Pr

oje
ctNOAH Ope
n-Fi
leRe
por
ts Vol. 4 (2015), No. x, pp. x-xx, ISSN 2362 7409

Differentiation of debris flow deposits through image classification of Landsat 8 images in

Nueva Ecija after Typhoon KoppuI

Carmille Marie J. Escapea,b,∗

a Nationwide Operational Assessment of Hazards, Department of Science and Technology, Philippines
b National Institute of Geological Sciences, University of the Philippines, Diliman, Quezon City, Philippines

Abstract
As Typhoon Koppu traversed the northeastern provinces of Luzon last October 2015, massive debris flow events were triggered in
the foothills of the Sierra Madre range, causing devastation in the municipalities of Gabaldon, Laur and Bongabon in Nueva Ecija.
These fast moving hazards had carried boulders as large as 6.5 m in diameter and had buried houses and destroyed structures situated
within the alluvial fans where human settlements had thrived. Fortunately, early warnings had been issued and the citizens were
able to act accordingly, thus no casualties were reported. Though it was a disaster that had been averted, this event could happen
again and thus, it is important to study the deposits and their extent in order to gain further data that could be useful in scenario
based hazard modelling. The extent and character of the debris flow deposits were identified from processing pre and post-typhoon
Landsat 8 images. Delineation of the deposit was done through change detection via image differencing and thresholding applied to
NDVI indices and Tasseled Cap Transformation indices. The differentiation between boulder to pebble and finer sediment deposits,
as well as stream delineation was done by unsupervised image classification of the NDVI index of the post-typhoon image and
the 3rd principal component produced from the Principal Component Analysis of the stacked red and NIR bands of the pre and
post-typhoon images. Accuracy assessment based from field data showed that the methodology was able to give a good delineation
between the different types of deposits.
Keywords: debris flow, Koppu, Nueva Ecija, principal component analysis, tasseled cap transformation

