Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Philippine Journal of Science

150 (4): 723-734, August 2021


ISSN 0031 - 7683
Date Received: 12 Oct 2020

Monitoring of Land Use and Land Cover Changes


in Mt. Pulag National Park Using Landsat
and Sentinel Imageries

Nova D. Doyog1, Roscinto Ian C. Lumbres1, and Zenaida G. Baoanan2*

1College of Forestry, Benguet State University


Km 6, La Trinidad, Benguet 2601 Philippines
2Department of Biology, College of Science
University of the Philippines Baguio, Baguio City 2600 Philippines

The degree of land use and land cover (LULC) conversion of Mt. Pulag National Park (MPNP),
Benguet Province, Philippines, from 1990–2020 was assessed to provide valuable information for
land resource management strategies. The increasing demand for vegetable farming threatens
the park with further degradation. The magnitude of changes in the LULC of MPNP was
determined using Landsat 5 for 1990–2010 and Sentinel-2 for 2020. The FLAASH (Fast Line-of-
sight Atmospheric Analysis of Hypercubes) algorithm was performed to enhance the reflectance
of the satellite images before data analysis. The maximum likelihood classification (MLC) was
used to classify the MPNP into five LULC changes: agricultural lands, barren lands, built-up
areas, forest lands, and grasslands with areal percentage cover of 15.88, 0.26, 1.34, 80.51, and
2.01%, respectively, in 2020, and overall accuracy of 90.58%. The LULC percentage of change
from 1990–2020 showed that agricultural areas increased by 8.96%, barren lands decreased
by 2.14%, built-up areas increased by 1.02%, forest lands decreased by 8.35%, and grasslands
increased by 0.51%. The results showed that the increasing area of human-induced LULC
class such as the agriculture lands and built-up areas is the primary cause of forest land loss.
However, the barren lands undergoing natural regeneration contribute to the gain of the forest
area. Therefore, the LULC conversion pattern serves as a warning that the park will continue
to lose a significant portion of the forest due to land conversion if no intervention is done.

Keywords: LULC, MLC, MPNP, optical data, remote sensing

INTRODUCTION Pielke 2005; Pielke et al. 2011). While we are supposed


to be aiming for positive LULC change through the
The LULC determines the environmental condition and expansion of the forest land cover to mitigate the impact
physical attributes of a particular area and its potential of climate change, this is not the typical scenario. The
for biomass productivity and biodiversity (Anderson increasing human population demands more built-up and
et al. 1976; Yesuph and Dagnew 2019). LULC data agricultural areas at the expense of barren and naturally
play essential roles in ecological assessments relative vegetated lands (Kafi et al. 2014; Toure et al. 2018). Poor
to many environmental concerns and issues affecting environmental governance may further influence land-use
worldwide land cover conversion (Feddema et al. 2005; alteration (Mahmud and Achide 2012). Both human and
*Corresponding Author: zgbaoanan@up.edu.ph
nature-induced disturbances contribute to LULC changes

