the potential of drones for vegetation monitoring: a case ... › wp-content › uploads ›...

The potential of drones for vegetation monitoring: a case study of the tundra in the Krkonoše Mountains

Lucie Kupková, Markéta Potůčková,Lucie Červená, Jakub Lysák, Markéta Roubalová, Jana Müllerová, Tereza Klinerová, Stanislav Březina, Záboj Hrázský, Jan Pačák, Přemysl Janata

Eurosite Remote Sensing Webinar, 4th June 2020. This webinar is financially supported by the European Union.

Project and TeamOperational program Environment: Development of methods for monitoring of the Krkonoše Mts. tundra vegetation changes using multispectral, hyperspectral and LIDAR sensors from UAV - drones (2019 –2023)

Administration of the Krkonoše NP (KRNAP): Stanislav Březina, Záboj Hrázský, Jan Pačák, Přemysl Janata

Charles University Prague, Faculty of Science: Lucie Kupková, Markéta Potůčková, Lucie Červená, Jakub Lysák, Markéta Roubalová

Institute of Botany Czech Academy of Science Průhonice: Jana Müllerová, Tereza Klinerová

Project advisor – Clive Hurford (habitat monitoring specialist; a chair of the Eurosite Remote Sensing Support Group)

Very close collaboration of NP (project manager, botanist, GI department), botanists from Institute of Botany and remote sensing specialists from Charles University


Starting points – why tundra?• Tundra ecosystems (alpine treeless) belong to the most valuable natural phenomena worldwide

• Biotopes above the treeline are very sensitive to various types of environmental factors (… climatic change)

• The decline of some valuable habitats and the expansion of shrubs (esp. dwarf pine) and grass vegetation in recent decades is well documented (Štursová and Kociánová 1995, Harčarik 2007, Hejcman et al. 2009, 2010 Březina et al. in prep.)

• Current management problems:

• reedgrass (Calamagrostis villosa) and purple moor-grass (Molinia caerulea) expand into the mat grass (Nardus stricta) stands

• Dwarf pine expansion

• Changes can be relatively fast in these areas and monitoring them is very important

• Information on management practices is needed… to sustain the zone without human interference? What type of management is better to sustain the high biodiversity?

• Earth observation – a potentially powerful tool for the monitoring (objectiveness, repeatability, data from an extensive area in one moment, several times during season…)

• Different types of biotopes in a small area – the possibility to test various vegetation structure, diversity

• Close collaboration with botanists – from NP, from BÚ… very important support for the deailed UAV vegetation mapping!


Goals• To evaluate and compare potential (advantages vs. drawbacks)

of different types of remote sensing data (multispectral, hyperspectral, LiDAR) for accurate mapping of tundra vegetation cover

• To test the accuracy of different classification methods (pixel and object-based) for different types of vegetation and different types of data. Are the accurate methods available in free software packages?

• To monitor the changes over 5 years (seasonal, inter-annual) –implications for management (type of management vs. zone without human interference)

– Detection of change… change vs. error of the used method

• To propose a methodology for tundra vegetation monitoring using RS data from drone

• To implement the methodology in the NP monitoring practice


Study area – Eastern tundra

• The Krkonoše Mts. – the highest mountain range in Czechia – is on the border of Czechia and Poland (50°44’ N, 15°44’ E)

• Tundra – two sub-alpine plateaus (1,300 to 1,400 m a.s.l.) above the timberline; our study – the Eastern part at the Czech side of the border (16 km2).

• “Arctic-alpine tundra”, displaying affinities to both subarctic and high-mountain regions … one of the most valuable ecosystems of the Krkonoše Mts.

• The vegetation is characterized by a mosaic of subalpine grasslands dominated by Nardus and Carex species and dwarf pine stands, subalpine peat bogs, slopes of corries (kar) and rocks, and sparse alpine grasslands on base-poor soils and other habitats of the Krkonoše lichen tundra at the highest elevations.

