High spatial resolution land use and land cover mapping of ...
Irish Land Mapping Observatory (ILMO) · 2019. 10. 19. · Land-cover mapping • Manual...
Transcript of Irish Land Mapping Observatory (ILMO) · 2019. 10. 19. · Land-cover mapping • Manual...
GIS & Earth Observation
Stuart Green [email protected]
How satellites, drones and digital mapping can help with environmental resource management
Land-cover mapping
• Manual Digitisation
• Photomorphic Labelling
• Vegetation Detection
• Unsupervised Classification
• Supervised Classification
– Maxiumum Likelihood
– Random Forest
– Support Vector Machine
• Object Orientated
• Hybrid Approaches
http://conservationmaven.com/frontpage/first-detailed-national-map-of-us-land-cover-vegetation-rele.html
Land-cover mapping – what is land cover
Habitats are described by Fossitt
GA1: Improved Agricultural Grasslands
GS4: Wet Grassland
GS3: Dry-humid acid grassland
land cover – Land use - Habitat
Land Use is fundamental to GHG reporting- its an ECV
• Forest land
• Cropland
• Grassland
• Wetlands
• Settlements
• Other lands (e.g. bare soil, rock, ice, etc.)
• http://oceanservice.noaa.gov/facts/lclu.html
• Land cover indicates the physical land type such as forest or open water whereas land use documents how people are using the land
Or landcover is what is under your feet- landuse is what you might do with that
https://ghginstitute.org/wp-content/uploads/2015/04/Understanding_Land_Use_in_the_UNFCCC.pdf
land cover – Land use - Habitat
We Often have Levels of classication- the higher the level-the higher the detail
https://www.epa.ie/researchandeducation/research/researchpublications/researchre
ports/Research_Report_254.pdf
Land-cover mapping – what is land cover
Why map Land cover?
Sensors recording optical
wavelengths have long been used
to distinguish land cover classes
based on reflectance patterns
Red (0.6-0.7µm) and NIR (0.9-
1.2µm) to create vegetation indices
During the growing season spectral
signature changes requiring
multiple images to capture
vegetation dynamics and
phenology – cloud can preclude
acquisition of optical images at
optimal times
Growing interest globally to support policy decisions (e.g. European environmental directives and ensure effective land management
One driver is need to provide carbon budget and greenhouse gas inventories – legally binding targets may be enforced in second commitment
period of Kyoto Protocol
Manual
Clicking lines around objects- just like last weeks practical
Automation
Turn an Image into a map- automatically
Replacing the digital numbers in each pixel (that tell us the average spectral
properties of everything in the pixel), with a single number code that represents:
• The majority land cover type in the pixel
• or a biophysical property in the pixel (e.g. amount of biomass)
• or a relative value for a Landover (percentage of pixel that is forestry).
https://landmapping.wordpress.com/
Mapping “Green”
Vegetation Indices are an indictor of the presesnce and abundance of
vegeation
• Biomass: The mass per unit area of vegetation.
• Cover: The vertical projection of the plant parts on
the ground surface per unit area
of ground. Usually expressed as a percent. No
species can have more than 100% cover.
• Leaf Area Index: The ratio of the area of leaves and
green vegetation in theplant canopy per unit area of
ground surface. LAI can exceed 1.The only way to
get true leaf area is to strip all the leaves off the
plants and measure their area. All other methods
provide an “index” of this value
• Normalized Difference Vegetation Index (NDVI):
An index of vegetation greenness derived from
remote sensing methods. Often used as an index
of biomass.
We can measure with the satellite sensors the ratio of red to NIR light- this ratio is called a
Vegetation Index
Mapping “Green”
• http://earthobservatory.nasa.gov/Features/MeasuringVegetation/
• http://wiki.landscapetoolbox.org/doku.php/remote_sensing_methods:net_primary_productivity
• ftp://ftp.biosfera.dea.ufv.br/users/francisca/Franciz/papers/Running%20et%20al.%20Bioscience%202004.pdf
Vegetation Indices
Normalized Difference Vegetation Index- NDVI
• The gigantic chlorophyll “absorption well” distinguishes vegetation from non-vegetation.
• Its size tells us chlorophyll concentration in the leaf and the canopy.
• Many vegetation indices are a simplistic attempt to estimate the size of this absorption well.
VI= NIR/R
Vegetation Indices
wavelength
400 600 800 1000 1200
reflecta
nce(%
)0.0
0.1
0.2
0.3
0.4
0.5
density 1
density 2
density 3
density 4
density 5
density 6
sunlit soil
https://www.youtube.com/watch?v=h3gjnD1QqUU
Vegetation Indices
A pixel by pixel mathematical process
NDVI=(B8-B4)/(B8+B4)
Sentinel 2
The answer is between
-1 and +1.
On land only concerned with
Between 0 and 1.
0.2= No vegetation
0.5= some vegetation
Irish grasslands range between
0.5 and 0.8
Simple Vegation maps with NDVI
Simply set thresholds and colour
Less than 0.2 – Water
Less than 0.45- Natural Vegetation
GT>0.45-Farmland
Time Series
Relief Cloud is always a problem…..
Watching how things change
Over the season is very powerful.
Using more than one observation
Will improve your classification
Phenology!!