723
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

(Bajocco et al. 2012), thereby altering the atmospheric, Vizcaya (Figure 1b). The satellite image of MPNP is
climatic, and biotic domains of the Earth at a particular shown in Figure 1c. Mt. Pulag used to be the second-
time (Feddema et al. 2005; Pielke 2005; Pielke et al. highest mountain in the Philippines at 2,926-m elevation
2011). until Mt. Dulang-Dulang in Bukidnon exceeded the record
with a 2,938-m peak (Peterson et al. 2008).
The constant change of land use alters CO2 storage.
Therefore, continuous monitoring of terrestrial ecosystems Buot and Okitsu (1998) determined the annual mean
is essential for sustainable forest management. An temperature of MPNP from 16 °C at 2,000 masl to 14
accurate LULC map supports a precise estimation of °C at the pine forest’s upper limit (2,300 masl). At 2,800
CO2 sequestered from the atmosphere and serves as an masl, the annual mean temperature drops to 12 °C. The
indicator of biodiversity status (Giri et al. 2005; Kafi et mountain has a unique vegetation zonation pattern since
al. 2014). it does not have a dipterocarp flora in its lower slopes,
while on top is defined grassland vegetation (Merrill and
Remotely sensed imagery is gaining a reputation as an Merritt 1910; Hipol et al. 2007).
efficient LULC information predictor (Jensen 2000;
Richards 2013; Shivakumar and Rajashekararadhya 2018)
based on its “unique view” of the spatial and temporal Satellite Images Used for the LULC Prediction
dynamics (Batty and Howes 2001; Roy et al. 2002; To monitor the LULC change of MPNP in this study,
Martha et al. 2008). The satellite images evolve with we utilized the satellite images downloaded freely from
the emergence of new instruments – from Landsat and the United States Geological Survey website ( www.
Sentinel to Worldview – resulting in increasing spatial, earthexplorere.usgs.gov/). We only chose satellite images
spectral, and temporal resolution. The data obtained from with no greater than 15% cloud cover to achieve higher
remote sensing are appropriate for investigating timely accuracy results. For 1990–2010, we used images from
and scheduled changes in vegetation cover and systematic Landsat 5 with 116 WRS path and 049 row on the
observation of environmental conditions (Gong et al. following dates, providing the respective land and scene
2008). cloud cover percentages in parenthesis: 10 Apr 1990 (9.0
and 14.00), 23 Mar 2001 (5.00 and 12.00), and 12 Feb
The field area chosen for this study is MPNP, being 2010 (5.00 and 12.00). Although Landsat 7 was launched
a valuable watershed in the Cordillera and harboring on 05 Apr 1999, the images are not clear due to high cloud
endemic flora and fauna (Fernando and Cereno 2010). cover. Hence, we resolved to use Landsat 5 imagery for
The place was declared as a national park by Presidential 2000 and 2010. All the raster files used were projected
Proclamation No. 75, s. 1987. It was closed to occupancy to the Universal Transverse Mercator system, Zone 51
through Republic Act No. 7586 (National Integrated North, and WGS 84 datum.
Protected Areas System or NIPAS Act of 1992)
to encourage protection of the area and its unique For 2020 LULC analysis, we used Sentinel-2 images
biodiversity. However, the accessibility to tourists – with 116/049 path/row acquired on 27 Feb 2019 and a
coupled with uncontrolled development and expanding scene cloud cover of 2.18. Figure 2a shows the extents
farming – are possible factors leading to continued land of the satellite images relative to the whole Philippines,
cover changes in the park. Hence, this study is conducted while the downloaded satellite images of the study site
to monitor the extent of LULC conversion of MPNP from are presented in Figures 2b–e.
1990–2020 using time-series satellite images, integrating
geographic information systems, remote sensing data, and Existing LULC Identification and Collection of
analysis techniques. We hope to bring to the attention of Training and Validation Data per LULC Class
the park management the current land use map of the area The different types of LULC information describe the
in aid of legislation. vegetation, water, and natural features of the ground
surface. Through visual observation of the satellite
images and verification from the Google Earth desktop
application, five LULC types were designated over MPNP
MATERIALS AND METHODS – namely, agricultural lands, barren lands, built-up areas,
forest lands, and grassland.
Study Site
The training and validation dataset used for classification
The MPNP (~ 11,550 ha) sits in the Cordillera
and accuracy assessment, respectively, were collected
Biogeographic Zone (Figure 1a) bounded by the
from the Google Earth desktop application. Each sample
municipalities of Bokod, Kabayan, and Buguias of the
was randomly picked through the visual inspection of the
Benguet province, Tinoc of Ifugao, and Kayapa of Nueva
study site. A random number of sample data were collected

724
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

Figure 1. Geographical location of the MPNP showing its specific location in the Northern part of the
Philippines (a), the political boundaries of the province surrounding the national park (orange
is the whole Benguet province; blue is part of the Ifugao province and green is part of the
Nueva Vizcaya; red is the boundary of MPNP) (b), and satellite image with the location
of the training (red) and validation (yellow) data representing the LULCs of MPNP (c).

Figure 2. The location of the MPNP in the Philippines (a) and the various satellite
images of the study site from 1990–2020 (b–e).

725
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

per LULC class. Google Earth’s imagery resolution ranges considered to avoid the prevalence sampling effect (Vach
from 15 m to 15 cm, providing very high quality, with 2005; Allouche et al. 2006)
historical and up-to-date details of the Earth’s surface.
A total of 955 pixels were collected to represent the
A supervised selection technique was conducted to ensure existing LULC classes: 24 for agricultural lands, 170 for
that each LULC class is valid for all periods. The LULC barren lands, 45 for built-up areas, 325 for forest lands,
class presumed to have not changed from 1990–2020 was and 295 for the grassland. The total points that were
purposely chosen as training and validation data. The satellite collected in the Google Earth Pro were divided into two:
image of MPNP acquired on 19 Feb 2019 was used to select 20% for the validation dataset for the accuracy assessment
the dataset for each LULC class through the historical of the classification and 80% for the training dataset. No
imagery tool of the Google Earth desktop application. similar sample points were used for both the training and
validation sample since a sample that was already used for
The overall minimum number of samples that were the training would be excluded as a validation sample. The
determined by Equation 1 is 64: distribution of the different training and validation samples
collected in the different locations is shown in Figure 1c.
(1) On the other hand, the appearance of each LULC viewed
on Google Earth in different periods is shown in Figure
3. Numerous points in different locations were selected
where A is the area (ha) of the study site, C is the
per LULC class since pixels in the same LULC class may
coefficient of variation, e is the error, and a is the sampling
have different reflectance.
size. A minimum of 10 samples per LULC class was

Figure 3. The specific locations of the collected dataset for agricultural lands, barren lands,
built-up areas, forest lands, and grassland in each particular period. The images
were outsourced from the Google Earth Pro.

