Novel Computational Tools to Analyse Fragmented Forests Véronique Lefebvre, Marion Pfeifer, Andrew Bradley and Robert Ewers ATBC 2013
Extracting fragments and their properties from maps � Fragmentation of forest affects biodiversity � To find out how we can: – Define what fragments are and delineate them on maps obtained from satellite images – Estimate potential biodiversity drivers for each fragment (size, shape, connectivity..) � We present novel image processing based methods to – Delineate fragments on maps – Estimate fragments properties � Novel methods can make use of prior knowledge on studied species and local area � And are designed to cope with resolution and Distinct fragments? pixel geometry constraints of raster maps Similar “shape”?
Part 1: Patch delineation by CCL Patch 1 ‘on’ pixel Common technique: Patch 2 ‘off’ pixel Connected Component Labelling (CCL) Habitat binary map Connected component labelling Problems: � Does not represent species perception of landscape - Nor experimenters’ perception of landscape � cannot use prior knowledge � very sensitive to forest classification
Part 1: Patch delineation by new method Patch 1 3 patches Patch 2 � Our method uses ecological knowledge � Connected component labelling New delineation method To disconnect weakly connected chunks of pixels � Delineation reflects species perception of landscape � definition of patches is adaptable 10 km
Effective fragment – Concept and definition How to decide where to “separate” connected pixels? � Chunks of forest may be connected but not perceived as such – A stretch of forest may be too narrow to be a corridor – Habitat suitability may vary with the distance to forest edge � Concept of effective fragment � To delineate effective fragments the method uses: – the Minimum corridor width (MCW) to find weak links – the Depth of Edge Influence (DEI) to find core area � MCW and DEI can be obtained from species abundance data, local knowledge and literature, and map classification confidence
Delineation technique Landscape map Distance map 1) Find cores and corridors from the distance map 2) Find where to cut all weak links (narrower than MCW) MCW / DEI Watershed segmentation with the watershed segmentation 3) Reconnect edge chunks (less then DEI) to most strongly connected core Reconnection � Can incorporate matrix element, e.g. water, pastures, urban, by adding weights to the distance map
Part 1: Patch delineation – Example Result Comoros Islands Forest Binary map of landscape and measurement locations Connected component labelling New delineation method 1 km Watershed delineation � landscape segmented into ecologically meaningful fragments
Part 2 – Fragment characteristics � To compare fragments between each other we can extract their geometrical characteristics from simple binary maps and fragments delineation: – Area Straightforward from – Core area patch delineation – Potential dispersal area from a fragment Forest map Potential dispersal area ? – And shape
Shape descriptors Compactness Compact shape: Smoothness effective in conserving resources Convoluted shapes: effective in enhancing interaction with the surroundings Smooth shapes: Higher resistance to disturbance Compactness and contour smoothness can describe different types of habitat � Commonly used shape descriptors do not distinguish between these 2 properties of shape Bogaert et al. 1999, Environmental and Ecological Statistics
Compactness How packed is the shape? Longest distance within fragment Circle of same area The compactness measure shows the “spread” of the fragment compared to the most packed shape. Advantages: - measures only compactness - does not use a perimeter estimation
Contour smoothness How wriggly is the contour line of a shape? Suggestion: � counting the number of indents in shape contour Method: indents � Counting the number of zero crossing of the contour curve derivatives � Compute the proportion of smooth perimeter: �� � ������������������ � ������������� � Advantage: - Only describes smoothness But it requires a perimeter estimation (which varies with resolution) Perimeter calculated using distances Inspired by: between mid-points of border pixels Bogaert et al. 1999, Environmental and Ecological Statistics
Comparison of common and suggested shape descriptors 6 2 3 On the 10 biggest fragments 4 of the Comoros forest distance from landscape top (km) 2 5 7 6 Effective fragments obtained 10 by the Watershed method 8 4 10 8 9 12 1 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 distance from landscape left (km)
Comparison of common and suggested shape descriptors Fragments ordered by: # 1 # 5 # 6 # 7 # 9 # 10 # 4 # 2 # 3 # 8 Area + - # 1 # 5 # 6 # 9 # 10 # 7 # 4 # 2 # 3 # 8 Shape factor (compactness) # 10 # 6 # 9 # 7 # 5 # 1 # 4 # 2 # 3 # 8 Compactness # 1 # 9 # 6 # 5 # 10 # 7 # 4 # 2 # 3 # 8 Fractal dimension # 5 # 1 # 7 # 10 # 9 # 6 # 4 # 2 # 3 # 8 Smoothness - + Shape factor and fractal dimension classifications mainly reflect area order
Variation of shape descriptors with area Random patches Forest patches from several landscapes New descriptors New descriptors Shape Factor Compactness Shape Factor Compactness 1 30 25 0.8 20 0.6 15 0.4 10 0.2 5 0 1 2 3 1 2 3 10 10 10 10 10 10 Fractal Dimension Fractal Dimension Smoothness Smoothness 1 3 0.8 2.5 0.6 2 0.4 1.5 0.2 0 1 1 2 3 1 2 3 10 10 10 10 10 10 Fragment area Fragment area Our metrics are less determined by area � easier comparison between fragments
Examples of randomly generated patches
Variation of shape descriptors with each other Compactness Shape factor w.r.t. fractal dimension w.r.t smoothness (new) 0 Forest 0 patches 0 0 0 Area 30 0 25 0 20 Random 0 15 patches 0 10 5 0 0 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Our descriptors are not functions of each other � reflect distinct shape properties
What can fragment delineation and descriptors do for forest fragmentation studies � Used to study biodiversity responses to fragments properties (area, potential dispersal area, compactness, smoothness) � But often not enough fragments are measured in a landscape Biodiversity index Trend? Forest map and plot locations compactness � Delineation and fragment descriptors can help selecting plot locations within several patches of widely different properties Fragment delineation is also useful in finding out fragments history � Robert Ewers’ talk at � 8:45 in this session
Thanks ! � To all researchers helping us to collect Biodiversity data for the BioFrag project: http://biofrag.wordpress.com/ � Thanks to the team ! – Andrew Bradley (the remote sensing pro) – Marion Pfeifer (the ecology reference who patiently teaches me everything) – Robert Ewers (the wise boss) � Thanks for your attention � The delineation method and metric code is available in Matlab with the image processing toolbox. It can be recoded in another language or included in an existing software. We are open to collaborations � v.lefebvre@imperial.ac.uk
Recommend
More recommend
