Building skills in using Unmanned Aerial Vehicles and object-based image analysis for ecological surveys

Size: px

Start display at page:

Download "Building skills in using Unmanned Aerial Vehicles and object-based image analysis for ecological surveys"

Gervais Hopkins
5 years ago
Views:

surveys Simon Pickles 1, Paula Lightfoot 2 Dr C.

Scott 3 1 North and East Yorkshire Ecological

1 Building skills in using Unmanned Aerial Vehicles and object-based image analysis for ecological surveys Simon Pickles 1, Paula Lightfoot 2 Dr C. Fitzsimmons 2, Prof. N. V. C. Polunin 2 and Dr C. Scott 3 1 North and East Yorkshire Ecological Data Centre, 2 Newcastle University, 3 Natural England

2 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Simon s Bit Could be seen as a case study following on from Cameron s workshop this morning. 1. North Why do we and want to operate East an Yorkshire SUAV? 2. How did we get one? 3. What opportunities has the SUAV created? 4. Why we need partnerships to maximise the usefulness of our SUAV? Ecological Data Centre (NEYEDC) Paula s Bit Case study of partnership work to investigate using the SUAV for intertidal habitat mapping Current Partners, Priorities and Projects

3 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Why do we want to operate an SUAV? North and East Yorkshire Ecological Data Centre is one of a small number of fully independent LERCs. It is the primary operating function of a UK registered charity, the Yorkshire and Humber North Ecological and Data Trust East (YHEDT). Yorkshire For several years now, NEYEDC has been trying to move away Ecological from the traditional reactive Data data Centre collation and (NEYEDC) dissemination model, to a pro-active model in which are activities are increasingly question driven. Unfortunately, most of the funders approve of the new model, but pay us for the old one. Not really your problem! Current Partners, Priorities and Projects

North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust A wide range of tools are becoming available that would allow us to address

4 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust A wide range of tools are becoming available that would allow us to address some of the key questions being posed by our partners and clients: - Habitat suitability modelling. Ecosystem functionality assessment Systematic conservation planning North and East Yorkshire e.g. Marxan can find the most spatially efficient network to achieve the targets at minimum cost. Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

5 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Three key obstacles keep emerging that limit the usefulness of these tools at a landscape scale: - The available habitat classification / attribution systems. The lack of consistent habitat condition assessment methodologies that can be applied both within and without protected areas. Quality and extent of habitat mapping at a scale which is meaningful for landscape scale projects. North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

6 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Defra Approach and Priority Actions? Our long term vision is for a more coordinated and coherent approach to habitat mapping and assessment, that combines and links data at different scales North and resolution and of sources East to provide Yorkshire a comprehensive habitat map for England. This will combine data from Earth Observation and field survey to provide Ecological a basic level of information Data on Centre the extent and (NEYEDC) condition of priority habitats, with additional information, at higher resolution for high value components such as protected sites. The map will support both local decision making and targeting of actions as well as reporting on outcomes for the Habitat Directive and Biodiversity 2020 strategy. Biodiversity monitoring and surveillance: the Defra network strategy and action plan for the monitoring and surveillance of terrestrial, coastal and freshwater biodiversity in England Draft Document, November 2014 Current Partners, Priorities and Projects

7 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust How did we get our SUAV? Careful what you wish for! 1. It all started out with a conversation in the pub. 2. Which led to a fantastical idea presented at a Christmas Trustee s meeting. 3. Followed by a Serendipitous request to apply for an undersubscribed Innovation Fund. 4. Which led to a desperate last minute bid. 5. Which 6 weeks later led to Desmond turning up at the office. North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

8 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

9 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust What opportunities has the SUAV opened up for NEYEDC this year? Support existing habitat mapping and condition assessment campaigns. North Support partner s and funding East bids. Yorkshire Act as an advocate for the UAV as a tool. Ecological Contribute to partnerships Data working Centre towards better (NEYEDC) data analysis and workflows that incorporate the use of UAV data with field work and remote sensed data from other platforms. Current Partners, Priorities and Projects

Support existing habitat mapping and condition assessment

10 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Support existing habitat mapping and condition assessment campaigns. North and East Yorkshire Ecological Data Centre (NEYEDC)

Data Trust Provide enhanced resolution images at critical

11 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Provide enhanced resolution images at critical points in time. North and East Yorkshire Ecological Data Centre (NEYEDC)

Ecological Data Trust Provide additional

12 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Provide additional information. North and East Yorkshire Ecological Data Centre (NEYEDC)

13 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Support partner s funding bids. North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

14 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Act as an advocate for the UAV as a tool. North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

