North Devon Biosphere Foundation

Making hedgerow and woodland wood-fuel resource

assessments using remote sensing datasets.

A project carried out for the North Devon UNESCO World Biosphere Reserve Sustainable Energy Group

Summary This study attempts to explore remote sensing technology and publicly available data for making some

biomass resource assessments. Various data sets have been used. The favoured one was the

Environment Agency Geomatics LiDAR data given for a surface model and terrain model. Other data did

not have the required accuracy.

From these data sets it seems likely that biomass assessments in hedgerows can be made. The figure

derived for the 27.25 km2 study areas was 500 GWh of heat energy being available from almost 360Km

of hedgerow.

The woodland assessment is more problematic with high errors and uncertainty in the data for the

deciduous woodland areas. Conifer areas tend to be managed and the resource is probable better

known. However most of the deciduous areas are not managed and this is where much of the resource

can be exploited sustainably….if the volumes are known.

For the immediate future, on site sampling assessments for woodlands will be the most accurate and

cost effective tool, until more work is done on the LiDAR processing.

Hedgerow assessments can be done using this technology, with more calibration.

Acknowledgements

The North Devon UNESCO Biosphere Reserve is grateful to the support from the EA Geomatics Group

for providing the data free of charge as a trial of a methodology. We are also grateful to DCC data

managers for the use of the GetMapping Data under licence. GIS Data in this report is based on OS data

provided under licence to DCC and supported by a Public Service Mapping Agreement.

© Crown Copyright and database right 2014. Ordnance Survey 100019783.

Brief: The North Devon Biosphere Reserve Sustainable Energy group has carried out a number of surveys and

reports for the energy use and possible solutions for reducing emissions from the Biosphere Reserve

area generally. One of the key conclusions of the report was to reduce the emissions, to improve the

local economy and to reduce fuel poverty was to increase the sustainable use of wood fuel heating. This

needed to be coupled with the improvement of energy efficiency.

Work in recent EU funded Cordiale Programme indicated ways in in which hedgerows, with altered

management regimes might be a useful source of wood fuel and as an income for the landowners.

The objective for this project is to identify ways that resources can be assessed using remote sensing

data that might be available in various forms.

The Biosphere Reserve Sustainable Energy Group have identified a community based project in South

Molton to explore the use of a district heating system to to heat the community owned swimming pool

and other community buildings nearby. This project will explore how resources may be assessed in the

surrounding area building on the experience gained from previous projects.

Wood fuel resource assessment methodologies to date: There are trade-offs with the cost and level of accuracy of the estimates. However the more accurate

the measurement, results in lower uncertainty and therefore a better standing price.

Direct Measurement

Foresters have been using a range of methods to measure standing resources of timber for decades.

These have been both invasive and non-invasive.

The methods might range from measuring every tree; time consuming and very costly though to

stratified sampling measures E.g. Tariff systems involve counting and measuring a sample of trees and a

sub sample of these trees are felled to measure more precisely the volume of trees in various diameter

at breast height (DBH) classes and relating these through statistical tables to provide an overall estimate

of the standing volume of crop.

Generalised stand volumes that are approximated from crown height, basal area (the total surface area

of the stems at DBH per Ha) and a form factor for the tree species.

Recent additions to the tool box for forest mensuration include the smart phone, with applications such

as iHypsometer which effectively replaces the relascope, clinometer and clipboard.

More semi quantitative methods have been developed for unmanaged woodlands for

landowners/managers to use as a ready reckoner to estimate Cordiale

Applying Hedgerow resource measurements

The recent work undertaken through Cordiale Project in the Tamar AONB iand more recent publications

from the Devon Hedgerow Group provide some guidance on yields for hedges under different

management/harvesting prescriptions and stagesii. These are reviewed below.

Remote Sensing

Stereographic aerial photogrammetry was a technique that has been used, that allowed some relative

and absolute heights to be gained, as well as broad species group composition. The Improvement of

photographic emulsions and latterly in to multispectral receivers meant that species can be identified

also.

The recent additions of technology such as LiDAR (Light incidence detection and ranging) can provide a

rapidly acquired data set used from a land based platform to give volumes in the forest, or from an

aerial platform (light plane or satellite) to give canopy heights and through various processing.

