16/17

16

DiscussionFull waveform LiDAR was found to be a very effective source of data for takingdetailed and precise measurements of the river channels and riparian

vegetation. The minimum density of four outgoing laser pulses per square metreproduced a ground point density of more than one point per square metre, evenin complex vegetation, generating extremely accurate and detailed groundsurface models. The high pulse density, full waveform LiDAR, returned up to 20points per square metre in areas of complex vegetation enabling the distributionof foliage to be mapped in three dimensions with a high level of detail.

The TNTmips software proved to be an extremely effective geospatial modellingenvironment for establishing automated workflows to produce consistent andrepeatable GIS output data from the classified LiDAR LAS files. The TNTmipssystem includes a Job Queue Manager which greatly facilitated management of

the data processing by optimising the use of networked multiple corecomputers. It also provided detailed information on the status of the varioustasks that had to be performed in sequence for more than 1610 individual sites.

TNTmips also proved to be a highly effective geospatial modelling environmentfor the development of the Variable Extraction Tool, due to its ability to integratea wide range of data types and formats, including LiDAR LAS files, and itsadvanced geospatial analysis functions.

The data from this project are intended for statistical analysis. Consequentlythere was a strong emphasis on the development of methodologies thatminimise subjectivity and bias by minimising opportunities for user interaction.

Variable extraction was performed entirely by GIS operators with no training orexperience in geomorphology. Throughout development and implementation ofthe variable extraction methodology, expert geomorphologists assessed thequality of the data and provided advice on opportunities for improvement of thealgorithms and work flows. Protocols were established for obtaining advice fromprofessional geomorphologists where decisions could not be made withconfidence by the GIS operators. The primary skill the GIS operators needed toperform the variable extraction was the ability to distinguish alluvial systemsfrom other types of river sections that do not apply to this project. Similarly,confluences and built structures such as bridges and dam walls can maketransects or entire sites unsuitable and the operators were trained to recognise

these in the DEM and flag them for discarding or checking.The methodology and algorithms are designed for alluvial systems and do notoperate effectively on river systems that are flat, topographically constrained orcontain large boulders. These types of river sections are difficult sites for bothgeomorphologists and an automated system to extract measurements from.The project was designed with redundancy for this purpose; of the 70 sitessurveyed in each valley, only 50 are required for the subsequent analysis basedon a biostatistical power analysis.

An independent assessment of the data by professional geomorphologists,currently under way, has reported that on less than 10 percent of sites, banklines were not correctly assigned to the active channel. Between 16 and 35


17/17

17

percent of sites have minor errors in the alignment of bank lines which will notadversely affect the site measurements. As discussed above, the interpolationof bank lines is less critical than the correct assignment of bank points on aprofile. Preliminary results suggest that between 58 and 75 percent of sites

were mapped as would be expected from an experienced geomorphologist. Itwas noted that the manual identification of active channels and delineation ofbank lines is a subjective process and that different geomorphologists would belikely to produce different results. Regardless of whether bank lines mapped bythe methods described here are consistent with the definitions generallyaccepted by geomorphologists, the fact that they are generated though arepeatable and objective process lends this data to statistical analysis.

This approach, and the intensely information rich LiDAR data, will enable thePhysical Form of the Basin river channels and associated vegetation to beassessed with a higher degree of rigour and repeatability than has previously

been possible. Future monitoring and evaluation of the Murray-Darling Basinwould most likely combine remotely sensed data with field calibration andcoincident collection of field data, such as channel pebble size, channel depthwhere water is present, and vegetation recruitment.

ReferencesDavies, P.E., Harris, J.H., Hillman, T.J., and Walker, K.F., 2008, SRA Report1:A Report on the Ecological Health of Rivers in the Murray-Darling Basin, 2004-2007. Prepared by the Independent Sustainable Rivers Audit Group for the Murray-Darling Basin Ministerial Council.

Davies, P.E., Harris, J.H., Hillman, T.J., and Walker, K.F., 2010, TheSustainable Rivers Audit: assessing ecosystem health in the Murray-DarlingBasin, Australia. Marine and Freshwater Research , 61, 764-777.

Johansen, K., Arroyo, L.A., Ashcraft, E., Hewson, M., and Phinn, S., 2009,Assessing then potential for remotely sensed statewide mapping of streamsidezone and physical form metrics in Victoria, Australia, 2 April 2009. Prepared forPaul Wilson, Department of Sustainability and Environment, State of Victoria.

Pickup G., 1976, Alternative Measures of River Channel Shape and theirSignificance. Journal of Hydrology (N.Z.) , 15, 9-16

15arspc Submission 84

Documents

Transcript of 15arspc Submission 84