Ecological Data Trust North and East Yorkshire

15 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

16 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

17 North and East Yorkshire Ecological Data Centre An operating function of the Yorkshire and Humber Ecological Data Trust Contribute to partnerships working towards better data analysis and workflows that incorporate the use of UAV data with field work and remote sensed data from other platforms. North and East Yorkshire Ecological Data Centre (NEYEDC) Current Partners, Priorities and Projects

18 Case study: collaboration to build skills and knowledge PhD research: Integrating ecological knowledge and multi-resolution remote sensing data to inform Marine Protected Area planning. Funded by NERC and Natural England, January 2014 to June Aim to develop and test object-based image analysis (OBIA) methods for mapping and monitoring marine habitats subtidal and intertidal.

imagery, producing GIS-ready outputs An increasingly popular method of

et al, 2014) OBIA methods are likely to play a key role in UK habitat

19 What is OBIA? User creates automated workflows to segment and classify layers of imagery, producing GIS-ready outputs An increasingly popular method of interpreting remote sensing data since around the year 2000 (Blaschke et al, 2014) OBIA methods are likely to play a key role in UK habitat monitoring (Medcalf et al. 2015) The application of OBIA to marine data is still in its infancy (Diesing, Mitchell and Stephens 2016)

Blaschke et al, 2010 Benz et al, 2004 Benefits of OBIA Remote sensing data is becoming available at ever higher resolution objects of interest may be larger than individual pixels.

Objects have more properties than single pixels do: Mean, mode, max, min, standard deviation, skewness etc of spectral values. Geometric features e.g. shape, size, orientation.

20 Blaschke et al, 2010 Benz et al, 2004 Benefits of OBIA Remote sensing data is becoming available at ever higher resolution objects of interest may be larger than individual pixels. Grouping pixels to form objects avoids the salt-and-pepper effect produced by pixel-based classification. Objects have more properties than single pixels do: Mean, mode, max, min, standard deviation, skewness etc of spectral values. Geometric features e.g. shape, size, orientation. Texture, e.g. rugosity. Context and hierarchy relation to neighbour objects, super-objects, sub-objects. This enables users to integrate their ecological knowledge and contextual information into the segmentation and classification process. Low resolution High resolution

Access to OBIA Increasing availability of

of training Quality of documentation and

21 Access to OBIA Increasing availability of commercial and open source OBIA software makes it (too?) easy to produce plausible, visually pleasing maps from remote sensing data. RSGISLIB Potential barriers to uptake by ecologists: Software, hardware and maintenance costs Cost and availability of training Quality of documentation and user community support Access to scientific papers about OBIA Time to develop and test robust, standardised methods Collaboration between academia and practitioners helps to build skills and knowledge, evaluate OBIA methods and facilitate their uptake.

22 Evaluating UAV and OBIA for intertidal habitat mapping Study site: Kettleness headland in Runswick Bay, North Yorkshire

23 Data Collection UAV imagery 17 flights between April - September flights with RGB camera, 8 with Red Edge. 1,500 images captured: Image size approx. 150 x 200 metres Pixel size approx x 0.04 m 60% overlap between images Ground truth data Stratified random sampling protocol based on shore height 144 x 0.25m 2 quadrats

24 UAV Data Processing Test area: East Kettleness: 194,470 m2 3rd July 2015

OBIA workflows OBIA workflows were developed in ecognition Developer v 9.0.

25 OBIA workflows OBIA workflows were developed in ecognition Developer v Three classification approaches: Two sets of training samples Standard Nearest Neighbour (2-fold cross validation) Random Forest (Breiman, 2001) Knowledge-based hierarchical rules using thresholds and membership functions Mean DSM Threshold Laminaria digitata Fucus vesiculosus Membership Function 45 F. spiralis F. vesiculosus F. serratus L. digitata

26 OBIA workflows OBIA workflows were developed in ecognition Developer v Three classification approaches: Two sets of training samples Standard Nearest Neighbour (2-fold cross validation) Random Forest (Breiman, 2001) Knowledge-based hierarchical rules using thresholds and membership functions Two levels of thematic resolution: Broadscale habitats red algae, green algae, brown algae, barnacles/bare rock Biotopes (Connor et al, 2004) Two datasets: Kettleness East, July 2015 Kettleness West, Sept. 2015

Dataset 2 (Sep. 2015) Dataset 1 (July 2015) Results broadscale maps Overall Accuracy Kappa Balanced Error Rate Standard Nearest Neighbour (training samples a) 80% 0.

27 Dataset 2 (Sep. 2015) Dataset 1 (July 2015) Results broadscale maps Overall Accuracy Kappa Balanced Error Rate Standard Nearest Neighbour (training samples a) 80% % Standard Nearest Neighbour (training samples b) 85% % Random Forest (training samples a) 95% % Random Forest (training samples b) 90% % Knowledge-based 84% % SNN (training samples a) 77% % SNN (training samples b) 87% % Random Forest (training samples a) 68% % Random Forest (training samples b) 88% % Knowledge-based 80% % Example broadscale habitat maps produced using the knowledge-based OBIA workflow: Mid Broad B FINAL Class_name Barnacle/rock Brown algae Cliffs Green algae Red algae Sea Shadow

Dataset 2 (Sep. 2015) Dataset 1 (July 2015) Results biotope maps Overall Accuracy Kappa Balanced Error Rate Standard Nearest Neighbour (training samples a) 41% 0.