Figure 1 LiDAR scanning captures ground and canopy level reflections

Such radar or light based survey methods can yield 2 layers:

• The Digital Surface Model (DSM): this is the surface model generated by all of the highest

readings over a cell area. It therefore includes all of the structures and features above the bare

ground.

• The Digital Elevation Model (DEM); this is the model populated by the minimum heights

acquired in the cell area. This effectively is the bare earth model of the area being investigated.

Figure 2 Difference between DSM (left) and DTM

LiDAR can be expensive to acquire and so it must be considered for multiple uses when it is gathered.

However it is getting cheaper to acquire. Similarly since the data capture excursions are the most

expensive element, other remote sensing should be carried out simultaneously if possible. This might

be high resolution aerial images and other Hyperspectral receivers to maximise the data capture and

expand the types of analyses available; for example Infrared, to show moisture content, plant health,

etc.

As will be shown in this trial, having contemporaneous data capture means that verification between

data points can be more certain. For example comparing LiDAR and aerial photographs taken even one

year apart will be impacted by forest or hedgerow management cycles.

Methods deployed over a landscape area

The Tamar Valley AONB undertook a hedgerow assessment using aerial images in a pioneering

programme to interpret the resource available and published the methodologyiii. The method uses 5

classifications of hedges that were easily identifiable from aerial and on farm surveys. These are

mapped using GIS and linked database to provide the resource assessment. The methodology is

participative, which has the benefits of direct engagement with the landowners, but was very time

consuming (6 months with a specialist)

The Cordiale Project also produced a ready reckoner for non-specialists for hedgerows and small

woodlandsiv. The tools provide 17 classifications of hedges from which the volume of extractable

biomass is estimated. The same classification system is used in the tool developed for

specialist/professionalsv which uses more intensive sampling methods for DBH measurements and

height measurements of sample plots as applied in standard forest mensuration. The specialist tool was

derived from the work undertaken by the Forestry Commission.vi

The most recent iteration of hedgerow assessment provides a classification of the hedges into 6

typologies and gives and estimated energy value of the chips after seasoning. There are variations due

to species composition. It is assumed that the wood to bark ratio in the harvested volume is taken into

consideration in the energy estimates.

This project will test a few publicly available remote sensing data sets and methodologies using off-the-

shelf technologies to analyse the data. The contract time for this project was less than 10 days.

Therefore by necessity, the assessments are going to be rapid.

The Project Area The area chosen for the assessment is in the South Molton area and slightly to the north. It was chosen

because of the proposed renewable heat project for the swimming pool. The area to the north of the

A361 has been chosen to focus the work on because

• it includes a number of woodland sites that are Plantations on Ancient Woodland Sites (PAWS)

and these will need some management that will be supported to re-instate to native woodland

cover

• there are blocks of woodland deemed to be in management and others not deemed to be in

management.

• there appears to be a variety in hedgerow structures across the farms in that area.

Figure 3 Map of the project area

Data sets available:

• FC National Forest Inventory 2013 (publicly available). The polygon data includes an estimate of

forest type (e.g. conifer, broadleaved, Mixed, Scrub, Coppice, open ground).

• Defra Rural land Registry Field Parcels (under licence to the Biosphere Reserve). This data set

can be used to generate field boundary layers that are more likely to be hedges, since the OS

Master Map layer is hard to separate agricultural boundaries from others.

• Bing aerial imagery automatically streamed through Arc GIS

• Get Mapping Digital surface model and Digital elevation model (DCC licensed)

• LiDAR coverage 2m resolution (limited cover) available in JPG and ASCI Grid for different prices.

Both data types are in DSM and DTM. (under licence/purchased from Environment

Agency/Geomatics).

Figure 4 Indication of LiDAR coverage. Source EA Geomatics-Group.co.uk

As can be seen from the screen shot above, LiDAR coverage is not complete. This data has been

identified as 2013 composite layer. This doesn’t necessarily mean that the flight was taken in 2013. It

therefore draws in a level of uncertainty. To professionally acquire the data for analysis for this project

for the ASCI grid would cost £5900 plus VAT. In JPG version the same areas is £1500 plus VAT.