• There are many endemic, glacial relicts and rare species as well as unique habitats and geomorphologic components


History of human impact• Humans have interfered with the Krkonoše subalpine tundra at

least for the last five centuries

• Roads network building and summer farming, and the consequent building of the first chalets above the timberline in the 16th century (Luční bouda chalet)

• Followed by clear-cuts of dwarf pine (Pinus mugo)

• The peak of grazing and regular hay management of the low productive grassy tundra was during the 2nd half of the 19th century

• There have been several phases of dwarf pine planting

• Tourism from 18th century

• A considerable part of the mountains was declared as the Krkonoše Mts. national park in 1963, with the “tundra” areadesignated as its core zone without human interference

• × Disturbances from grazing and hay production

Increase in biodiversity (we deal with biodiversity monitoring)


Research plots within the Eastern Tundra 1

2

3

4


1

2

3

4

1 Bílá louka meadow (acidophilous alpine grasslands, species poor stands dominated by matgrass – Nardus stricta, reedgrass – Calamagrostis villosaand purple moor-grass – Molinia caerulea, Norway Spruce - Picea abies)2 Luční hora Mt. (fine grain mosaic dominated by wavy-hair grass - Avenellaflexuosa, moss layer, close alpine grasslands, alpine heathlands, small shrubs, stone sea – boulder scree)3 Úpské rašeliniště bog (boreal ombrotrophic raised bog, hummocks, moss layer and hollows, dwarf pine, grasses, lakes)4 Kar Úpské jámy corrie (species rich subalpine tall grasses, Vaccinium heathlands, small springs, mosses, dwarf pine, subalpine deciduous shrubs, boulder scree… Krakonošova garden… rich biodiversity)

Research plots


Bílá louka meadow

• acidophilous alpine grasslands on the plateau, formed by dense and very species poor stands dominated by matgrass(Nardus stricta), sometimes mixed with other grasses and herbs

• reedgrass (Calamagrostis villosa) and purple moor-grass (Molinia caerulea) often expand into the mat grass stands, forming monodominant stands

• several individuals of Norway spruce (Picea abies) with adjacent bilberry (Vaccinium myrtillus)

Luční hora Mt.

• northern slope of Luční hora Mt.wind-exposed alpine grasslands (strong winds, thin snow layer, deep frosts)

• fine grain mosaic dominated by wavy-hair grass Avenella flexuosa

• well-developed moss layer between the turfs (Cetrariaand Cladonia)

• closed alpine grasslands in shallow terrain depressions (Nardus stricta etc.)

• alpine heathlands low shrubs of heather (Callunavulgaris) and bilberry

• small stands of dwarf pine (Pinus mugo), Norway spruce and boulder scree

Úpské rašeliniště bog

• boreal ombrotrophic raised bogon the plateau

• a mosaic of hummocks (Eriophorum vaginatum,Trichophorum cespitosum, with a mixture of heather and bilberry, and a well developed moss layer), and hollows (Carexlimosa and C. rostrata, and Eriophorum angustifolium, more or less filled with mosses and water)

• species poor communities with dominant Trichophorumcespitosum accompanied by other raised bog species

• well-developed moss layer at moister sites

• larger stands of dwarf pine

Kar Úpské jámy corrie• steep eastern slope of Úpská

jáma corrie• upper open part mostly species

rich subalpine tall grasslands, dominated by Calamagrostis villosa or Molinia caerulea

• subalpine Vaccinium heathlands on the open slopes and gaps in dwarf pine stands

• bare stony scree fields• several smaller acidophilous high

- mountain springs with a prevalence of mosses

• poorly developed subalpine tall-forb vegetation along smaller streams

• eastern part dense dwarf pine stands and deciduous shrubs


Research set-up

• Four research plots with different types of vegetation

• Each plot extent: 100 × 100 m

• Drone data acquisition: 4 times in the season (mid of June, July, August, September)