726
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

Preprocessing of Satellite Images first to classify and merge the pixels of the same image
Landsat image preprocessing is a prerequisite when using attributes such as reflectance, color, and texture. This
satellite data, especially if taken in two or more different study used the image segmentation toolbox in ENVI for
periods. The satellite images need to be registered with the segmentation using the edge segmentation and full
each other for comparison. Atmospheric correction is lambda merge algorithm with 10 and 96 as parameters,
needed to remove the effects of the different noises caused respectively. A suitable combination of the parameters is
by sensor errors, the responsitivity of satellite sensors, needed to achieve the best segmentation and grouping
changes in brightness, and atmospheric interventions of pixels.
(Vicente-Serrano et al. 2008), and convert the radiances
acquired by the instruments to the reflectance of the
Satellite Image-based LULC Classification
object’s surface (Gao et al. 2006). The principle of
Technique
atmospheric correction algorithm is based solely on the
This study utilized the MLC algorithm for the classification
spectral radiance that can be expressed as the function of
of the LULC of MPNP based on the Bayes theorem
three components – namely, the radiance that is coming
expressed as Equation 2:
from the surface and captured directly by the sensor, the
radiance that is the effect of the adjacent environments,
(2)
which is coming from the surface being scattered by
the atmosphere into the sensor, and the radiance that
is reflected by the atmosphere without even touching where (P(ω|i) is the likelihood function, P(i) is the a priori
the surface. The FLAASH model approach is proven information, i.e. the probability that object i occurs in the
to provide better results (Lee et al. 2015; Somdatta and focused area, and P(ω) is the probability that ω is observed.
Chakrabarti 2011; Guo and Zeng 2012) compared to DOS
(Dark Object Subtraction), QUAC (Quick Atmospheric The general procedures in MLC are as follows (Ahmad
Correction), 6S, and ATCOR 2-3. The FLAASH model and Quegan 2012):
approach is based mainly on radiative transfer (Guo (1) determining the different land cover types within
and Zeng 2012). It is the most advanced model as it the study area;
also accounts for adjacency effects (Adler-Golden et
al. 1999) during the process of atmospheric correction. (2) choosing the land cover information-based
The adjacency effect compensates for the low resolution training pixels for each of the desired classes;
obtained from Landsat 5 imagery.
(3) estimating the mean vector and covariance matrix
In addition to atmospheric correction, the pansharpening of each class using the chosen training pixels; and
technique was applied to enhance the spatial resolution
(4) classifying each pixel in the image into one of the
of the image while retaining the spectral attributes. It was
established land cover types.
applied to the Sentinel-2 image but was not performed
for the Landsat 5 due to the absence of the panchromatic
band. Although the Sentinel-2 Multispectral Instrument Accuracy Assessment of the LULC Classification
does not have a panchromatic band, the resolution of its The 20% of the data that were not used as training data
four bands at 10 m was used to enhance the resolution of were used for validation assessment of the classification.
the other bands, which have 20–60 m spatial resolution. For the assessment of the accuracy of the LULC
The Sen2Res plugin for SNAP was used in this study to classification per assessing period, the confusion matrix
enhance the resolution of all the bands of the Sentinel-2 was used in which the user, producer, overall accuracy,
to 10 m. and kappa coefficient were assessed. The overall accuracy
was calculated by getting the sum of all the correctly
classified pixels, represented from upper left to lower
Satellite Image Segmentation
right diagonal, and divided with the total number of all
Image classification is either with (supervised) or
the pixels. The user accuracy on the other hand was equal
without (unsupervised) the presence of training data. In
to the sum of correctly classified pixels of a particular
unsupervised classification, image classification is being
class divided by the total number of classified pixels in
processed to cluster pixels based only on the available
the class, while the producer accuracy was the sum of
statistics of the image while in supervised classification,
correctly classified pixels of a particular class divided by
the classification and clustering of pixels are based on the
the number of training pixels of the same class (Patel and
attributes of the provided pre-defined training data. The
Kaushal 2010). The three were all expressed in percentage
MLC, as one of the supervised classification techniques,
thus needed to be multiplied by 100. On the other hand, the
requires the availability of training samples. Before the
kappa coefficient that measured the agreement between
LULC classification, the satellite images were segmented