1. Introduction reported dead, 83 injured and 4 missing [6]. Because of its

calamitous effect, the Philippine Atmospheric, Geophysical
Last 14 October 2015, Typhoon Koppu (Local name: Lando) and Astronomical Services Administration (PAGASA) had
entered the Philippine Area of Responsibility as a tropical decided to decommision the name [7].
storm, bringing with it a maximum sustained winds of 65 kph
[1]. By 17 October, it had escalated into a Category 4 Typhoon
(based on the Saffir-Simpson hurricane wind scale) with 1.1. Geomorphology of debris flows and alluvial fans in Nueva
maximum sustained winds of 240 kph [2]. On the afternoon Ecija
of the same day, the super typhoon center had made landfall At the foothills of the Sierra Madre mountain range, the
over Casiguran, Aurora where its maximum sustained winds locals within the municipalities of Gabaldon and Laur, Nueva
had reached 185 kph with central pressure of 920 hPa [3][4]. Ecija had reported experiencing flooding that later on brought a
As it hit the Eastern Luzon provinces, it had brought heavy to mixture of sediments, boulders and other detritus. Fortunately,
extreme rainfall that necessitated the raising of Public Storm they were able to immediately evacuate and there were no
Warning Signals 2 - 4 within that region [5]. By the time it recorded casualties [8]. This rapidly moving mass of slurry,
had weakened into a Low Pressure Area as it moved through soil, rock, water, air and other components (e.g. uprooted
Balintang Channel on the afternoon of 21 October, the typhoon trees) is a geological hazard known as a debris flow. They are
had already caused severe flooding, numerous landslides typically the result of the mobilization of poorly consolidated
including a huge debris flow event in Nueva Ecija, and a materials situated at the elevated regions of a drainage area.
slew of casualties. It had caused great devastation in infras- The motility could either be caused by rapid snowmelt runoff
tructure and agriculture in Regions I-V and CAR (Cordillera or torrential rainfall [9].
Administrative Region), equivalent to 10 billion pesos. Over
a 130 thousand houses were damaged and 48 people had been This debris flow event which affected the municipalities of
Gabaldon, Laur and Bongabon in Nueva Ecija, was deposited
I Published online on x month 2015 at
in a geomorphic feature known as an alluvial fan. These
http://blog.noah.dost.gov.ph/PERMALINK/
fan-shaped features are typically present in valleys contiguous
∗ Corresponding author to a mountainous region where a channel emerges. They form
Email address: carms.escape@gmail.com (Carmille Marie J. Escape) by the accumulation of sediments that are deposited due to the
1
rapid loss of carrying capacity that a channel or transporting satellite images.
medium experiences once it stems out from the highlands [10].
When two or more neighboring alluvial fans converge, they
are collectively called as a bajada [11]. Because they occur
in piedmont zones, have channels that serve as water source 2. Methods
and often have long periods of stasis or inactivity that could
Various techniques such as object based image analysis, use
span from decades to centuries, alluvial fans had become ideal
of different indices, change vector analysis and image differ-
location for urban settlement. Unfortunately, they are also
encing, had been employed to interpret landslides from satellite
sites for two very disastrous hazards - alluvial fan flooding
images [18][19]. Since debris flows are also considered as
and debris flow, which often takes people by surprise because
mass wasting events like landslides, these techniques can be
of the seemingly benign appearance of the topography [12][13].
used to interpret them. In fact, numerous studies regarding
interpretation of landslides using optical remote sensing also
Due to the presence of the Philippine Fault that transects the included debris flows [20][21]. However, characterization of
provinces of Aurora and Nueva Ecija in a northwest-southeast the debris flow deposits using remote sensing techniques had
trend, the geomorphology of the affected region was con- not been given much focus. Therefore, the methods (Fig.3)
ducive for the formation of alluvial fans and bajadas. From used in this study were fused together from prior studies that
the mountainous region of Sierra Madre, numerous creeks aimed to characterize landslides [22] [23].
emerged into the flatland traversed and formed by the fault.
The rapid change in elevation resulting to progressive loss of
flow power, availability of sediment source from high relief
and short transport distance, allowed for the formation of al-
luvial fans and bajadas for almost each of the creeks (Fig.1(A)).

1.2. Debris flows in Nueva Ecija

As Typhoon Koppu raged into eastern Luzon, it had induced
numerous landslides in the Sierra Madre range. The loose
deposits of these landslides became the source material for
the debris flow (Fig.1(B))[14]. The torrential rainfall had
also oversaturated the soil with water, allowing it to be easily
mobilized at great velocities. Debris flows occurred at almost
all of the alluvial fans situated in the mentioned municipalities.
According to interviews conducted during a post-disaster field-
work (28-30 October 2015), the debris flow events occurred
between 9 AM to 12 Noon of 18 October. Boulders as large as
6.5 m in diameter were observed and a maximum thickness of
3.5 m of debris flow deposit was measured in Barangay Ligaya,
Gabaldon (Fig.1).

Figure 1 illustrates the extent of the debris flow events.