66 24% SNN (training samples a) 28% 0.10 77% SNN (training samples b) 27% 0.12 76% Random Forest (training samples a) 59% 0.

28 Dataset 2 (Sep. 2015) Dataset 1 (July 2015) Results biotope maps Overall Accuracy Kappa Balanced Error Rate Standard Nearest Neighbour (training samples a) 41% % Standard Nearest Neighbour (training samples b) 56% % Random Forest (training samples a) 66% % Random Forest (training samples b) 63% % Knowledge-based 70% % SNN (training samples a) 28% % SNN (training samples b) 27% % Random Forest (training samples a) 59% % East Biotope A FI Class_name Cliffs Coff Cor Random Forest (training samples b) 54% Low accuracy 0.41 due to misclassification 64% of the red alga Osmundea as Corallina caused by a change Knowledge-based 31% % in this species spectral signature from July to East Biotope A FINAL Example biotope maps produced using the knowledge-based OBIA workflow: September (bleaching). Class_name Eph.Ent Fser.R FspiB FvesB FvesR Cliffs Ldig Coff Osm Cor Rkp.FK Eph.Ent Sea Fser.R Sem FspiB Shadow FvesB bare rock FvesR unclassifie Ldig Osm Rkp.FK

29 Dataset 1 (Kettleness east, July 2015) Dataset 1 (Kettleness east, July 2015) Results consistency Percentage of map area with the same classification in both maps Dataset and thematic scale Standard Nearest Neighbour Random Forest Dataset 1: Broadscale habitat map 79% 92% Dataset 2: Broadscale habitat map 74% 83% Dataset 1: Biotope map 32% 62% Dataset 2: Biotope map 23% 72% SNN (training samples A) SNN (training samples B) Agreement between map A and map B (79%) SNN (training samples A) SNN (training samples B) Agreement between map A and map B (32%)

30 Implications and Next Steps OBIA of UAV imagery shows great promise for surveys in the intertidal environment: Small scale Difficult to access Highly dynamic: needs frequent revisits High spatial complexity Seaweeds have distinctive spectral qualities Communities inhabit vertical zones All three classification methods produce reliable intertidal broadscale habitat maps. The knowledge-based approach is more cost-effective and consistent as it does not rely on collection of training samples. The knowledge based approach shows promise for biotope mapping but further research is needed to improve reproducibility, i.e. to overcome variation in the imagery due to seasonal change and lighting conditions. The real value of OBIA for intertidal surveys lies in the development of large scale rapid assessment methods to target UAV/ground survey effort more effectively. Currently researching OBIA methods to detect change in intertidal habitats from time-series aerial and LiDAR data, and building collaboration to seek funding to develop these proofs of concepts into cost-effective monitoring methods.

31 References Benz, U.C. (2004) Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry & Remote Sensing, 58, Blaschke, T. (2010) Object based image analysis for remote sensing. ISPRS Journal of Photogrammetry and Remote Sensing, 65. Blaschke, T., Hay, G.J., Kelly, M., Lang, S., Hofmann, P., Addink, E., Feitosa, R.Q., van der Meer, F., van der Werff, H., van Coillie, F. & Tiede, D. (2014) Geographic Object-Based Image Analysis - Towards a new paradigm. ISPRS Journal of Photogrammetry and Remote Sensing, 87, Breiman, L. (2001) Random forests. Machine Learning, 45, Connor, D.W., Allen, J.H., Golding, N., Howell, K.L., Lieberknecht, L.M., Northen, K.O. & Reker, J.B. (2004) The Marine Habitat Classification for Britain and Ireland Version Diesing, M., Mitchell, P. & Stephens, D. (2016) Image-based seabed classification: what can we learn from terrestrial remote sensing? ICES Journal of Marine Science. Medcalf, K., Parker, J., Breyer, J. & Turton, N. (2015) MEOW Phase 3: Cost effective methods to measure extent and condition of habitats. A report produced by Environment Systems Ltd for Defra and the JNCC. Peterborough.

32 Thank you! Any Questions? With thanks to:

GEOMATICS. Shaping our world. A company of

GEOMATICS. Shaping our world. A company of GEOMATICS Shaping our world A company of OUR EXPERTISE Geomatics Geomatics plays a mayor role in hydropower, land and water resources, urban development, transport & mobility, renewable energy, and infrastructure