• Get Mapping 2006 25cm Aerial photography (DCC Licenced). This can be used as in the early

stages of the Cordiale Project on hedgerow identification.

Figure 5 Sample images of the 25cm resolution aerial ortho-rectified imagery.

DSM DTM Analysis. The principle approach for early quality testing of the data source is to subtract the DEM from the DSM.

If the quality is adequate for further work, the resulting raster image will reflect all the above ground

structures, with plausible heights.

Get Mapping Data

The GetMapping DSM and DTM Tiles SS80 were provided by DCC. These tiles were to the south of the

study area but were provided for a trial of the methodology and feasibility of using this dataset. The

raster resolution is 5m pixels, which can be adequate for forest purposes for identifying canopy height

but might be dubious for individual tree recognition. The width of hedgerows is 5 m or less therefore

picking-up hedgerows would be uncertain.

The following page shows the DTM (figure 7), DSM (figure 8) and the residual above ground images

Figure 9). As can be seen the residual layer picks out the hedgerows and woodland, however the lines

are not continuous.

Furthermore, there are values in the residual layer that are negative. The range is from -13.03 to +30.3.

It is suspected that this error was from the original data capture where the steep slopes and woodlands

might present false values, especially within the valleys.

Figure 6 Histogram of points of the residual raster of Get Mapping

For reasons of high uncertainly and low resolution, this data set was discarded.

Histogram of Minus_ss80ds1: Field = VALUE

-13.03399658203125 - -3.530230068559487 -1.2559588138460169 - -1.2220144667607413 1.2898672175496584 - 1.3238115646349335 3.8017489018600577 - 3.8356932489453328

170,000

160,000

150,000

140,000

130,000

120,000

110,000

100,000

90,000

80,000

70,000

60,000

50,000

40,000

30,000

20,000

10,000

0

Figure 7 DTM of Get Mapping product

Figure 8 DTM of GetMapping

Figure 9 Residual layer of Get Mapping SS80. Note the discontinuity of the hedgerows.

Figure 10 Google Earth image for same site (the image source is Get Mapping 2010).

LiDAR Data

The LiDAR data available does not give complete coverage of the area. This is due to the reason for

which the flights are originally commissioned. To date it has been largely used for modelling hydrology

and flood zones. The image below shows the extent of the data used in this analysis.

Figure 11 LiDAR Data coverage for the project area

Jpg Format.

The same process as above was carried out on the LiDAR data provided by Geomatics for the study area.

This is the cheaper of the 2 LiDAR products.

Figure 12 Presentation of LiDAR JPG data

Since the layers are Red Green Blue channels that include hill shading, the subtraction process results in

a single band raster, that shows the features, but there is no meaningful extractive data than can be

reverted back into any absolute heights. The JPG files are good for illustrative measures but not good

for any analysis work. This data set was also discarded.

0 0.2 0.4 0.6 0.80.1

Kilometers

0 0.2 0.4 0.6 0.80.1

Kilometers

LiDAR DATA ASCI Grid Format

The same process was applied to the ASCI grid LiDAR data and the results can be seen below.

Figure 13 Histogram of heights in LiDAR residual feature height data. NB very few negative height values

Figure 14 Sample of the LiDAR residual feature height mapping

Based on these results the ASCI data was taken to the next stage of analysis.

Histogram of Residual feature height

Residual Height (m)

