Upload
others
View
21
Download
0
Embed Size (px)
Citation preview
GEOENGINE - MSc in Geomatics Engineering
1
Master Thesis
Neda Mohammadi Naghadeh
“Large-scale characterization of human settlement patterns using binary settlement masks
derived from globally available TerraSAR-X data”
Duration of the Thesis: 6 months
Completion: January 2014
Supervisor: Dipl.-Ing. René Pasternak
Examiner: Prof. Dr.-Ing. Alfred Kleusberg
ABSTRACT: The goal of master thesis is to demonstrate a method which facilitates quantitative and qualitative
analysis of human settlements patterns on a national and global level. As at the beginning of 21th
century the number and bigness of urban areas are going to increase, so it is vital challenge to
distinguish where the urban patterns are going to develop, because urban land use change affects
environmental changes. According to these facts, this study will utilize of Geographical
Information Systems (GIS) and Landscape Metrics in order to find a way for this purpose and also
to try enhance them. Landscape metrics enable to quantify a landscape (here: urban area or in
general human settlements) with respect to spatial dimension, alignment and pattern at a specific
scale and resolution.
Introduction: Remote sensing techniques can be applied to the analysis of human and environmental dynamics
within urban systems to aid in sustainable planning and management of these areas.
Remote sensing has great potential for gaining comprehensive and accurate land-use information
for analysis and planning of settlement areas. This research involves a methodology using
information from Terra-SAR data to describe settlement patterns.
The mission of thesis follows Remote sensing and Geospatial tools in order to classify large scale
settlement patterns in Global area, therefore in this study we use GUF (Global Urban footprint)
data, landscape metrics and Geospatial analysis to quantify and analysis the building density in
two testsites. Result will be discussed in relation to the type of settlements distribution and be
displayed in classified maps.
GEOENGINE - MSc in Geomatics Engineering
2
Figure 1: The procedure of task
Study area and land cover data set: As study areas, two test sites were chosen, “Munich” which contains the city of Munich and its
suburbs, and “Emden” which comprises the cities of Emden, Oldenburg, Wilhelmshaven, Aurich
and Groningen (located in the Netherland) and also settlements around these cities.
As it is clear from the figures below (Figure (2) & Figure (3)), the settlement patterns in the
“Munich” and “Emden” have noticeable differences in terms of heterogeneity of populated areas.
In Munich urban growth is concentrated in center, then settlements distribute as radial in suburb,
while in Emden these dense patterns propagate in whole area in more homological situation, as the
area of dense settlement patterns in Emden are not as large as agglomerated urban patterns in
Munich.
TerraSAR-X data
Python and ArcGIS for
landscape partition
FRAGSTATS for
spatial pattern analysis
Analysis of data and
classified map
GEOENGINE - MSc in Geomatics Engineering
3
9999999
Source: worldatlasbook.com
Figure 2: Location of test site “Munich” in state of Bavaria and GUF (Global Urban Footprint
layer) data with the resolution of 20m
GEOENGINE - MSc in Geomatics Engineering
4
Figure 3: Location of test site “Emden” in state of Lower Saxony and GUF (Global Urban
Footprint layer) data with the resolution of 20m
Source: worldatlasbook.com
GEOENGINE - MSc in Geomatics Engineering
5
Methodology:
1. Landscape Partition The methodology of this study is landscape partition, as landscape was subdivided in several sub-
landscapes in order to handle data more efficiently. As type of settlements differ entirely in the
limited area. Therefore each test site was converted into square tiles to square tiles which are equal
in terms of size and area in “Munich” and “Emden”. The landscape partition was done by ArcGIS
and Python script programming language, it was started by making clip raster by the size of 900
pixels or 18000 meter (pixel size: 20m). In the following of this process, by tiles with 9000, 4500
and 2250 meters in row and column, however because of this way some parts of test sites must be
left and will not be used for final results and analysis.
2. Choosing Geospatial Software for Analysis
A wide range of geospatial software is available commercially, as freeware and as open source.
Here we chose freely available software for this investigation, which is free in license and useful
for landscape metrics and spatial analyst goals, additionally it is suitable for the data type in this
study (raster data).
FRAGSTATS is a computer software program designed to compute a wide variety of landscape
metrics for categorical map patterns. Also in the recent years it was developed for variety types of
Geographical data. FRAGSTATS is a public-domain GIS implementation of a set of spatial
statistics that addresses a fundamental problem in GIS applications and it is a spatial pattern
analysis program for categorical maps.
