Maximum likelihood classification of satellite imagery achieved over 90% accuracy in detecting landslide scars from a 2010 event in Madeira, Portugal. Several classification methods were tested, and maximum likelihood performed best with a 93.76% overall accuracy. Post-classification tools like sieving and clumping further improved results. Manual delineation using high-resolution orthophotographs validated and corrected the scar mapping. This automated process produced an extensive, accurate landslide inventory more efficiently than traditional manual methods.
1. A COMPARATIVE ASSESSMENT OF SUPERVISED PIXEL–BASED CLASSIFICATION METHODS IN THE DETECTION OF LANDSLIDE SCARS
M. LOUSADA1, C. LIRA1, P. PINA1, A. GONÇALVES2, A. P. FALCÃO2, S. HELENO2, M. MATIAS2, A. J. DE SOUSA1, M. J. PEREIRA1, R. OLIVEIRA3 AND A. B. ALMEIDA3
1CERENA, 2ICIST, 3CEHiDRO
IST/UTL - INSTITUTO SUPERIOR TÉCNICO, LISBOA, PORTUGAL
INTRODUCTION
On February 20th, 2010 heavy rainfall culminated in violent floods and mudslides in Madeira Island, Portugal. This extreme event triggered thousands of landslides in
both inhabited and uninhabited zones, resulting in extensive personal and material damages. Two main areas were heavily affected: Funchal and Ribeira Brava. Aiming
to estimate the volume of sediment displaced during the event, a landslide inventory was an urgent necessity. This paper focuses on the procedures used in the
cartographic inventory of the landslides, particularly in the assessment tests for the most accurate automatic classification procedure and the applied post-processing
methods. GeoEye-1 satellite imagery from February 23rd and 28th, 2010, was the basis for our classification procedures.
METHODS
A) Several training areas or regions of interest (ROIs) were selected, LANDSLIDE SCAR
different sets of ROIs were tested and the best were used to LANDSLIDE TRACK
GRASS
create a total of 11 classes . Results indicated that the Maximum
BARE SOIL
Likelihood algorithm presents the best accuracy and quality on CLOUDS
landslide scar contours. GRAVEL
FOREST
B) Computation of the confusion matrices allowed the comparison ROOFS
INDUSTRY
of results and the evaluation of the accuracy of landslide scar
SHADOW
classification with a ground truth image built from ROIs. ROADS
Landslide Scars, polygons manually
corrected/validated
Maximum Likelihood GeoEye image Mahalanobis Distance Minimum Distance GeoEye image Parallelepiped
Ortophotomaps with 0.4m spatial
(pan-sharpening 0.5 m/pixel) (pan-sharpening 0.5 m/pixel)
resolution
A) CLASSIFICATIONS IN B) CONFUSION MATRICES D) MANUAL DELINEATION
C) POST-CLASSIFICATION OF
GEOEYE -1 IMAGERY AND VALIDATION OF
Landslide Scars, polygons obtained from WITH GROUND TRUTH MAXIMUM LIKELIHOOD
Max-Like classification
(R-G-B-NIR band, 2 m/pixel and IMAGE BUILT FROM ROIS SIEVING AND CLUMPING
LANDSLIDE SCARS WITH
C) In a post-classification stage tools as
panchromatic 0.5 m/pixel
resolution)
ORTOPHOTOMAPS sieve, clump and majority analysis
Maximum Likelihood
classification were used in the Maximum
Likelihood classification and tested
Clump Sieve with different thresholds to
GeoEye image ( 0.5m/pixel)
suppress or clump isolated or small
groups of pixels, improving widely
the quality of the final landslide scar
D) To correct contours and obtain the eventually missing scars , manual layout.
delineation and corrections were edited, with high resolution
ortophotomaps (1:5000 scale with 0.4 m ) from may, 2010.
Ground Truth (Percent)
RESULTS CONCLUSION
Class Lnds. scar Lnds. track Grass Bare soil Clouds Gravel Forest Roofs Industry Shadow Roads Total
Maximum Likelihood Mahalanobis Distance Commission Omission Commission Omission User Acc. Prod. Acc. Prod. Accuracy User Accuracy
The final results are very satisfactory, as the
Class (Percent) (Percent) (Pixels) (Pixels) (Percent) (Percent) (Pixels) (Pixels)
Overall Accuracy = (969054/1033569) 93.76% Overall Accuracy = (836836/1033569) 80.97%
Landslide scar
Landslide track
92,58
3,26
3,81
95,91
0
0
0,91
0,01
0
0
0
0
0,06
0
1,03
0,67
0,07
0
0
0
0,12
0
0,6
0,71 Kappa Coefficient = 0.915 Kappa Coefficient = 0.749
Landslide scar
Landslide track
11,17
3,31
7,42 693/6206
4,09 244/7366
442/5955
304/7426
92,58
95,91
88,83 5513/5955
96,69 7122/7426
5513/6206
7122/7366
methodology produced an overall accuracy over 90%
Grass 0,52 0,05 99,82 0 0 0,01 0,35 0 0 0 0 0,89
Bare soil 2,4 0,09 0 99,08 0 0 0,02 0 0 0 0 0,84
Grass
Bare soil
13,59
2,36
0,18 1246/9168
0,92 204/8634
14/7936
78/8508
99,82
99,08
86,41 7922/7936
97,64 8430/8508
7922/9168
8430/8634 in the detection of landslides for the study area. This
Clouds 0,13 0,07 0 0 99,01 0,07 0,04 0 0,19 0 0,32 26,32
CLASSIFICATION
Clouds 0,1 0,99 283/272044 2709/274470 99,01 99,9 271761/274470 271761/272044
Gravel
Forest
0,12
0,3
0,03
0 0,18
0 0
0
0
0
58,01
0,47
0,15
93,39
2,22
0,01
15,77
0,1
0,48
2,29
7,22
0,11
3,81
31,8
Gravel 18,76 41,99 7390/39395 23164/55169 58,01 81,24 32005/55169 32005/39395 enabled an extensive inventory of the scars, with a
Roofs
Industry
0,69
0
0,04
0
0
0
0
0 0,99
0 0,01
3,35
0
0
93,23
1,4
0,31
80,41
0
0
0,08
1,14
0,66
2,82
Parallelepiped Minimum Distance Forest
Roofs
2,08
2,15
6,61 6841/328725
6,77 146/6782
22797/344681
482/7118
93,39
93,23
97,92 321884/344681 321884/328725
97,85 6636/7118 6636/6782 substantially less time-consuming and less expensive
Industry 16,25 19,59 4745/29192 5955/30402 80,41 83,75 24447/30402 24447/29192
Shadow
Roads
0
0
0
0
0
0
0
0
0
0
2,27
35,8
5,99
0
0
1,43
0,03
3,13
97,22
0
0
91,01
28,87
2,68
Overall Accuracy = (336155/1033569) 32.52%
Kappa Coefficient = 0.243
Overall Accuracy = (831048/1033569) 80.41%
Kappa Coefficient = 0.742 Shadow 7,34 2,78 21917/298406 7894/284383 97,22 92,66 276489/284383 276489/298406
process than the traditional manual delimitation
Total 100 100 100 100 100 100 100 100 100 100 100 100
Roads 75,25 8,99 20806/27651 676/7521 91,01 24,75 6845/7521 6845/27651 methods.