Urs Treier & Signe Normand - UAS4RS Conference 2016

Tundra change at the dawn ofdrone ecologyUrs Treier & Signe NormandEcoinformatics and Biodiversity & Arctic Research CenterAarhus University, DenmarkPhoto: Normand-Treier

Drone ecology1) potential & challengesThe use of drones2) to answer questions in ecologyAdds to our ability to understand and monitor distribution of species ecosystem processes and underlying environmentalfactorsQuestions are complex and environmental variation is high: many independent measurements (data points) needed generalisation/up-scaling, i.e. many versus few sites design: random, stratified random (ecological gradient), remote areas, collection under optimal conditions notpossible1)Koh LP, Wich SA 2012 Trop Conserv Sci, Ogden 2013 BioScience, Anderson & Gaston 2013 Front Ecol Environ2014 J Unmanned Veh Syst, Clothier et al 2015 Risk Anal, Sandbrook 2015 Ambio2) Chapman

Tundra change across space and timePhoto: Normand-Treier

Tundra change across space and timeWhere? Why?Dynamics?Photo: Normand-TreierMap, measure,modelMonitorUnderstand& predict

SCALE GAPChallenge scale gapAll species are distributedin space – but within limitsPhoto: Normand-TreierThe distribution of allindividuals in spaceconstitutes the range

SCALE GAPChallenge scale gapAll species are distributedin space – but within limitsThe distribution of allindividuals in spaceconstitutes the rangeThe magnitude of changevaries across the rangeInitial manifestation ofvegetation dynamics Linking scales challenging due to scale gapPhoto: Normand-Treier

Scale gap lack of data?

Scale gap lack of datasmå områderGAPStoreområder –grovopløslighedsmåområder –finopløslighed Bridging scales: high resolution data across large areas drones

Arctic tundra change large scaleClimate and ”greening” trends (1982-2010)Change in summer temperatureChange in “greening” (NDVI) What drives the large scale vegetation change?www.climate.govc. 8x8kmkmpixelspixel8x8Guay et al 2015 http://daac.ornl.gov

Arctic tundra change small scaleObserved changes at local scaleIncreased cover / abundance1987pixel 8x8 kmIncreased heightGAPExpansion - recruitmentSturm et al. 2001 Nature, Myers-Smith et al. 2011 Ambio, Elmendorf et al. 2012 Nature Climate Change2009

Arctic tundra change small scaleObserved changes at local scaleIncreased cover / abundanceWhat drives the largescale vegetationchange?orDo local and large scalechanges correlate?Upscaling:Link observationsacross scale extrapolate andgeneralise1987Increased heightExpansion - recruitmentSturm et al. 2001 Nature, Myers-Smith et al. 2011 Ambio, Elmendorf et al. 2012 Nature Climate Change2009

Arctic tundra change aim & questionsWhere? Why?Dynamics?Map, measure,modelUnderstand& PredictMonitorQuestion: What ecological factors determine spatial variation in: abundance? functional traits (e.g., height)? demographic parameters (e.g., recruitment)?Method: space-for-time substitution & projections with modelsNow - assessment across environmental gradients

Arctic tundra change aim & questionsWhere? Why?Dynamics?Map, measure,modelUnderstand& PredictMonitorQuestion: How does arctic tundra change over time and what isthe spatial variability?Method: time series (monitoring), which is challenging: Accuracy and standardisation of (image) data Repeatability and transferabilityFuture – observe changes across time and space

Greenlandic gradients sites and trends2016 browning/greening (1982-2012)arctic growing season (GS-NDVI),GIMMS 3g, 8-km resolution201420152015201520132014Guay et al 2015 http://daac.ornl.gov 4 field campaigns, 7 focal areas large scale and small scaleenvironmental gradients

Fieldwork 2013 Nuuk Fjord – western GL2013

Environmental gradients plot-based dataFigur: A. B. Overgaard, Nabe-Nielsen et al. in prep.; Photo : Normand-Treier

Environmental gradients species abundanceStation 1Nabe-Nielsen et al. in prep.Station 2Station 3Station 4Station 5

Environmental gradients plot-based RSRGB NIRimages of108 plots

Plot-based RS classifying speciesblue channelgreen channelred channelNIRNDVI

Plot-based RS classifying speciesblue channelgreen channelred channelNIRNDVI model performance: overall 60%, 2 species 80% success

Plot-based RS classifying species Single image classification could be optimised but We need transferability of models across many images Image standardisation and test for transferability108 NIR & VISrandomspatial & temporal gets/polygonstrain, validatepredict

Plot-based RS colorimetric calibration

Plot-based RS colorimetric calibrationWavelength (380-730 nm)Berns R et al 2005 ICOM-CC meeting, Ritchie et al 2008 Appl Eng Agric, Fischer et al 2012 Flora