FRAGSTATS computes three groups of metrics: Patch metrics, Class metrics and Landscape
metrics. Patch metrics are computed for every patch (definition of patch: Surface area that differs
from its surroundings in nature or appearance) in the landscape; the output file contains a row
(observation vector) for every patch (defined as environmental units), where the columns (fields)
represent the single metrics.
Class metrics measure the aggregate properties of the patches belonging to a single class or patch
type. Landscape metrics are computed for entire patch mosaic; the resulting landscape output file
contains a single row (observation vector) for the landscape, where the columns (fields) represent
the individual metrics.
GEOENGINE - MSc in Geomatics Engineering
6
3. Analysis of Landscape Patterns with FRAGSTATS metrics In urban remote sensing or in other words analysis of settlement patterns, a few special metrics
of FRAGSTATS will be used.
For Patch metrics, metrics such as Patch Area (AREA), Perimeter (PERIM), Radius of Gyration
(GYRATE) could be suitable for this study, as Radius of Gyration (GYRATE) is a measure of
patch extent, so this metrics displays the bigness of patches. It is clear that every landscape includes
of several patches which the largest one is the sign of the most dense area in this task.
In Class metrics, metrics like Total (Class) Area (CA), Percentage of Landscape (PLAND),
Largest Patch Index (LPI), Total Edge (TE), Edge Density (ED), Number of Patches (NP), Patch
Density (PD), Landscape Shape Index (LSI) and Euclidean Nearest-Neighbor Distance (ENN) are
practical for analysis of residential area in this task. Percentage of Landscape (PLAND) and Class
Area (CA) give information about the area of settlements. Number of Patches (NP) and Patch
Density (PD) focus on the subdivision of aggregation, so NP and PD are considerable for number
and density of settlements, whereas the sizes of patches (area of settlements) do not have equal
bigness, they are not so practical for settlement pattern analysis through the large areas (not in
equal tiles), but approximately an idea could be given in a limited area. The Largest Patch Index
(LPI) gives information about the type and existence of a spatially dominant urban core.
FRAGSTATS computes several statistics representing the amount of perimeter (or edge) at the
patch, class, and landscape levels. Edge metrics usually are best considered as representing
landscape configuration. At the patch level, edge is a function of patch perimeter (PERIM). At the
class and landscape levels, total edge (TE) is an absolute measure of total edge length of a
particular patch type or of all patch types. The edge density determines landscape configuration,
with large values displaying a more organic, convoluted urban pattern. The nearest neighbor
standard deviation valuates uniform or regular distribution of patches against a more irregular or
uneven distribution in a landscape. Clearly low values of ENN reveals the dense settlement
patterns, however FRAGSTATS metrics for ENN calculates the average amount, such as ENN-
MN or ENN-SD.
In the table below (Table (1)) some special metrics of FRAGSTATS which were used in this task
are explained by formulas.
GEOENGINE - MSc in Geomatics Engineering
7
Table 1. Spatial metrics used in this study
Subject Metric Formula Units Range
Patch metrics
Patch Area (AREA) 𝑨𝑹𝑬𝑨 = 𝒂𝒊𝒋 (𝟏
𝟏𝟎, 𝟎𝟎𝟎) hectares
Area > 0,
without limit
Perimeter (PERIM) 𝑷𝑬𝑹𝑰𝑴 = 𝑷𝒊𝒋 meters PERIM > 0,
without limit
Radius of gyration
(GYRATE) 𝐆𝐘𝐑𝐀𝐓𝐄 = ∑
𝐡𝐢𝐣𝐫
𝐳
𝐳
𝐫=𝟏
meters GYRATE ≥ 0
Class metrics
Class Area (CA) 𝑪𝑨 = ∑ 𝒂𝒊𝒋
𝒏
𝒋=𝟏
(𝟏
𝟏𝟎, 𝟎𝟎𝟎) hectares
CA > 0,
without limit
Percentage of
Landscape
(PLAND)
𝑷𝑳𝑨𝑵𝑫 = 𝑷𝒊 =∑ 𝒂𝒊𝒋
𝒏𝒋=𝟏
𝑨(𝟏𝟎𝟎) percent 0<PLAND≤100
Largest Patch Index
(LPI) 𝑳𝑷𝑰 =
𝒎𝒂𝒙𝒋=𝟏𝒏 (𝒂𝒊𝒋)
𝑨(𝟏𝟎𝟎) percent 0<LPI≤100
Total Edge (TE) 𝑻𝑬 = ∑ 𝒆𝒊𝒌
𝒎
𝒌=𝟏
meters TE > 0,
without limit
Edge Density (ED) 𝑬𝑫 =
∑ 𝒆𝒊𝒌𝒎𝒌=𝟏
𝑨(𝟏𝟎, 𝟎𝟎𝟎)
meters
per
hectare
ED > 0,
without limit
Number of Patches
(NP) NP=n_i none NP > 1,
without limit
Patch Density (PD) 𝑷𝑫 =𝒏𝒊
𝑨(𝟏𝟎, 𝟎𝟎𝟎)(𝟏𝟎𝟎)
number
per 100
hectares
PD > 0,
constrained by cell
size
Landscape Shape
Index (LSI) 𝐋𝐒𝐈 =
. 𝟐𝟓 ∑ 𝐞𝐢𝐤∗𝐦
𝐤=𝟏
√𝐀 none LSI > 1,
without limit
Euclidean Nearest-
Neighbor Distance
(ENN)
𝑬𝑵𝑵 = 𝒉𝒊𝒋 meters ENN> 0,
without limit
GEOENGINE - MSc in Geomatics Engineering
8
4. Final Result based on interpolation
As it was explained before, in our study we divided the landscape to equal tiles in four steps.
