Impacts of spatial resolution
on land cover classification
Chanida Suwanprasit and Naiyana Srichai
Prince of Songkla University Phuket Campus
APAN 33rd Meeting 13-17 February 2012
2/20
Outline





Introduction
Objective
Methodology
Results
Conclusions
3/20
Spatial Resolution
is a measurement of the spatial detail in an image, which is a
function of the design of the sensor and its operating altitude
above the Earth’s surface (Smith, 2012).
Classification Factors





Number of mixed Pixel
Number of ROIs
Scale or spatial resolution
Spectral resolution
Temporal resolution
5/20
Objective
 To examine effects of pixel size on
land use classification in Kathu district,
Phuket, Thailand
Study area: Kathu, Phuket
Kamala
Kathu
Patong
7/20
6/20
Data set specification
Imagery Source
LANDSAT 5 TM
Resolution (m)
Band
Spectral Type
30
1 (Blue)
0.45 – 0.52 m
30
2 (Green)
0.52 – 0.60 m
30
3(Red)
0.63 – 0.69 m
30
4 (NIR)
0.78 – 0.90 m
30
5 (NIR)
1.55 – 1.75 m
60
6 (TIR)
10.40 – 12.5 m
30
7(MIR)
2.80 – 2.35 m
15
1 (Blue)
0.45 -0.52 m
15
2 (Green)
0.53 – 0.60 m
15
3 (Red)
0.62 – 0.69 m
15
4 (NIR)
0.77 – 0.90 m
THEOS
Landsat 5 Spectral Bands
Band 1 (Blue)
Band 4 (NIR)
10/20
Band 2 (Green)
Band 3 (Red)
Band 5 (NIR)
Band 7 (MIR)
THEOS Spectral Bands
Band 1 (Red)
Band 2 (Green)
Band 3 (Blue)
Band 4 (NIR)
11/20
9/20
True Color
THEOS
Landsat 5
RGB (4,3,2)
THEOS
13/20
Landsat 5
Process Overview
12/20
Data Set
THEOS
Control points
Training
area
Landsat 5
Unsupervised
K-Mean
Supervised
SVMs
Test area
Land use Classification Map
THEOS
LandSat 5
Classes
• Forest
• Built-up
• Road
• Water
• Agriculture
• Grassland
• Bare land
Unsupervised Classification:
K-Mean (7 Classes)
THEOS
14/20
Landsat 5
Support Vector Machines : SVMs
THEOS
Landsat
Forest
Bare land
Built - up
Grassland
Water
Road
16/20
Class Confusion Matrix
THEOS
Class
17/20
Landsat-5
Prod. Acc.
(%)
User Acc.
(%)
Prod. Acc.
(%)
User Acc.
(%)
Forest
97.47
96.81
100.00
100.00
Built-up
62.37
71.18
97.02
97.57
Road
74.89
64.62
90.15
90.59
Water
99.87
99.29
83.25
78.71
Bare land
76.78
91.31
60.88
66.78
Grassland
89.49
95.23
96.02
91.85
Agriculture
92.21
84.22
76.69
75.37
Overall Accuracy
90.65% (Kappa Co.= 0.88)
89.00% (Kappa Co.=0.87)
Conclusion
 THEOS gave a higher classification accuracy than Landsat 5
for identifying land use in this study.
 More Spectral bands from Landsat 5 with 30m is not appropriated for
selecting clearly ROIs than THEOS with 15m resolution.
 The better resolution image greatly reduce the mixed-pixel problem,
and there is the potential to extract much more detailed information on
land-use/land cover structures.
18/20
References
 Duveiller, G. and P. Defourny (2010). "A conceptual
framework to define the spatial resolution requirements for
agricultural monitoring using remote sensing." Remote
Sensing of Environment 114(11): 2637-2650.
 Randall B. Smith (2012). "Introduction to Remote Sensing
Environment (RSE)". Website: http://www.microimages.com.
19/20
20/20
Acknowledgement
 Faculty of Technology and Environment
Prince of Songkla University, Phuket Campus
 Geo-Informatics and Space Technology Development
Agency (Public Organization)
 UniNet
Thank you for your kind attention