Costal region of northern Peru, the pacific equatorial dry forest there is recognised for its unique endemic biodiversity

Transcription

1

2

3 Costal region of northern Peru, the pacific equatorial dry forest there is recognised for its unique endemic biodiversity Highly threatened ecosystem affected by agroindustrial expansion in the region and climate variations. Although highly threatened, the forest provides livelihoods and ecosystem services to local communities, mainly through one of the key stone species Prosopis used for its hard wood and its fruits.

4 A disease, probably a combine effect of climate change and other threats to the habitat is affecting Prosopis which is dying fast. Kew is involved with the local communities to provide a monitoring system to help understand threats to the habitat and provide viable adaptation strategies. For this monitoring system it is essential to first identify the main key stone species, Prosopis so we can then measure the extent of the damage and eventually monitor the species health

5

6 Video is using UAV images to get the conservation message through Figures and statistic: quantification is an essential step to gain information from the imagery Huge amount of data: automated procedures

7 There are two phases to the project: 1. Identify and quantify individual species from the UAV images using an automated approach. Here quantification is the challenge 2. Monitor their health and damage caused by the disease For this, calibration is the challenge

8 12 tree species, 4 dominant ones: Prosopis pallida: The key stone species. Evergreen tree of up to 10 meters high with a very distinctive star shaped crown Prosopis palida Cordia lutea: Bush that can get confused with Prosopis but it is deciduous and only up to 2 meter high Cordia lutea Capparis scrabrida: Evergreen tree but very distinctive for its compact and round shaped crown Vallesia glabra: Small bush growing under the canopy of Prosopis Capparis scabrida Valesia Glabra

9 Very high resolution images allowing to see individual crowns (8 cm resolution) Red edge camera: discriminate between species Overlapping images stereoscopic DEM. 3-dimentional information Red Edge camera DEM

10 Capparis scabrida Vallesia glabra Prosopis pallida Cordia lutea

11 Image analysis environment that allow us to take this information in to account. Return number of trees (identify and count individual trees) Object Oriented Image Analysis: objects vs pixels Delineate tree crowns (image segmentation) The challenge. Iterative process first segmentation of the image and then group this initial objects into more meaningful ones Once individual crowns are delineated: Trees are our objects Statistical parameters of for further classification (size, compactness, border index, basic ratios, heights from DEM)

12 Further classify tree objects into different tree classes The aim at this stage is to identify Prosopis candidates which are then validated on the ground Map of the distribution of Prosopis across the Landscape

13 The next step: measure and monitor Prosopis die-back Ideally looking at different levels of infestation Comparative analysis (monitor) Images to provide meaningful, consistent and comparable information about the vegetation calibration Dead Prosopis Alive Prosopis Infected Prosopis

14 Calibration : corrections related to geometry, radiometry, spectral response and includes instrument calibration in labs Radiometric calibration Convert brightness values in our images (DN) to meaningful estimates of the physical properties of vegetation (surface reflectance values) So far all previous analysis done on the image DN values Data to be comparable across flights, illumination conditions or even across time: DN to reflectance This information can be correlated with ground information on the levels of infestation

15 Standard digital consumer cameras not designed for radiometric fidelity Modified with filters (NI or Red Edge) Refer as broad band : bands usually cover a rather large spectrum and often overlap Spectral response of the camera: three bands in the Blue, Green and Red Edge compared with WorldView-2 Red Edge camera WorldView 2

16 Careful handling of the raw images and using appropriate calibration targets, they can be calibrated and use and measurement devices Still limitations related to: Broad band nature of the images Different flights to cover spectral bands Need of accurate georeference

17 New multispectral sensor about to be released Multispectral + RBG sensor in just one flight Scientific grade measurements Affordable

18 1. Quantification: Extracting information from satellite images nearly always a challenge Land use/ land cover, vegetation mapping, change detection Many off the shelf products. No need to customised information 2. Calibration This is probably an extreme example for calibration It does not have to be a challenge : many surface reflectance products Many application don t need calibrated data

19