727
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

classification and truth values ranged from 0–1. According additional technique applicable for Sentinel imagery. The
to McHugh (2012), a kappa value from 0–0.20 have no LULC of MPNP for a particular period was determined
agreement, 0.21–0.39 is minimal, 0.40–0.59 is weak, based on the reflectance of the training data that was
0.60–0.79 is moderate, 0.80–0.90 is strong, and above 0.90 observed in the satellite image.
is very strong and has nearly perfect agreement.
The general reflectance of each LULC across the different
bands of the satellite image conveyed that trees had the
Change Detection and Determination of the highest reflectance followed by the agricultural area,
Magnitude of Change grassland, then barren land. The reflectance was due to
The changes over the five different LULC classes of MPNP the vegetation that was present in the forest, agricultural
were analyzed for three decades covering 1990–2020. The areas, and grassland.
magnitude of change (K) is a change in the size of the
LULC class (expansion or reduction). A negative value The MLC supervised classification was used in this study.
represents a decrease in LULC class size while a positive The training datasets required for classifying the target
value indicates an increase (Mahmud and Achide 2012). pixels were obtained from the Google Earth Pro image.
It is a reliable source of training dataset or ground truth
The K is expressed by the following formula (Equation 3): data since it has high resolution resembling the real ground
𝐾 = 𝐹 − 𝐼 (3) situation. Through the MLC technique, the LULC of
MPNP from 1990–2020 was determined.
On the other hand, the percentage of change (A) is The assessment of the performance of the MLC in
expressed as Equation 4: classifying the different LULC of MPNP in a particular
assessing period was conducted using the confusion
(4)
matrix through the computation of the user’s, producer’s,
and the overall accuracies. The kappa coefficient was also
where K is the magnitude of change, A is the percentage used as one of the evaluation criteria. The computation
of change, F is the first date, and I is the reference date. results per decade are shown in Tables 1–4.
It can be seen in the confusion matrices throughout the
decades that the user’s accuracy ranges from 45.45% – as
observed from built-up areas in 1990 (Table 1) – to 100%
RESULTS from barren lands in 2000 (Table 2), 2010 (Table 3), and
2020 (Table 4), and from grassland in 2000 (Table 2).
Performance of MLC Classified Satellite Image- With the computed 100% accuracy, there was no error
based LULC committed. The grassland obtained the highest producer’s
In this study, the MPNP was classified into five LULC accuracy as well, with the computed value of 96.61%
types using the multi-temporal Landsat images that (Tables 1 and 2). The lowest producer’s accuracy of
were atmospherically corrected. Pansharpening was an 54.17% was recorded in built-up LULC in 1990 (Table

Table 1. Confusion matrix of the LULC classification in MPNP for the year 1990.
Observed LULC class
Built- User's accuracy (%)
1990 Agriculture Barren Forest Grassland Total
up
Agriculture 13 0 2 0 0 15 86.67
Predicted LULC

Barren 0 29 2 0 31 93.55
Built-up 6 0 5 0 0 11 45.45
Forest 5 5 2 62 2 76 81.58
Grassland 0 0 0 1 57 58 98.28
Total 24 34 9 65 59 191
Producer's accuracy (%) 54.17 85.29 55.56 95.38 96.61
Kappa coefficient 0.82
Overall accuracy 86.91

728
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

Table 2. Confusion matrix of the LULC classification in MPNP for the year 2000.

Observed LULC class


User's accuracy (%)
2000 Agriculture Barren Built-up Forest Grassland Total

Agriculture 13 0 4 3 0 20 65.00
Predicted LULC

Barren 0 30 0 0 0 30 100.00
Built-up 2 0 5 0 0 7 71.43

Forest 9 4 0 62 2 77 80.52

Grassland 0 0 0 0 57 57 100.00

Total 24 34 9 65 59 191

Producer's accuracy (%) 54.17 88.24 55.56 95.38 96.61

Kappa coefficient 0.83

Overall accuracy 87.43

Table 3. Confusion matrix of the LULC classification in MPNP for the year 2010.
Observed LULC class
User's accuracy (%)
2010 Agriculture Barren Built-up Forest Grassland Total
Agriculture 16 0 4 2 0 22 72.73
Predicted LULC

Barren 31 0 0 31 100.00
Built-up 2 0 5 0 0 7 71.43
Forest 6 3 0 62 3 74 83.78
Grassland 0 0 0 1 56 57 98.25
Total 24 34 9 65 59 191
Producer's accuracy (%) 66.67 91.18 55.56 95.38 94.92
Kappa coefficient 0.85
Overall accuracy 89.01

Table 4. Confusion matrix of the LULC classification in MPNP for the year 2020.
Observed LULC class
User's accuracy (%)
2020 Agriculture Barren Built-up Forest Grassland Total
Agriculture 16 0 2 2 0 20 80.00
Predicted LULC

Barren 0 32 0 0 0 32 100.00
Built-up 8 0 7 0 0 15 46.67
Forest 0 2 0 62 3 67 92.54
Grassland 0 0 0 1 56 57 98.25
Total 24 34 9 65 59 191
Producer's accuracy (%) 66.67 94.12 77.78 95.38 94.92
Kappa coefficient 0.87
Overall accuracy 90.58

1) and 2000 (Table 2). The overall accuracy in LULC coefficient, ranging from 0.82 in 1990 to 0.87 in 2020.
classification shows this pattern in the increasing order: These values are considered to be very strong and nearly
86.91% in 1990, 87.43% in 2000, 89.01% in 2010, and in perfect agreement (McHugh 2012).
90.58% in 2020. The same trend follows for the kappa