Although the deposits of debris flows are usually poorly sorted,
the distribution of coarse-grained (boulders to pebbles) sedi-
ments versus that of finer grained sediments in the proximal
and distal part of an alluvial fan can give information about the
mechanism of flow [15] within the event [16][17]. The flow
mechanism can explain the various characteristics of a debris Figure 3: The process flow created for this study.
flow deposit such as its geometry, presence of inverse grading,
imbrication and friable sediment blocks and clumps. It is
therefore important to be able to characterize the debris flow
deposits in order to understand the behaviour of an event, upon 2.1. Data
which the observations gathered could be used as parameter The satellite images used for this study were Landsat 8
for modelling scenario-based debris flow hazards. images with spatial resolution of 30 m for the multispectral
bands and 15 m for the panchromatic band. It has a 16
It was therefore the aim of this study to identify and delineate day revisit capability and can be freely downloaded from
various debris flow deposits through different remote sensing GloVis, EarthExplorer and LandsatLook Viewer. Three
techniques applied in pre and post Typhoon Koppu Landsat 8 images were used; two of them served as pre-typhoon images
2
(acquisition date: 06 September 2015 and 08 October 2015) change detection method was also employed, but its prod-
while the other one served as post-typhoon image (acquisition uct was used for image classification. PCA is a geometric
date: 24 October 2015) (Fig.4). The acquired images are transformation technique that is used for data reduction. A
Level 1T images which are terrain corrected and are in the set of correlated variables are linearly transformed into a set
form of quantized, calibrated and scaled Digital Numbers (DN). of uncorrelated variable (principal components) in a different
orthogonal coordinate system [28] [29]. It compresses the
Upon inspection and review of the satellite images. It had spectral components by identifying components from which
been decided that the focus of the study will be upon the 06 the most variance can be attributed. PCA was applied by
September image which will serve as the pre-typhoon/”passive stacking the images of pre and post red and NIR bands for a
state” image, and the 24 October 2015 image which was taken total of four components. Visual inspection showed that the 3rd
six days after the occurrence of the debris flow events. The 08 component depicted the most variance within the debris flow
October image was still processed as a pre-typhoon image but it deposits. Due to this variance, it was the principal component
cannot be considered as a ”passive state” image because it was (PC) used for image classification.
taken after the occurrence of Typhoon Peipah (Local Name:
Kabayan), which caused mobilization of some materials within For the image differentiation, another data transformation
the study area. was used. The Tasseled Cap Transformation (TCT) [30] is
similar to PCA in such a way that it takes linear combinations
2.2. Data Preprocessing of the bands and summarizes and interprets them into a new
In order to use the images, they were first transformed from set of bands. However unlike PCA, TCT’s first component is
Digital Number to Top of the Atmosphere reflectance using the an index of the overall brightness of the image. The second
equation provided by USGS (U.S. Geological Survey) [24]. component is for the greeness while the third component is
These are the reflectance values with contributions from clouds an index of wetness. The rest of the bands usually contain
and atmospheric aerosols and gases. Atmospheric correction ”noise” and are not as useful. The TCT brightness, wetness
was done by the application of Dark Object Subtraction (DOS) and greeness index were produced for the pre(06 Sept) and
[25] which built upon the assumption that there are pixels post(24 Oct) satellite images. It was done by calculation using
within the image that are under a complete shadow, and that the equation derived by Baig (2014) and was applied to bands
their radiance received by the satellite is due to atmospheric 2-7 (Eq.1) (Fig. 5).
scattering (path radiance) [26]. After the correction, cloud
masking was performed using the Quality Assessment (QA)
Band which has pixels that contains integers that reflect TCT index = (A ∗ band2) + (B ∗ band3) + (C ∗ band4)
(1)
surface, atmosphere and sensor conditions. Using the table +(D ∗ band5) + (E ∗ band6) + (F ∗ band7)
provided by USGS and shown in Fig.1, cloud masking was
performed automatically. For the pixels with uncertainties Where:
regarding presence of clouds, it was necessary to manually TCT index = Index needed (Brightness, Greeness, Wetness)
review whether they are indeed clouds or not. band n = reflectance of the specified band

Index A B C D E F
Pixel value Cloud Pixel value Cloud
Brightness 0.3029 0.2786 0.4733 0.5599 0.508 0.1872
61440 Yes 28672 No Greeness -0.2941 -0.243 -0.5424 0.7276 0.0713 -0.1608
59424 Yes 28590 No Wetness 0.1511 0.1973 0.3283 0.3407 -0.7117 -0.4559
57344 Yes 26656 No
Table 2: Coefficients used for calculation of the TCT indices based on Baig
56320 Yes 24576 No (2014)
53248 Yes 23552 No
39936 Maybe 20516 No
Another index that was produced is the Normalized Dif-
36896 Maybe 20512 No
ferential Vegetation Index (NDVI) which measures the health
36864 Maybe 20480 No
and estimates green biomasses [31]. NDVI is very helpful in
31744 No
distinguishing vegetation from bare earth and had been widely
Table 1: The QA Band interpretation of pixel value in relation to cloud presence
used for landslide detection and change analysis [18]. It is
as provided by the USGS Landsat team [27]. calculated by using the red and NIR bands of the satellite
image (Eq.2).