Plot-based RS colorimetric calibration An example from medicine Tested/compared different correction algorithmssame tongue captured using different camerassame tongue & camera, different light conditionimages after colour correctionimages after colour correctionWang W, Zhang D - 2010 - IEEE Trans Inf Technol Biomed

Fieldwork 2014 Zackenberg & BlæsedalenCan parameters derived from droneimagery explain variation in: vegetation cover? Species richness?Imagery biodiversity indicatorsZackenbergBlæsedal20142015Guess how many plant species you mightfind in this landscape in a 2m circle?2015201520132014Blæsedal

Fieldwork 2014 Data13 fights of ca. 5000 m2 (100x50m) - height 50m - 0.5 cm resolution 700 images per flight (NIR plus VIS) above permanent plots at 20, 100, 200, 300 m. asl.Illustration: J. Nabe-Nielsen

Fieldwork 2014 some impressions

Fieldwork 2014 some impressions

Fieldwork 2014 Tundra vegetation Zackenberg

Fieldwork 2014 Tundra “vegetation” BlæsedalCanon EOS 550D / 100D (18MP)with Canon EF 50mm f/1.8 lensNIR modified( 830nm)

Ultra-high resolution example from DKresolution: 5 mm; area: 10000 m2dune vegetationPhoto: Normand-Treier

Ultra-high resolution example from DKdune vegetation

Fieldwork 2015 3 areas along the Arctic Circle201520152015Guay et al 2015 http://daac.ornl.gov

Fieldwork 2015 3 areas along the Arctic Circle8 km resolution1 gimms3g pixs256 Modist pixs71111 Landsat pixs500 mresolution20151 Modis pixs278 Landsatpixs2015201530 m resolution1 Landsat pixsGuay et al 2015 http://daac.ornl.gov

Fieldwork 2015 3 areas along the Arctic Circle10 sitesNDVI trends201520152015Guay et al 2015 http://daac.ornl.govBuffer-averaged yearly max NDVI1.000.1020002015gimms3gmod13q1mod43a4ilandsat

Fieldwork 2015 Questions & expectationsNDVI changeNDVI changeMean SLAMean NDVIMean coverMean heightWhat characterize sites with different rates of NDVI change?NDVI changeexpectationalternativeexpectationNDVI changeTemperatureTemperatureMean SLAMean NDVIMean coverMean heightWhat explains within site variation in NDVI?Space-for-time substitution: what characterizes warmer sites?TemperatureTemperatureBack in time: synchrony temp., NDVI, growth and time

Data GCP, ground truth & trait sampling 2015 10 ground control points, GNSS survey grade ( ) andreflectance panel for image standardisation ( ) 10 ground truth sampling squares, 50x50cm ( ) 25 trait data sampling plots, 2x2m ()

Data Sampling 2015Visuel cover and height estimatesVIS & NIR imagesVegetation class coverEnvironmental parameters moisture, pH, aspect, Dendroecology growth, recruitmentTrait sampling SLA, height, woodGenetic samplesPhoto: Sigrid Nielsen

Data multispectral camera

Data some resultsRGBNDVIDSMNDRENIR REDNIR REDNIR RENIR RE

Data good resolution ( 2cm)RGBNDVIDSMNDRENIR REDNIR REDNIR RENIR RE

Data GCP, ground truth, trait samplingRGBNDVIDSMNDRENIR REDNIR REDNIR RENIR RE

Challenge difficult terrain, adaptive flight planRGBNDVIDSMNDRE354 m a.s.l.308 m a.s.l.Slope 40

Summary take-homeDrones have great potential toimprove our ability to answerecological questions, by: Mapping the current distributionof individuals across larger areas Setting baseline to monitorfuture change Scaling observations from plot tolandscape, region, biomepixel 8x8 kmGAP

Summary take-homeChallenges: Extrapolation requires manyrandom representative samples Change detection requiresstandardization and transferabilityin space and time Extraction of ecologicalrelevant parameters fromimages Comparability of parametersacross remote sensing productspixel 8x8 kmGAP

Thanks Samira Kolyaie, Sigrid Nielsen, Margrete Christiansen, BjarkeMadsen, Damien Georges, Jacob Nabe-Nielsen, AchilleasPsomas, Christian Ginzler, Ruedi Boesh, Niklaus ZimmermannFoto: Normand-Treier

Tak!

1) Koh LP, Wich SA 2012 Trop Conserv Sci, Ogden 2013 BioScience, Anderson & Gaston 2013 Front Ecol Environ 2) Chapman 2014 J Unmanned Veh Syst, Clothier et al 2015 Risk Anal,