• Multispectral (MS), Hyperspectral (HS) and LiDAR data

• MS and HS data pixel size: 3 cm

• Spectral resolution:MS data 3/5/8 bands, HS data 270 bands

• Lidar data scanning density: ca 900 pts./m2

• 4 years of field monitoring

• 5 years of methods testing and results evaluation/comparison


Used instruments

• Drone: DJI Matrice 600 Pro (price: 6,400 EUR)

• Gimbal Ronin (HS camera)

• IMU and GNSS (positional accuracy ca 5 m)

1. MS camera Micasense RedEdge – M®

Spectral bands: Blue, green, red, red edge, near-

IR – multispectral scanner (MSS)

Price: 6,000 EUR

2. RGB camera Sony A7 ILCE-7 24,3 Mpx; lens-

Voigtlander Color – Skopar 21 mm

Price: 920 EUR + 520 EUR

HS camera – Headwall Nano-Hyperspec®

(enables vegetation health evaluation)

• Pushbroom scanner• 640 pixels• lens 17 mm (possible also 12 mm a 23 mm)• 270 bands (398,784 nm – 1 001,84 nm,

spectral sampling 2,24 nm)• Radiometric resolution: 12bit

Price: 40,000 EUR

LiDAR RIEGL miniVUX-1UAV

• Compact & lightweight (1.55 kg / 3.4 lbs)• 360° field of view• Multiple target capability – up to 5 target

echoes per laser shot• Scan speed up to 100 scans/sec• Measurement rate 100,000

measurements/sec

Price: 72,000 EUR

Used methods – MS data acquisition• Data acquisition carried out by our colleagues from KRNAP

(they are very experienced)

• Both cameras (Micasense RedEdge – M and Sony A7 ILCE-7) mounted on the drone

• Scanning from a height of about 50 m above the ground (constant for Luční hora, Bílá louka, Úpské rašeliniště)

• At the area kar Úpské jámy corrie a descending flight trajectory according to the terrain configuration (final altitude also about 50 m above the ground)

• Each area was flown sequentially in parallel flight lines

• forward overlap of 85% and sidelap 65% for the Micasensecamera

• 85% forward overlap and sidelap for the RGB Sony camera

• Calibration panels (ground control points – GCP) for radiometric calibration and geometric transformation used -black and white chessboard 30×30 cm, placed in the corners of the area

Flight trajectories

Calibration panel (GCP)


Used methods – MS data preprocessing

• SW Agisoft Metashape (price ca 3,400 EUR)

• Radiometric calibration – based on calibration panels –automatic detection of the panels in the image + photo in the field – csv file – calibration coefficients

• Geometric correction

• calibration panels (ground control point – GCP) in the corners of the area – placed into the iron pipes – the position of the iron pipes was focused by geodetic GPS with a mean position error of 0.008 m

• the GCP points marked on all RGB images

• automatic optimization of the internal and external orientation parameters

• final average position error of up to 5 cm

• for RGB images applied functions Build dense point cloud, build mesh

• final mosaics created separately for RGB data (from Sony camera) and for multispectral data (from Micasense RedEdge –M camera)


Used methods – field vegetation mapping• Training and validation vegetation data collected by our colleagues from the Botanical

Institute on the same dates as the data from the drones

• Trimble GeoExplorer GeoXH 6000 or Trimble GeoExplorer GeoXH 2008 GPS used

• Positional accuracy after post-processing around – avg 10–15 cm, in some cases 30–50 cm

• Collector for ArcGIS with orthophoto from UAS (taken several days before the vegetation data collection)

• Record a point in the center of vegetation patch and create a buffer around it

• Manual correction of the collected polygons position in GIS based on the orthophoto (small polygons not used in case of uncertain localization)

Outputs:

1. Classification legend (two levels: general – 15 categories, detailed – 50 categories)

2. Training and validation polygons (882 polygons – ca 1/3 for used for training and 2/3 for validation, total area 5901 m2)