729
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

The accuracy assessment – combining the user’s, built-up that was increased by 0.53% resulting to 0.85%.
producer’s, overall accuracy, and kappa coefficient – In 2010, the area of the forest continued to decline by
implies that the LULC classification applied for MPNP 3.33% resulting in area coverage of 83.18% although it
was reliable. was still the major cover of MPNP. On the other hand,
the agriculture LULC lands continued to increase by
3.69% and already covered 13.61% of MPNP, while the
Percentage Area Distribution of LULC of Mt. Pulag
grasslands covered 1.71% of the area. The built-up areas,
in a Particular Year
on the other hand, covered 0.92%, which was increased
The resultant maps in Figure 4 indicate the changing
by 0.07%. The remaining area was still barren, occupying
LULC in MPNP from 1990–2020. During the year 1990,
0.58%. In 2020, the forest area continued declining to
MPNP was covered mainly by forest occupying 88.86%
80.51% due to the expansion of the other LULC classes.
of the total area followed by agriculture LULC (6.93%),
The agricultural areas, grasslands, barren areas, and
barren lands (2.39%), grasslands (1.50%), and built-up
built-up covered 15.88%, 2.01%, 0.26%, and 1.34%,
area (0.32%). After 10 years (2000), the majority of the
respectively.
national park was still covered by forest but was decreased
by 2.35% resulting in 86.51%, then agriculture (9.92%) Overall, the forest lands decreased from 1990–2020 by
that was increased by 2.99%, grassland that was increased 8.35%. The agriculture areas increased by 8.95%, the
by 0.11% resulting in an area of 1.61%, barren lands grassland increased by 0.51%, the barren lands decreased
that were decreased by 1.29% resulting to 1.11%, and by 2.14% while the built-up increased by 1.02%. The
LULC changes for three decades resulted in 0.28,
0.30, 0.02, 0.07, and 0.034% per year for forest lands,
agriculture areas, grassland, barren lands, and built-up
areas, respectively.

The Extent of LULC Conversion in MPNP from


1990–2020
The magnitude of change (K) is a change in the size of
the LULC (expansion or reduction).
The magnitude of change of the LULC size was analyzed
on a decadal basis from 1990–2020. Table 5 shows the
summary of the aerial percentage of each LULC per
period. Table 6, on the other hand, shows that agriculture,
built-up areas, and grasslands are constantly increasing
throughout the years, as indicated by the positive values.
The forest lands and barren lands are decreasing, as
indicated by the negative values. The agricultural areas
had the highest magnitude and percentage of change with
values of 397.70 ha (43.19%) from 1990–2000, 490.43
ha (37.20%) from 2000–2010, and 301.15 ha (16.65%)
in 2010–2020. The highest percentage of change was
observed from 1990–2000. The grassland increased by
14.62 ha (7.33%) from 1990–2000, 13.29 ha (6.21%),
and 39.52 ha (17.40%) from 2010–2020. The built-up
areas increased by 70.61 ha (166.07%) from 1990–2000,
8.55 ha (7.56%) from 2000–2010, and 56.61 ha (46.52%)
from 2010–2020. The forest decreased by 312.24 ha
(–2.64%) from 1990–2000, by 442.46 ha (–3.85%) from
2000–2010, and by 354.76 ha (–3.21%) from 2010–2020.
The decrease of the area of the barren lands was 170.68 ha
(–53.75%), 69.82 ha (–47.53%), and 42.52 ha (–55.17%)
from 1990–2000, from 2000–2010, and from 2010–2020,
respectively.
Figure 4. The LULC map of the MPNP from 1990–2020 with the
corresponding areal percentage of each LULC class.

730
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

Table 5. Areal percentage of each LULC class in the different periods.


1990 2000 2010 2020
LULC class
Area(ha) % Area % Area % Area %

Agriculture 920.79 6.93 1318.48 9.92 1808.92 13.61 2110.06 15.88

Barren 317.56 2.39 146.88 1.11 77.06 0.58 34.55 0.26

Built-up 42.52 0.32 113.13 0.85 121.68 0.92 178.29 1.34

Forest 11806.80 88.86 11494.56 86.51 11052.10 83.18 10697.34 80.51

Grassland 199.30 1.50 213.92 1.61 227.21 1.71 266.73 2.01

Total 13286.97

Table 6. Magnitude and percentage of LULC change from 1990–2020.


1990–2000 2000–2010 2010–2020
LULC class
Magnitude (ha) % Magnitude (ha) % Magnitude (ha) %
Agriculture 397.70 43.19 490.43 37.20 301.15 16.65
Barren –170.68 –53.75 –69.82 –47.53 –42.52 –55.17
Built-up 70.61 166.07 8.55 7.56 56.61 46.52
Forest –312.24 –2.64 –442.46 –3.85 –354.76 –3.21
Grasslands 14.62 7.33 13.29 6.21 39.52 17.40
Note: the positive values indicate increasing trend of change while negative values mean decreasing trend.