Therefore this master thesis must demonstrate that, partition will describe urbanization and
structure of settlement patterns in an effective way.
The figure (figure (4)) below reveals the interpolation for different size of tiles and how these
partitions increase the accuracy of task. As it is clear, in the final step of partition with smaller
tiles, we face with better results.
In addition based on the goal of the classification in terms of settlement types, the suitable metrics
could be chosen. Here because of type of data and classification which reveals density, was
PLAND (percentage of landscape), clearly according to the purpose of study could be applied on
other metrics. However the same result could be reached with LPI or NP/CA almost, as larger
values for LPI reveal the dense area and vice versa less values of NP/CA display the dense
settlements. Classification is done for three types of settlement patterns and contains the
adjustment based on urban systems and structures. Nevertheless if interpolation could be applied
for Global data, it would be better to use unique classification ranges.
5. Conclusion and future study
Overall, Remote Sensing and GIS are useful tools for analysis of large scale settlement patterns,
as the capability for landscape classification is one of the most important applications of Remote
Sensing.
Figure below (figure (4)) reveals how the methodology of partition could be useful to reach thigh
accuracy with smaller tiles (not so tiny based on the resolution) and more steps of division. Another
thing for high accuracy is the structural pattern of the settlements in terms of distribution,
agglomeration and shape of build-up lands, which affects the classification of settlement patterns.
As a future study for this task, it is useful to implement and develop a Cellular Automata (CA)
Algorithm for urban patterns. Another option for future research could be the comparison of
different resolutions of Terra-SAR data to reach classification with better accuracy.
Test site: “Munich”
GEOENGINE - MSc in Geomatics Engineering
9
Figure 4: Classification of settlement types in “Munich“
REFERENCES
GEOENGINE - MSc in Geomatics Engineering
10
McGarigal, K., SA Cushman, and E Ene., (2012), FRAGSTATS v4:
Spatial Pattern Analysis Program for Categorical and Continuous Maps.
Computer software program produced by the authors at the University of
Massachusetts, Amherst. Available at the following web site:
http://www.umass.edu/landeco/research/fragstats/fragstats.html. (last access: 11/10/2013).
McGarigal, K., (2012), Fragstats .help.4.pdf. Available at the following
web site:
http://www.unibuc.ro/prof/patru stupariu_i_g/docs/2013/noi/25_15_29_40fragstats.help.4.pdf
(last access: 13/09/ 2013).
Esch, T., Taubenböck, H., Roth, A., Heldens, W., Felbier, A., Thiel, M., Schmidt M, Müller, A.,
Dech, S., (2012), TanDEM-X mission—new perspectives for the inventory and monitoring of
global settlement patterns. In: Journal of Applied Remote Sensing. 6(1): 1-21, doi:
10.1117/1.JRS.6.061702.
Esch, T., (2010), Delineation of urban footprints from TerraSAR-X data by analyzing speckle
characteristics and intensity information. In: IEEE TRANSACTIONS ON GEOSCIENCE AND
REMOTE SENSIN. 48(2): 905–916.
Taubenböck, H., Wegman, M., Wurm, M., Ullmann,T., Dech, S., (2010),
The global trend of urbanization-Spatiotemporal analysis of mega cities
using multi-temporal remote sensing, landscape metrics and gradient
analysis. In: Proceeding of the conference SPIE Europe Remote Sensing, Earth Resources and
Environmental Remote Sensing/GIS Applications, Toulouse, France, 20. September, edited by
Michel, U., Civco, D.L., 7831:78310I 1-20, doi: 10.1117/12.864917.