2.3. Change detection techniques NIR − red

NDV I = (2)
For change detection, image differencing was the primary NIR + red
method that was used. However, the Principal Component In order to show the change between the pre and post
Analysis (PCA) which is a multi-temporal visual composite typhoon images, image differencing was done using the TCT
3
Greeness index, Wetness index and NDVI. The pre and post
were differentiated (Fig. 6) and a threshold was applied in the
resulting image upon the basis that the 0 value represents pixels
with no to least change (Fig.7). The results were then stacked
together to produce a change detection map.

2.4. Image classification and post processing

For image classification, the 3rd PC layer together with the
24 Oct TCT wetness index and NDVI layer were all classified
using the ISO (Interactive Self-Organizing Data Analysis)
cluster unsupervised classification technique. It is an iterative
procedure that assigns arbitrary cluster centers, classifies the
pixels, calculates the cluster mean and covariance until the
minimum change between iterations is achieved [32]. The
classification was done under 10 classes condition and the
results were then sieved and clumped in order to remove
the ’salt and pepper’ effect. The resulting images were then
filtered to only retain the necessary informations, such that the
wetness index was for the stream, the 3rd PC was for both the
boulder to pebble and finer sediments groups and the NDVI
Figure 9: Result of the stacked layers of image difference that shows the extent
was for combined deposits (Fig. 8). These were then stacked of change for each category (TCT Greeness, TCT Wetness and NDVI value).
to produce a classified map of the debris flow deposits.

An accuracy assessment was done by comparing the results done in future works. The specs in the elevated regions could
of the classified map to field data that was gathered from 28-30 be easily removed by masking the elevated regions through
October. The kappa value was also calculated in order to show slope selection (so as not to eliminate streams where the
the extent of how the result correctly represented each classes source materials came from). This could be done so that the
[33]. results would only focus on the flatter areas where the debris
flow deposits are located. However, masking was not done
in this study because the primary goal is to develop an effec-
3. Results and Discussion
tive, semi-automated method with minimal human intervention.
3.1. Change Detection
Some changes weren’t also detected but this could be
Figure 9 shows the result for change detection in a portion of attributed to the threshold that was applied for each image
the study area. The decrease in NDVI value perfectly outlines difference. The threshold used were the most conservative
the debris flow deposits since NDVI is a good tool for demon- thresholds based on visual inspection. However, for future
strating a drastic change from vegetation to bare soil material. work other qualitative methods for designating an unbiased
On the other hand, the decrease in TCT wetness delineated threshold should be considered [34].
the ’passive state’ stream. This decrease was brought by the
replacement of the ”wet” river by the relatively drier deposits.
The decrease in TCT greeness on the other hand, indicates the
3.2. Image Classification
areas which were previously ’greener’ but are now replaced
by materials that are significantly less toned. This decrease is For image classification, only the post-typhoon image was
very prominent especially in the streams at elevated regions used. Figure 10 is the result of the stacking of the unsupervised
and portions of alluvial fans. This was probably because those classification results for TCT wetness, 3rd PC and NDVI, as
areas showed the highest drastic change. The drastic change well as the field sample locations that were used for accuracy
could be attributed to the fact that those areas have the highest assessment. The TCT wetness index was used to delineate the
volume of ’brown’ deposits - since they are closest to the stream after the typhoon. Since the satellite image used was
source materials - that covered the ’green’ pixels (stream banks taken six days after the event, it could be assumed that this
and rice paddies). stream delineation depicts a drier condition and is therefore
a closer delineation of the streams’ ”passive state”, though if
Some ’salt and pepper’ specs are still observable especially compared to the pre-typhoon image, there’s still a very obvious
in the elevated regions and those that are on the fringes of stream widening. Flooding on the lower right side was also
clouds and cloud shadows. Though cloud masking was done, detected. This could be attributed to a flooded sag pond that is
fringes and cloud shadows were still a problem. Refinement the product of the presence of the Philippine Fault.
of cloud masking and detection in cloud shadows should be
4
 Classified Sources Reference Sources Total
Water Boulder - Pebble Finer Seds Combined
Water 3 1 0 0 4
Boulder - Pebble 1 14 0 0
Finer sediments 0 3 5 0 8
Combined 1 0 0 1 2
5 18 5 1 29

Table 3: Accuracy assessment result using the field data gathered 10 days after
the typhoon.