3. Vegetation map for each area

• For 2020 GPS with RTK will be used, central point of vegetation patch will be measured and a radius for each mapped vegetation polygon will be recorded as the point attribute – final positional accuracy should be in centimeters

Vegetation map for Bílá louka meadow plot


Classification legend typ Description

af stands dominated by Avenella flexuosa, minimal cover of other species (mainly Anthoxanthum alpinum, Cx bigelowii and herbs up do 10%)

afs stands dominated by Avenella flexuosa, high cover of other species (mainly Anthoxanthum alpinum, Cx bigelowii and herbs)

bahno histosol, dry peatland water holes

bet individual or a group of Betula sp.

bor Vaccinium myrtillus stand, shrubby vegetation

brus Vaccinium vitis-idaea stand, shrubby vegetace

ca stand dominated by Calamagrostis arundinacea, minimal cover of other species (up to 10%)

cetr Cetraria islandica stand

cv stand dominated by Calamagrostis villosa, minimal cover of other species (up to 10%)

cxbig stand dominated by Carex bigelowii

cxlim Carex limosa stand with histosol

desch stand dominated by Deschampsia cespitosa, minimal cover of other species (up to 10%)

erang stand dominated by Eriophorum angustifolium with histosol

ervag stand dominated by Eriophorum vaginatum

hup Huperzia selago stand

jerab individual or a group of Sorbus aucuparia

junc stand dominated by Juncus filiformis, minimal cover of other species (up to 10%)

klec Pinus mugo stand

klen individual or a group of Acer pseudoplatanus

lis lichens (mainly Cladonia sp.)

listnace deciduous trees, see description for details

luz Luzula luzuloides stand

mech mosses

mokvskal wet outcrops covered by mosses

mol stand dominated by Molinia caerulea, minimal cover of other species (up to 10%)

nard stand dominated by Nardus stricta, minimal cover of other species (up to 10%)

niva tall-forb vegetation

ras Sphagnum sp. stand

rostr stand dominated by Carex rostrata with histosol

scirp stand dominated by Scirpus sylvaticus

skala rock outcrop

smrk individual or a group of Picea abies

sut blosckfields, mostly bare, with lichens

ten fine grain mosaic of Trichophorum cespitosum, Eriophorum vaginatum and Nardus stricta

trich stand dominated by Trichophorum cespitosum, minimal cover of other species (up to 10%)

vaculig Vaccinium uliginosum stand, shrubby vegetation

voda water

vrba individual or a group of Salix sp.

vres Calluna vulgaris stand, shrubby vegetation

vyfuk mosaic of rocks, bare soil, mosses and vegetation, wind-exposed alpine grassland

smes see detail description

pram vegetation of subalpine springs

sw Swertia perennis stand, vegetation of subalpine springs

rasros stand of Sphagnum sp. with significant cover of Drosera rotundifolia

Typ Description

1 Boulder scree and artificial surfaces

2 Norway spruce stands

3 dwarf pine

4 subalpine Vaccinium vegetation

5 closed subalpine grasslands

5a Nardus stricta

5b species rich stands with abundance of dicots

6 subalpine tall grasslands

6a Calamagrostis villosa

6b Molinia caerulea

6c Deschampsia cespitosa

7 subalpine tall-forb grasslands

8 alpine heathlands

9 wetlands and mires

10 open water

11 wind-exposed alpine grasslands

12 individuals or small groups of trees

Used methods – classification of vegetation

• MS and HS data from all 4 acquisition dates resampled:

pixel size 9 cm, HS data: 54 bands

• Detailed legend according to the categories present in individual areas

• Supervised classification using field training vegetation data

MS (from both cameras RGB, MSS; RGBMS = RGB+MSS) and HS data the same methods:

1. Pixel based methods: Maximum likelihood - MLC, Support Vector Machine - SVM (SW

ENVI and ArcGIS), Random Forest - RF (SW R or ArcGIS)