DISCUSSION brought about by the combined expansion of agricultural


area and built-up in 2020, leading to greater forest loss.
Factors attributed to LULC conversion such as natural The increasing coverage of grassland areas likewise results
and human activities may either have positive or negative in reduced forest cover.
changes with corresponding impacts. An example of
positive change is the decrease in the barren land, which In the natural setting, negative LULC conversion is
is an indicator of regenerating vegetation. Looking back the shift from vegetated to non-vegetated LULC or the
at Figure 4, an orange color – which represents the barren shift from rich to poor biodiversity. The positive LULC
land in 1990 – has been replaced by green color, indicating conversion is the opposite. In the case of MPNP, the shift
regrowth of trees in 2020. However, this gain in forest from one particular LULC to another within the span of
cover is lesser compared to human-induced changes 30 years is summarized in Table 7. The table shows that

Table 7. Transition tables per LULC class for the years 1990, 2000, 2010, and 2020.
Original LULC Current Shift in LULC (%)
(1990) LULC 1990–2000 2000–2010 2010–2020
Agriculture Agriculture 100.00 100.00 100.00
Barren 46.44 52.25 44.83
Barren
Forest 53.56 47.75 55.17
Built-up Built-up 100.00 100.00 100.00
Agriculture 3.36 4.27 2.73
Built up 0.60 0.08 0.50
Forest
Forest 95.91 89.44 96.41
Grassland 0.12 6.21 0.36
Grassland Grassland 100.00 100.00 100.00

731
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

there was no land conversion in the agriculture, built-up, The pixels in the training data were registered and used
and grassland for 30 years whereas portions of the forest to classify the rest of the pixels. Despite the limitations in
and barren lands were converted to other land uses. The using optical datasets, the overall accuracy was still high
rate of conversion from barren to forest was highest from and more cost-effective for LULC classification.
2010–2020. The rate of conversion from forest to other
land uses was also highest in 2000–2010, where 10.56% The combination of optical data and MLC supervised
of the forest area was converted to agriculture, built-up, classification produced reliable LULC maps of MPNP,
and grassland as compared to 4.09% in 1990–2000 and which facilitated measurements of the magnitude and
3.59% in 2010–2020. percentage of change. Anthropogenic-related activities
caused the alarming decline of the forest cover warranting
Our findings provide empirical data to prove that the effective intervention. It is, therefore, recommended that
LULC transformations in MPNP are mainly human- the local government pay attention to the LULC changes
induced. Abino et al. (2015) and Nauta et al. (2003) in Mt. Pulag to prevent further destruction and to protect
reported the same trend for the Marikina sub-watershed the remaining biodiversity and the ecological services
and Laguna de Bay, Laguna, Philippines, respectively. provided by the park.
Most of the negative conversions in key biodiversity
areas or national parks are attributed to human activities
(Santillan et al. 2019; Soriano et al. 2019). It seems ironic
since these areas are supposed to be protected and free ACKNOWLEDGMENTS
from human intervention. But with the growing population
This study was supported by CHED-DARE TO
estimated to reach 9 billion people by 2050 (Roser and
(Commission on Higher Education–Discovery Applied
Ortiz-Ospina 2013; Sylvester 2018), the agricultural
Research and Extension Trans/Inter-Disciplinary
demand is predicted to rise by 69% (Alexandros 2012).
Opportunities) through the project, ”Popularizing Access
The Benguet province, where MPNP lies, is one of the top
to Biodiversity Information Data and Conservation
vegetable-producing provinces in the country. Farming is
Opportunities (PABIDACO).” Gratitude is also extended
therefore the main reason for land conversions not only
to the Protected Area and Management Office of the
in Benguet but to many other mountainous regions in
MPNP for permission to conduct the study.
the Cordillera (Lapniten 2020). Without proper planning
and management, even the protected areas may not be
exempted from human encroachment.
MPNP is also a famous ecotourism destination in Northern STATEMENT ON CONFLICT OF
Luzon. Irresponsible trekkers may also cause damage to INTEREST
the park such as the reported grassfire in January 2018
(Bigtas 2018). There is no conflict of interest that the authors need to
declare.
The decreasing forest cover poses threat to the unique
flora and fauna thriving in the area. The ecological goods
and services are likewise affected. The people relying on
the ecosystem services provided by the park are on the REFERENCES
losing end.
ABINO AC, KIM SY, JANG MN, LEE YJ, CHUNG
Strengthening the community-based forest management JS. 2015. Assessing land use and land cover of the
systems may alleviate the problem of land conversions. Marikina sub-watershed, Philippines. Forest Sci Tech
Bringing awareness to the people on the importance of 11(2): 65–75.
protecting natural resources may lead to a paradigm shift AHMAD A, QUEGAN S. 2012. Analysis of Maximum
towards sustainability. Likelihood Classification on Multispectral Data. Appl
Math Sci 6: 6425–6436.
A D L E R - G O L D E N S M , M AT T H E WA M W,
CONCLUSION BERNSTEINA LS. LEVINE RY, BERK A,
RICHTSMEIER SC, ACHARYA PK, ANDERSON
The LULC transformations in MPNP from 1990–2020 GP, FELDE JW, GARDNER JA, HOKE ML, JEONG
were monitored and classified using Landsat 5 and LS, PUKALL B, RATKOWSKI AJ, BURKE HK.
Sentinel-2 imagery and MLC supervised classification 1999. Atmospheric correction for shortwave spectral
technique. The LULC of Mt. Pulag was classified imagery based on MODTRAN4. In: Proceedings on
depending on the reflectance that was registered by a pixel.