User’s Accuracy Producer’s Accuracy

Water 75.00% 60.00%
Boulder - Pebble 93.33% 77.78%
Finer Seds 62.50% 100.00 %
Combined 50.00% 100.00%

Table 4: User and producer accuracy result

The accuracy assessment (Tab.3) was calculated using field

data that were gathered by a team of geologists 10 days after the
typhoon. The time difference between the image aquisition and
Figure 10: Result of the stacked layers of unsupervised classification of TCT the field data acquisition may have a small effect but based on
wetness index, 3rd PC and NDVI.
the calculation of the accuracy assessment and kappa value, the
results seem to be favorable and valid. The highest user accu-
Since the 3rd PC component showed the most variation racy was for the boulder to pebble sediments while the highest
within the debris flow deposits, it was used to differentiate producer accuracy as achieved for both the finer sediments and
between boulder to pebble deposits and finer sediments. the combined deposits (Tab.4). The kappa statistics shows the
The boulder to pebble deposits are very well delineated in relevance of the method, grading it 65% better than random or
the alluvial fan regions. This was expected because coarser chance.
sediments brought by a debris flow event are usually deposited
in the proximal region of an alluvial fan. However, there are 4. Conclusions
coarse deposits that could be observed in the upper left side
of the image on plain region. This result could be interpreted Overall, the method proposed here appears to be useful
as a representation of the smaller coarse deposits (cobble to for the characterization of debris flow deposits even though
pebble deposits) which could be carried further away from the the method was devised in such a way that it will be semi-
alluvial fan. Future efforts could include further refinement of automatic and involves minimal intervention. NDVI was very
distinguishing between larger and smaller coarse deposits. useful for the delineation of the rapid change in the satellite
image and it was therefore useful for delineating the overall
Meanwhile, the NDVI result is very dominant in the flatter debris flow deposit outline. The TCT wetness and greeness
regions of the study area. They could be interpreted as combi- indices were very useful for change detection especially
nation of different deposit types including other detritus such concerning the vegetated-to-flooded areas. The 3rd PC compo-
as uprooted trees and even murky or still wet deposits. Since nent was useful for distinguishing between debris flow deposits.
NDVI merely marks a drastic change from vegetation, it is safe
to assume that the change was brought by a wide variety of There is however still a problem regarding cloud shadows
materials. However, it is very apparent in the elevated regions and cloud masking. Also, errors were also observed in the
that NDVI was able to detect the landslides that deposited the elevated regions. The problem with cloud mask and shadows
source material for the debris flows. could be resolved in future work by incorporating more rigor-
ous techniques for masking and identification under shadowed
Possible refinement of the result could be done by again con- areas. Errors in the elevated regions could be removed by
sidering only the flatter regions though slope masking. How- masking through slope selection. This way, the results are
ever it was not done because of the same reason that was ex- limited on only the flatter areas where deposition occurred.
pressed above.
Future work should involve usage of more quantified thresh-
3.3. Accuracy Assessment olds as well as other types of classification methods including
supervised techniques. There should also be future endeavors to
0.79 − 0.39 apply the proposed methods on other known debris flow events
Kappastatistics = = 0.65 (3) to further verify its viability.
1 − 0.39
5
5. Acknowledgment [20] G. Metternicht, L. Hurni, R. Gogu, Remote sensing of landslides: An
analysis of the potential contribution to geo-spatial systems for hazard as-
sessment in mountainous environments, Remote sensing of Environment
The author would like to thank the Nueva Ecija debris flow 98 (2) (2005) 284–303.
team for allowing her to use their field data and pictures. [21] T. R. Martha, N. Kerle, V. Jetten, C. J. van Westen, K. V. Kumar, Charac-
terising spectral, spatial and morphometric properties of landslides for
semi-automatic detection using object-oriented methods, Geomorphol-
ogy 116 (1) (2010) 24–36.
References [22] A. Mondini, F. Guzzetti, P. Reichenbach, M. Rossi, M. Cardinali,
F. Ardizzone, Semi-automatic recognition and mapping of rainfall in-
duced shallow landslides using optical satellite images, Remote Sensing
[1] Philippine Atmospheric, Geophysical and Astronomical Services Ad- of Environment 115 (7) (2011) 1743–1757.
ministration, Severe weather bulletin #1 tropical cyclone alert: Tropical [23] K. Cheng, C. Wei, S. Chang, Locating landslides using multi-temporal
storm ”lando”, Website (October 2015). satellite images, Advances in Space Research 33 (3) (2004) 296–301.