-0.98020172119140625 - -0.82999813556671143

-0.37938737869262695 - -0.22918379306793213

0.37163054943084717 - 0.52183413505554199

1.1226484775543213 - 1.2728520631790161

1.8736664056777954 - 2.0238699913024902

2.6246843338012695 - 2.7748879194259644

3.3757022619247437 - 3.5259058475494385

4.1267201900482178 - 4.2769237756729126

4.8777381181716919 - 5.0279417037963867

5.628756046295166 - 5.7789596319198608

6.3797739744186401 - 6.529977560043335

7.1307919025421143 - 7.2809954881668091

7.8818098306655884 - 8.0320134162902832

8.6328277587890625 - 8.7830313444137573

9.3838456869125366 - 9.5340492725372314

10.134863615036011 - 10.285067200660706

10.885881543159485 - 11.03608512878418

11.636899471282959 - 11.787103056907654

12.387917399406433 - 12.538120985031128

13.138935327529907 - 13.289138913154602

13.889953255653381 - 14.040156841278076

14.640971183776855 - 14.79117476940155

15.39198911190033 - 15.542192697525024

16.143007040023804 - 16.293210625648499

16.894024968147278 - 17.044228553771973

17.645042896270752 - 17.795246481895447

18.396060824394226 - 18.546264410018921

19.1470787525177 - 19.297282338142395

19.898096680641174 - 20.048300266265869

20.649114608764648 - 20.799318194389343

21.400132536888123 - 21.550336122512817

22.151150465011597 - 22.301354050636292

22.902168393135071 - 23.052371978759766

23.653186321258545 - 23.80338990688324

24.404204249382019 - 24.554407835006714

25.155222177505493 - 25.305425763130188

25.906240105628967 - 26.056443691253662

26.657258033752441 - 26.807461619377136

27.408275961875916 - 27.55847954750061

28.15929388999939 - 28.309497475624084

28.910311818122864 - 29.060515403747559

29.661329746246338 - 29.811533331871033

30.412347674369812 - 30.562551259994507

31.163365602493286 - 31.313569188117981

31.91438353061676 - 32.064587116241455

32.665401458740234 - 32.815605044364929

33.416419386863708 - 33.566622972488403

34.167437314987183 - 34.317640900611877

34.918455243110657 - 35.068658828735352

35.669473171234131 - 35.819676756858826

36.420491099357605 - 36.5706946849823

37.171509027481079 - 37.321712613105774

Count (NB log scale)

1

10

100

1,000

10,000

100,000

1,000,000

Resource Assessment

Extracting the Height data for the areas of interest.

The treatment for woodland data and hedgerow data needs to be different due to the structural nature

of the features. The variance that one might expect between woodlands of different management

status, stocking density and species coupled with the variance one would expect with hedges would

render any statistical analysis ineffective.

The process for both however involves isolating the LiDAR data to polygons of woodland (generated

form the NFI data) and polygons for the hedgerow data. For the latter the field polygons used in Single

Farm Payments are used, converted to poly lines, given a 2.5 metre buffer. Using this methodology

firstly ensures that only farmed hedges are assessed and that there is no double capture of the data.

Each raster data set captured in the polygons is subjected to statistical analysis to identify any

groupings.

Results Analysis: Woodland

Observed Data Limitations for the woodland.

There are areas of forest that do not appear to have been filtered correctly to account for the tree cover

in the LiDAR DSM layer; this will lead to under estimation of volumes in those patches. On initial

investigation, it is the broadleaf wooded areas that are underestimated. This might be a more likely

occurrence if the flight is done after the leaves have fallen in autumn and bud burst in the spring.

Without the leaf area to reflect the scanning beam, a higher percentage of beams will strike the ground,

giving an appearance of less dense or even no woodland.

Figure 15 Indication of under measuring over broadleaf woodland areas in filtered LiDAR data

Despite the above data issue, some effort was made to attempt to make some analysis from the

information.

The best that could be drawn from the data are the maximum heights within the various woodland

blocks. Normally one would use the relative proportion of ground readings compared to elevated

canopy readings to give an estimation of woodland cover/density. This does not give the stem density

(i.e. stems per hectare) .

In an effort to measure stem density (Number of stems per hectare) the triangular information network

was created. (this process along took 28 hours of computing time.) A manual count from the TIN and a

manual count from the same area of young conifer stand led to an underestimate of total stems by

almost da factor of 2.

Figure 16 Woodland blocks and the registered maximum feature height from LiDAR data

No meaningful volumetric measure can be derived from this data as it is presented and with the current

software. A more reliable measure would be the mean height of the highest 20% of the readings in the

polygons. A better analysis might be done with the LiDAR data if the unfiltered data was available.

Results Analysis: Hedges:

The polygons were treated to a statistical analysis of the LiDAR data contained within them for

maximum height, mean and median height, coefficient and variance and volume. After various

multivariable analyses, the bulk volume per unit length was derived as the best method to create

classifications. There adjustments made to the classes by the mean and maximum height

measurements.