732
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

SPIE’s International Symposium on Optical Science, GAO BC, DAVIS CURTISS O, GOETZ AFH. 2006.
Engineering, and Instrumentation Vol. 3753; 27 A review of atmospheric correction techniques for
Oct 1999; Denver, CO, United States. https://doi. hyperspectral remote sensing of land surfaces and
org/10.1117/12.366315 ocean colour. In: Proceedings of IEEE International
Symposium on Geoscience and Remote Sensing; 13
ALEXANDROS N, BRUINSMA J. 2012. World
Jul–04 Aug 2006; Denver, CO, USA. p. 1979–1981
Agriculture towards 2030/2050: The 2012 Revision;
doi: 10.1109/IGARSS.2006.512
ESA Working Paper No. 12-03. Rome, Italy: FAO.
Retrieved from on 28 Feb 2021 http://www.fao.org/ GIRI C, ZHU Z, REED B. 2005. A comparative analysis
fileadmin/templates/esa/Global_persepctives/world_ of the Global Land Cover 2000 and MODIS land cover
ag_2030_50_2012_rev.pdf data sets. Remote Sens Environ 94: 123–132.
ALLOUCHE O, TSOAR A, KADMON R. 2006. Test the GONG J, SUI H, MA G, ZHOU Q. 2008. A review
accuracy of species distribution models: prevalence, of multi-temporal remote sensing data change
kappa and the true skill statistic (TSS). J Appl Ecol detection. Int Arch Photogramm Remote Sens Spat
43: 1223–1232. Inf Sci XXXVII(B7): 757–762. 
ANDERSON R, ERNEST EH, JOHN TR, RICHARD GUO Y, ZENG F. 2012. Atmospheric correction
EW. 1976. A land use and land cover classification comparison of Spot-5 image based on model FLAASH
system for use with remote sensor geological survey and model QUAC. ISPRS-Int. Arch Photogramm
professional paper 964. 1998, 2000, 2001. In: Bajocco Remote Sens Spat Inf Sci XXXIX(B7): 7–11. 10.5194/
S, de Angelis A, Perini L, Ferrara A, Salvati L. 2012. isprsarchives-XXXIX-B7-7-2012
The impact of land use/land cover changes on land
HIPOL RM, TOLENTINO DB, FERNANDO ES, CADIZ
degradation dynamics: a Mediterranean case study.
NM. 2007. Life strategies of mosses in Mt. Pulag,
Environ Manage 49(5): 980–989. doi: 10.1007/s00267-
Benguet Province, Philippines. Philipp J Sci 136(1):
012-9831-8
11–18.
BATTY M, HOWES D. 2001. Predicting temporal
JENSEN JR. 2000. Introductory digital image processing:
patterns in urban development from remote imagery.
a remote sensing perspective. 3rd ed. New Jersey:
In: Remote Sensing and Urban Analysis. Donnay
Prentice-Hall Inc. 526p.
JP, Barnsley MJ, Longley PA eds. London: Taylor &
Francis. p. 185–204. KAFI KM, SHAFRI HZM, SHARIFF ABM. 2014.
An analysis of LULC change detection using
BIGTAS JA. 2018. Look: this is Mt. Pulag after it was
remotely sensed data: a case study of Bauchi City.
accidentally set on fire. GMA News Online. Retrieved
IOP Conf Series: Earth Environ Sci 20: 012056.
on 28 Feb 2021 from https://www.gmanetwork.com/
doi:10.1088/1755-1315/20/1/012056
news/lifestyle/healthandwellness/641014/this-is-mt-
pulag-after-it-was-accidentally-set-it-on-fire/story/ LAPNITEN K. 2020. Averting an agricultural and
ecological crisis in the Philippines’ salad bowl.
BUOT IE, OKITSU S. 1998. Vertical distribution and
Mongabay News & Inspiration from Natures Frontline.
structure of the tree vegetation in the montane forest
Retrieved on 28 Feb 2021 from https://news.mongabay.
of Mt. Pulog, Cordillera mountain range, the highest
com/2020/03/averting-an-agricultural-and-ecological-
mountain in Luzon Is., Philippines. Veget Sci 15:
crisis-in-the-philippines-salad-bowl/
19–32.
LEE SB, LA HP, EO YD, PYEON MW. 2015. Generation
FEDDEMA JJ, OLESON KW, BONAN GB, MEARNS
of simulated image from Atmospheric Corrected
LO, BUJA LE, MEEHL GA. 2005. The importance of
Landsat TM Images. J Korean Soc Surv Geodesy
land-cover change in simulating future climates. Sci
Photogramm Cartogr 33(1): 1–9.
310: 1674–1678.
MAHMUD A, ACHIDE AS. 2012. Analysis of land use/
FERNANDO ES, CERENO RP. 2010. Biodiversity
land cover changes to monitor urban sprawl in Keffi-
and natural resources management in the Mt. Pulag
Nigeria. Environ Res J 6: 129–134.
National Park. In: Lapitan PG, Fernando ES, Suh
MH et al. eds. Biodiversity and Natural Resources MARTHA SG, EMILY HW, NELS B. 2008. Integrating
Conservation in Protected Areas of Korea and the multi-temporal spectral and structural information to
Philippines, Chapter 4. ASEAN-Korea Environmental map wetland vegetation in a lower Connecticut River
Cooperation Unit, Seoul National University, Korea. tidal marsh. Remote Sens Environ 112: 40–48.
p. 120–177.