2. Object based classification - OBIA: Support Vector Machine or Random forest classifier

(SW ENVI or e-Cognition)

(Jensen, R.H. Introductory Digital Image Processing. A Remote Sensing Perspective. 2005, Pearson

Prentice Hall)

• Images from each date and data type classified (1) separately and also (2) in

multitemporal composite = images (all bands) from all dates were merged and classified

as one image (32 bands for RGBMS data and 216 bands for HS data for each research plot)

• 184 classification outputs (for MS data and HS data) including multitemporal composites

Bílá louka

meadow

Luční hora Mt. Úpské

rašeliniště bog

Kar Úpské

jámy - corrieafs af bare soil/mud borcv afs cxlim cacxbig bor erang deschdesch desch ervag erangmol Dwarf pine junc dwarf pinenard lis dwarf pine listPicea abies nard mol mokvskal

blockfields nard mol

Calluna vulgaris rostr nard

vyfuk Picea abies niva

ten pram

trich pramsw

trichcesp rasrost

vaculig scirp

water Picea abies

Calluna vulgaris blockfields

Calluna vulgaris

Classification legend for individual areas


Used methods – accuracy assessment• Training and validation data collected in the field – polygons:

▪ 1/3 used for training and 2/3 for validation (802 polygons, total area 5,901 m2)

• Random sample points generated within the validation polygons stratified according the summarized area of training + validation data

• Number of validation points for each mosaic: 3,393 (calculated based on reliability of validation accuracy on 2% level (Foody, G.M. Sample size determination for image classification accuracy assessment and comparison. Int. J. Remote

Sens. 2009, 30, 5273–5291)

• Minimal number of points for 1 category: 50

• Points generated in the center of pixels

• Minimal distance between the points: 13 cm (no points were generated in neighboring pixels)

• Standard methods for classification accuracy assessment used:

• Overall accuracy (OA) – share of correctly classified pixels on the total number of pixels

• Kappa index – compares the result of the classification with the classification created by a random process of classifying pixels into individual classes, 1 means a perfect match and 0 represents a purely random result

Measures for individual categories accuracy evaluation

• Producer’s accuracy (PA) – the probability that a reference class is classified correctly

• User’s accuracy (UA) – the probability that a pixel in the classification actually represents this reference (field) class

(Jensen, R.H. Introductory Digital Image Processing. A Remote Sensing Perspective. 2005, Pearson Prentice Hall)


Results – Competitive species – seasonal changes in phenology

June

July

August

Nardus stricta (NS) Calamagrostis villosa (CV) Molinia caerulea (MC)

Seasonal RGBMS images

NS

NS

NS

NS

NS

NS

CV

CV

CV

?

?

?

CV

MC

MC

MC

MC

MC

June

July

August

Results – Examples of the acquired dataTrue color RGB

June

August

True color HS False color HS

Deschmpsia cespitosa

CV

NS

MC

DC


Results – Classification outputsBílá louka meadow Luční hora Mt. Úpské rašeliniště bog

Results – Classification outputs: Kar Úpské jámy – corrie


Results – Accuracy – after the first year! … reliability will be improved during the next three years

0

10

20

30

40

50

60

70

80

90

Avg OA for all methods Avg PA for all methods Avg UA for all methods

Bílá louka meadow MS data Bílá louka meadow HS data Luční hora Mt. MS data

Luční hora Mt. HS data Úpské rašeliniště bog MS data Úpské rašeliniště bog HS data

• Best results reached for Bílá louka meadow –HS data rather better in average values

• Luční hora Mt. – better results for MS data

• In average results of HS and RS data comparable… we will continue with MS data(RGBMS)

Comparison of average accuracy (OA, PA and UA) of RGBMS (RGB and MSS camera) and HS data in all areas of interest for all months (in %)


Results – Which term can provide the best results for RGBMS data?