The classifications were sampled at random at inspected against the 2006 aerial photography. There

appeared a reasonable correlation. However the 2006 aerial image data was the most recent

photographic data available as a layer on the GIS. This data is 7 years out of sync with the LiDAR data

and therefore the verification cannot be entirely relied upon.

Figure 17 Hedge Data processed and interim classified according to Walton, 2014

Figure 18 Zoomed in view of Hedgerow mapping and classification

The following table indicates the heat energy potentially available within the footprint of the LiDAR data

acquired (27.25 Km2). Under normal circumstances one would try to produce some statistical

qualification and error margins.

The results were ground trothed by visiting 2 farms which indicated a range of hedge volume classes.

The hedges were measured and classified against the system proposed in the Devon Hedge Group

publication.

The data were also refined by adjusting the outlier data points of extreme high values and negative

values.

Figure 19 Distribution of bulk volume per unit length of the hedges

The final energy analysis after re-calibrating form the ground truthing is as follows:

Hedgerow

Class

Sub-total length (m) Subtotal Energy MWh

1 62589 0

2 43166 10813

3 107342 80604

4 64692 97289

5 64432 203654

6 16887 106075

Totals 359,110 498,437

Graph of Hedges_Final

Vol_per_m

[0.002; 0.516) [2.056; 2.57) [4.11; 4.624) [6.164; 6.677) [8.218; 8.731) [10.785; 11.299) [13.353; 13.866) [15.92; 16.433) [18.487; 19.001) [21.055; 21.568) [23.622; 24.135)

Count

280

270

260

250

240

230

220

210

200

190

180

170

160

150

140

130

120

110

100

90

80

70

60

50

40

30

20

10

0

Figure 20 Mapping of the reclassified hedges

Conclusions of the study: The figures are quite impressive, with over 360Km of harvestable hedgerow potentially yielding

500GWh of heat if chipped and dried. This is just a snap shot and does not give an indication of

maximum sustainable annual yield. Such a calculation would need to take into account biodiversity

needs etc. Many of the much higher yielding hedges are either next to the woodlands or are alongside

the river banks. Both of these are essential ecological components providing shade and connecting

habitat.

Given that the same data set seems to be under representing the deciduous woodland, the gross

volumes occupied by the trees in the hedgerows may also be under presented.

To sum up;

• The study has explored the use of various remote sensing data available to public bodies to

assess the resource available in the landscape.

• LiDAR data appears to be the most suitable data for its horizontal resolution and vertical

accuracy. Raster data with pixel sizes more than 2 metres will not be able to give a suitable

estimate.

• The season within which the data is gathered will impact on the accuracy of the assessment.

• LiDAR can be a very suitable tool for the assessment of hedgerow resources in an area. Some

ground truthing will be needed to ensure the model is calibrated correctly.

• Conifer woodland measurements are likely to be more accurate than deciduous woodland.

• Onsite measurements of height, DBH and or Basal Area are needed to support any information

gathered by the Environment Agency with LiDAR on an aerial platform for woodlands.

• The snap shot given by this type of analysis does not show the age structure and what the long

term annual sustainable yield will be from the landscape from neither the hedgerows nor the

woodlands.

Further work recommended:

• Field measurements of the woodlands.

• Seek to acquire the raw LIDAR data to explore better bespoke filtering for woodlands.

• Further field measurements of the hedges to calibrate the model more precisely.

• Engage with the landowners in the possible market opportunities for harvesting.

REFERENCES and END NOTES i http://www.hedgelink.org.uk/wood-fuel.htm ii Wood Fuel From Hedges, ISBN 978-1-84785-042-3

iii http://www.cordialeproject.eu/en/toolkit/tools/tool_01_hedgerows_mapping_methodology/

iv http://www.cordialeproject.eu/images/uploads/toolkit-tools/12_b1-

Calculation%20of%20biomass%20volume%20from%20woods%20and%20hedges%20for%20non-

specialists_Farm%20toolkit.pdf v

http://www.cordialeproject.eu/en/toolkit/tools/tool_12_feasibility_assessment_tools_for_woodfuel_production_

and_woodfuel_b/ vi http://www.biomassenergycentre.org.uk/portal/page?_pageid=74,373197&_dad=portal&_schema=PORTAL and

http://www.forestry.gov.uk/fr/INFD-7SUE6F