UNSUPERVISED CLASSIFICATION
Using statistical clustering- the computer allocates every pixel into a
“class”
Class 1
Forest
Class 3
Agriculture
Class 2
Developed
The computer performs a clustering exercise on the image: The
user tells the computer how many clusters to look for and the
computer then analysis the image to \produce this number of
statistically sound clusters.
Most commonly the ISODATA algorithm is used
Images are Just Numbers
We can also plot these numbers on a graph….
Remember these
images are just arrays
of numbers and we
can do maths on
these numbers!
23
40
55
Plot pixels on a graph
Band 1
Band 2
• Two bands of data.
• Each pixel marks a
location in this 2d spectral
space
• Our eye’s can split the
data into clusters.
• Some points do not fit
clusters.
Rory Hutson
Remote Sensing Group
Plymouth Marine Laboratory
Isodata
Where in the landscape?
Unsupervised Classification
This is simple k-means clustering- all the computer is doing is finding
pixels with similar “colours”
2. Minimum Distance calculations:
Each pixel is associated with
closest mean
.
Band 1
. . . . . . .
.
. .
. . .
.
.
.
.
. . .
. . . . . .
.
. . .
. . . . . .
. . . .
. . . . .
. . . . . . .
.
. . . . . . .
. .
.
. .
. .
.
.
.
. . . .
. . . . .
.
. .
. .
. .
.
. .
.
Cluster Means 1. Means are initialized along diagonal
3. New mean calculated for each
cluster and means migrate to new
locations
1
2
4. Iterations continue until
convergence or maximum iterations
is reached 5. Each cluster associated with a
value. Each pixel given this value
2
• Unsupervised classification- the objective is to group multi-band spectral response patterns into clusters that are statistically separable
• Our example uses 3 bands – More bands can be used, but it can’t be shown in this 3-D plot
• A = Agriculture; D= Desert; M = Mountains; W = Water
Classify Data into Groups Unsupervised classification using
20 different categories was carried
out. Now, the task will be to group
these categories into some kind of
smaller grouping. In our case we
have been using 5 classes:
Agriculture, Developed, Natural,
Forest, and Water.
Obviously, the red is water, we
can see the Lake. Also, the
purple looks like a city, so we
would call that developed.
The rest of the colors are
anyone’s guess. So, the laborious
process of assigning a category to
the different classes (colors) will
now begin.
Assign a name to each group
Simple Map
Developed
Water
Agriculture
Forest
After about 20 minutes, I was able to assign
the classes with four of the categories to
create a final land use map.
But where is Natural? This is sometimes a
problem in digital image processing. Natural
can look like the other classes
And, based on the digital numbers, we were
unable to discriminate the spectral
differences
This is known as spectral confusion. We
may be able to discriminate between the
extractive and developed if we chose more
classes, but even then it might not be
enough. So, that is part of the struggle we
will have as image processors.
MERMS12 L6 NDVI
Image Pixels
Enlarged 10 Times
Thematic Mapper Image
The limit to zooming – the image has been captured at a max
Resolution- you cannot go beyond this- and in fact well before
This resolution is reached its hard for us to recognise objects-
We loose all that other context infromation we use to interpr
Images
MERMS12 L6 NDVI
• This image of the dying Aral Sea was taken by Envisat's Medium Resolution Imaging Spectrometer (MERIS) instrument on July 9. The whitish land surrounding the remaining waters of the evaporating Aral Sea is a salt-covered dry waterbed known as the Aralkum Desert.
MERMS12 L6 NDVI
Enhancements are used to make it easier for visual
interpretation and understanding of imagery.
Remote sensing devices must be designed to cope with
levels of target/background energy which are typical of all
conditions likely to be encountered in routine use.
With large variations in spectral response from a diverse
range of targets (e.g. forest, deserts, snowfields, water, etc.)
no generic radiometric correction could optimally account for
and display the optimum brightness range and contrast for
all targets. Thus, for each application and each image, a
custom adjustment of the range and distribution of
brightness values is usually necessary.
MERMS12 L6 NDVI
In raw imagery, the useful data often populates only a small portion of the available
range of digital values (commonly 8 bits or 256 levels).
Contrast enhancement involves changing the original values so that more of the
available range is used, thereby increasing the contrast between targets and their
backgrounds. The key to understanding contrast enhancements is to understand the
concept of an image histogram.
A histogram is a graphical representation of the brightness values that comprise an
image. The brightness values (i.e. 0-255) are displayed along the x-axis of the graph.
The frequency of occurrence of each of these values in the image is shown on the y-
axis.
MERMS12 L6 NDVI
By manipulating the range of digital values in an image, graphically represented
by its histogram, we can apply various enhancements to the data. The simplest
type of enhancement is a linear contrast stretch. This involves identifying lower
and upper bounds from the histogram (usually the minimum and maximum
brightness values in the image) and applying a transformation to stretch this range
to fill the full range.
MERMS12 L6 NDVI
A uniform distribution of the input range of values across the full range may not
always be an appropriate enhancement, particularly if the input range is not
uniformly distributed. In this case, a histogram-equalized stretch may be better.
This stretch assigns more display values (range) to the frequently occurring
portions of the histogram.