0

10

20

30

40

50

60

70

80

90

Avg OA Avg PA Avg UA

Comparison of average accuracy OA, PA and UA for individual months for RGBMS data (in %)

June July August September MT composite

• Best values for multitemporal composites

• Best individual month: July• Worst month: June


Results – RGB vs. MSS vs. RGBMS data

70

75

80

85

avg RGB avg MSS avg RGBMS

Average overall accuracy for MT composites in % (MLC, SVM and RF methods)

0

10

20

30

40

50

60

70

80

90

100

Bílá louka meadow Luční hora Mt. Úpské rašeliniště bog Kar Úpské jámy - corrie

Comparison of overal accuracy for different classifiers and differentcombinations of MS data (in %)

MLC RGB MLC MSS MLC RGBMS SVM RGB SVM MSS

SVM RGBMS RF RGB RF MSS RF RGBMS OBIA RF RGBMS

• RGB data significantly worse results in avg OA for all methods and all areas in comparison with MSS and RGBMS

• MSS and RGBMS data often comparable results, for corrie RGBMS rather better, for bog in some cases better MSS

• In general best results for Bílá louka meadow, second best for LH mountain, the worst for corrie…. higher number of categories (diversity of habitats) worse classification result

• MLC and SVM methods comparable, OBIA sometimes good, RF worse for more diverse vegetation


afs

stands dominated by Avenella flexuosa, high cover of other species

(mainly Anthoxanthum alpinum, Cx bigelowii and herbs)

cv

stand dominated by Calamagrostis villosa, minimal cover of other

species (up to 10%)


desch

stand dominated by Deschampsia cespitosa, minimal cover of other

species (up to 10%)

mol

stand dominated by Molinia caerulea, minimal cover of other species

(up to 10%)

nard

stand dominated by Nardus stricta, minimal cover of other species (up

to 10%)


Results – Best classification output for RGBMS data – Bílá louka meadow

! Classification on practically species level

MT composite (all terms), MLC classifier (available in free software)

! Overall accuracy: 93.5%, avg. PA = 71.3%, avg UA = 77.6

0,00%

20,00%

40,00%

60,00%

80,00%

100,00%

120,00%

afs cv cxbig desch mol nard smrk

PA and UA for all classes in %

PA UA

Results: Vegetation categories – classification accuracy – comparison of classification methods – Example of Bílá louka meadow RGBMS data

0

10

20

30

40

50

60

70

80

90

100


avg MLC avg SVM avg OBIA avg RF

Average producer’s accuracy in %

0

10

20

30

40

50

60

70

80

90

100


avg MLC avg SVM avg OBIA avg RF

Average user’s accuracy in %

afs

stands dominated by Avenella flexuosa, high cover of other species

(mainly Anthoxanthum alpinum, Cx bigelowii and herbs)

cv

stand dominated by Calamagrostis villosa, minimal cover of other species

(up to 10%)


desch

stand dominated by Deschampsia cespitosa, minimal cover of other

species (up to 10%)

mol

stand dominated by Molinia caerulea, minimal cover of other species (up

to 10%)

nard

stand dominated by Nardus stricta, minimal cover of other species (up to

10%)


• PA and UA best results often MLC • Very good results PA: cv, desch, mol, nard (around

80%)• UA over 90%: cv, mol; over 80%: desch, nard• Bad results: afs, cxbig, smrk


Limitations of drone technology

• Administration, permissions, rules, training

• Dependence of the weather (rain, wind, clouds)

• Short flight time (about 20 minutes) - fast battery discharge – small acquisition area (relatively time consuming for larger areas in comparison with other RS technologies) x technology enabling long flight time exists (for example Wingtra drone – 1 hour flight possible… more expensive)

• Extensive overlap of images/flight lines required (many flight lines/images… in the dependence on required spatial resolution and flight altitude)

• Disturbance of animals (peregrine falcon - Falco peregrinus in corries – June nesting time)

• Movement of vegetation (especially tall grass) – unsharpened data

Advantages of drone technology

• Maximal operability (in comparison with other RS technologies)… possibility to acquire data anytime in the case of suitable weather