URL http://www.pagasa.dost.gov.ph/index.php/tropical- [24] USGS, Using the usgs landsat 8 product, Website (August 2015).
cyclones/weather-bulletin/165-tropical-cyclones/ URL http://landsat.usgs.gov/Landsat8_Using_Product.php
severe-weather-bulletin/lando-2015-bulletin/1800-lando [25] C. Song, C. E. Woodcock, K. C. Seto, M. P. Lenney, S. A. Macomber,
[2] Joint Typhoon Warning Center, Prognostic reasoning for super typhoon Classification and change detection using landsat tm data: when and how
24w (koppu) warning nr 19”, Website (October 2015). to correct atmospheric effects?, Remote sensing of Environment 75 (2)
URL http://weather.noaa.gov/pub/data/raw/wd/wdpn31. (2001) 230–244.
pgtw..txt [26] P. S. Chavez, Image-based atmospheric corrections-revisited and im-
[3] National Disaster Risk Reduction and Management Council, Ndrrmc up- proved, Photogrammetric engineering and remote sensing 62 (9) (1996)
date severe weather bulletin no. 12 re typhoon ”lando”, Tech. rep., Na- 1025–1035.
tional Disaster Risk Reduction and Management Council (2015). [27] USGS, Landsat 8 quality assessment band, Website (2014 December).
[4] Japan Meteorological Agency, Rsmc tropical cyclone advisory 171800, URL http://landsat.usgs.gov/L8QualityAssessmentBand.
Website (October 2015). php
URL http://weather.noaa.gov/pub/data/raw/wt/wtpq20. [28] J. Richards, Thematic mapping from multitemporal image data using the
rjtd..txt principal components transformation, Remote Sensing of Environment
[5] National Disaster Risk Reduction and Management Council, Ndrrmc up- 16 (1) (1984) 35–46.
date severe weather bulletin no. 13 re typhoon ”lando”, Tech. rep., Na- [29] I. Jolliffe, Principal component analysis, Wiley Online Library, 2002.
tional Disaster Risk Reduction and Management Council (2015). [30] R. J. Kauth, G. Thomas, The tasselled cap–a graphic description of the
[6] National Disaster Risk Reduction and Management Council, Ndrrmc spectral-temporal development of agricultural crops as seen by landsat,
update sitrep no. 22 re preparedness measures and effects of typhoon in: LARS Symposia, 1976, p. 159.
”lando”” (i.n. koppu), Tech. rep., National Disaster Risk Reduction and [31] D. J. Hayes, S. A. Sader, Change detection techniques for monitoring
Management Council (2015). forest clearing and regrowth in a tropical moist forest, Photogrammetric
[7] Philippine News Agency, Pagasa plans to decommission ’lando’ from list Engineering and Remote Sensing 67 (9) (2001) 1067–1075.
of typhoon names, Online News Article (October 2015). [32] A. Ahmad, S. F. Sufahani, Analysis of landsat 5 tm data of malaysian land
URL http://www.mb.com.ph/pagasa-plans-to-decommission- covers using isodata clustering technique, in: Applied Electromagnetics
lando-from-list-of-typhoon-names/ (APACE), 2012 IEEE Asia-Pacific Conference on, IEEE, 2012, pp. 92–
[8] ABS-CBN News, ’gabaldon a disaster that did not happen’, scientist 97.
says, Online news article (October 2015). [33] M. L. McHugh, Interrater reliability: the kappa statistic, Biochemia Med-
URL http://rp1.abs-cbnnews.com/focus/10/22/15/ ica 22 (3) (2012) 276–282.
gabaldon-disaster-did-not-happen-scientist-says [34] P. A. Rogerson, Change detection thresholds for remotely sensed images,
[9] D. J. Varnes, Landslide types and processes, Highway Research Board Journal of Geographical Systems 4 (1) (2002) 85–97.
Special Report (29).
[10] T. C. Blair, J. G. McPherson, Alluvial fan processes and forms, in: Geo-
morphology of desert environments, Springer, 1994, pp. 354–402.
[11] R. L. Bates, J. A. Jackson, et al., Dictionary of geological terms, Vol. 584,
Anchor Books, 1984.
[12] N. Santangelo, A. Santo, G. Di Crescenzo, G. Foscari, V. Liuzza, S. Scia-
rrotta, V. Scorpio, Flood susceptibility assessment in a highly urbanized
alluvial fan: the case study of sala consilina (southern italy), Nat. Hazards
Earth Syst. Sci 11 (2011) 2765–2780.
[13] J. F. Hubert, A. J. Filipov, Debris-flow deposits in alluvial fans on the west
flank of the white mountains, owens valley, california, usa, Sedimentary
Geology 61 (3) (1989) 177–205.
[14] T. Takahashi, Mechanical characteristics of debris flow, Journal of the
Hydraulics Division 104 (8) (1978) 1153–1169.
[15] R. M. Iverson, The physics of debris flows, Reviews of geophysics 35 (3)
(1997) 245–296.
[16] Y. K. Sohn, C. W. Rhee, B. C. Kim, Debris flow and hyperconcentrated
flood-flow deposits in an alluvial fan, northwestern part of the cretaceous
yongdong basin, central korea, The Journal of Geology 107 (1) (1999)
111–132.
[17] E. Blissenbach, Geology of alluvial fans in semiarid regions, Geological
Society of America Bulletin 65 (2) (1954) 175–190.
[18] J. Nichol, M. Wong, Satellite remote sensing for detailed landslide inven-
tories using change detection and image fusion, International Journal of
Remote Sensing 26 (9) (2005) 1913–1926.
[19] D. N. Petley, W. D. Crick, A. Hart, The use of satellite imagery in land-
slide studies in high mountain areas, in: Proceedings of the Asian confer-
ence on remote sensing, 2002.

6
Figure 1: Landsat 8 image of the study area showing the debris flow that devastated the municipalities of Gabaldon, Laur and Bongabon in Nueva Ecija. (A)
Pre-Lando satellite image (06 September 2015) showing the transecting Philippine Fault and the alluvial fans and bajadas that formed in the flatland. (B)Post-Lando
satellite image (24 October 2015) showing the extent of the debris flows that occurred in each alluvial fan and bajada

Figure 2: Images depicting the devastation caused by the debris flows that occurred during Typhoon Koppu. (A) A house buried halfway by debris flow deposits.
(B) Thick debris flow deposits were measured within various locations. (C) Debris flows are very strong that they can displace concrete bridges. (D) Debris flow
deposits at the side of a cleared road (human for scale). Photos courtesy of the Nueva Ecija debris flow team.

7
Figure 4: Landsat 8 satellite images that were inspected and used for the study. Red box indicates the study area.

Figure 5: Tasseled Cap Transformation indices for the 24 Oct Landsat 8 image

8
 Figure 6: Sample of image differencing using TCT wetness and NDVI.

Figure 7: Application of threshold allowed for the distinguishing of changes for TCT greeness, wetness and NDVI.

9
Figure 8: ISO cluster unsupervised classification for the 24 Oct Landsat 8 image.

10