• Extremely good time resolution (multi-seasonal, inter-annual)

• Extremely good spatial resolution (cm)

• Possibly extremely good spectral resolution – HS camera (hundreds of bands)

• Possibility to combine different types of sensors (MS, HS, LiDAR, thermal)

• Zoom-in/out – different landscape scales… landscape structure

• Charm of a new technology ☺

Conclusions – drones advantages and drawbacks


Conclusions• Results after the first year – research will continue and the results will gradually improve (collection of the

botanical field training/validation data with significantly better accuracy, improvement of mosaics –especially for HS data, improvements of OBIA classification, consideration of whether to discharge some categories with bad classification accuracy, etc.)

• Close collaboration of RS people and botanists – very important! Communication of project management, botanists and RS specialists on weekly basis for the whole year, many long meetings, collaboration in the field

• For some areas of interest very good results at the level of individual species – Bílá louka – great value of extremely high spatial resolution advantage of the drone – can fly at low altitude above ground

• The higher the diversity of the vegetation (habitats) = the less accurate the classification results (after the 1st year)

• Great added value – multitemporal composites (seasonal changes kept) the drone operability brings possibility to get multi-seasonal data – significant accuracy improvement (possibility to monitor inter-annual changes!¨)

• MS data – results comparable with HS data – good news for the conservation practice!!! (because of the lower price, MS data easier acquisition, preprocessing and processing)

• (UAV + RGB: ca 7,800 EUR; UAV + MSS: ca 12,400 EUR; UAV + RGBMS: ca 13,900 EUR; UAV + HS: 46,400 EUR; UAV + LiDAR: 78,400 EUR)

• HS data – assumption of increasing the accuracy with the better quality of mosaics (data acquisition during stable light conditions) and more accurate training vegetation data. Main value of HS data in other applications – e.g. evaluation of vegetation health; irreplaceable… MS data cannot be used for evaluation of vegetation state and health status (our other project)


Future work• Further improvements of classification accuracy

- especially for areas with high variety of habitat/species

• Improvement of botanical maps and comparison with classification output (better positional accuracy of training data needed)

• Assessment of error of used methodology x inter-annual change (what is change and what is error/inaccuracy?)… spatial co-registration of images from different dates with high positional accuracy needed

• What are the best classification methods for different vegetation classes

• Quantification of area of individual categories

• Spatial overlay of classification outputs - quantification of the area covered by the same categories classified by different methods

• Practical outputs for the Krkonoše NP Administration, for the management practice: method of vegetation monitoring using RS data from drone, monitoring plan, training

• Implementation of the proposed method in conservation practice of the KrkonošeNP – based on experiences of the whole team (GIS/RS department and botanists from all institutions)

… our colleagues from the Krknoše NP are very advanced already now ☺


The potential of drones for vegetation mapping:Take home message…

• Close collaboration and communication between botanists and RS specialists is essential (in the field and also during the data analysis)

• Remote sensing specialist/s can be member/s of the team of nature conservation practitioners

• Drone technology (combination of drone and RGB/MSS camera) enables high accuracy and detailed vegetation mapping at the species level, at least for some habitats

• It is a good time to use drone technology because the technology is really accessible for theconservation practice (price, manipulation, data acquisition, pre-processing and analysis; accurate classification methods – Maximum Likelihood - available in free software packages – QGIS, Grass)

• Operability – anytime (good weather), any area, several times during season, the whole area in the same time; objectiveness, repeatability… not possible to do this using traditional forms of vegetation mapping

• If you are aware of limitations and advantages of the technology it can improve the way that we monitor vegetation - biodiversity/spatial distribution and changes/health - in nature conservation

• Be sure that our Eurosite Remote Sensing Support Group is here for you and ready to help… ☺

Thank you for your attention!

[email protected]

the potential of drones for vegetation monitoring: a case ... › wp-content › uploads ›...

Documents