733
Philippine Journal of Science Doyog et al.: LULC Changes in MPNP
Vol. 150 No. 4, August 2021

MERRILL E, MERRITT M. 1910. The flora of Mt. Pulag. SOMDATTA C, CHAKRABARTI S. 2011. Preprocessing
Philipp J Sci 5: 287–403. of Hyperspectral Data: A case study of Henry and
Lothian Islands in Sunderban Region, West Bengal,
MCHUGH M. 2012. Interrater Reliability: The Kappa
India. Int J Geomat Geosci 2(2): 490–501.
Statistic. Biochemia Medica: Časopis Hrvatskoga
Društva Medicinskih Biokemičara / HDMB  22 SORIANO M, HILVANO N, GARCIA R, HAO AJ,
(October): 276–282. ALEGRE A, TIBURAN C. 2019. Land Use/Land
Cover Change Detection and Urban Sprawl Analysis
NAUTA T, BONGCO A, SANTOS-BORJA A. 2003. Set-
in the Mount Makiling Forest Reserve Watersheds
up of a decision support system to support sustainable
and Buffer Zone, Philippines. Environ 6(9): 1–20.
development of the Laguna de Bay, Philippines. Mar
doi:10.3390/environments6020009
Pollut Bull 1(6): 211–218.
SYLVESTER G. 2018. E-Agriculture in action: drones
PATEL N, KAUSHAL B. 2010. Improvement of user’s
for agriculture. Bangkok: Food and Agriculture
accuracy through classification of principal component
Organization of the United Nations and International
images and stacked temporal images. Geospatial
Telecommunication Union. 112p.
Inform Sci 13(4): 243–248.
TOURE SI, STOW DA, SHIH HC, WEEKS J, LOPEZ-
PETERSON AT, BROOKS T, GAMAUF A, GONZALEZ
CARR D. 2018. Land cover and land use change
JCT, MALLARI NAD, DUTSON G, BUSH SE,
analysis using multi-spatial resolution data and
CLAYTON DH, FERNANDEZ R. 2008. The avifauna
object-based image analysis. Remote Sens Environ
of Mt. Kitanglad, Bukidnon Province, Mindanao,
210: 259–268.
Philippines. Fieldiana Zool 114: 1–43.
VACH W. 2005. The dependence of Cohen’s kappa on the
PIELKE SR RA. 2005. Land use and climate change. Sci
prevalence does not matter. J Clin Epidemiol 58(7):
310: 1625–1626. doi: 10.1126/science.1120529
655–661.
PIELKE SR RA, PITMAN A, NIYOGI D, MAHMOOD
VICENTE-SERRANO SM, PÉREZ-CABELLO F,
R, MCALPINE C, HOSSAIN F, ET AL. 2011. Land
LASANTA T. 2008. Assessment of radiometric
use/land cover changes and climate: modeling analysis
correction techniques in analyzing vegetation
and observational evidence. WIRES Clim Change 2:
variability and change using time series of Landsat
828–850.
images. Remote Sens Environ 112: 3916–3934.
RICHARDS JA. 2013. Remote sensing digital image
YESUPH AY, DAGNEW AB. 2019. Land use/cover
analysis, 5th ed. Verlag Berlin Heidelberg: Springer.
spatiotemporal dynamics, driving forces and
494p.
implications at the Beshillo catchment of the Blue Nile
ROSER M, ORTIZ-OSPINA E. 2013. World population Basin, North Eastern Highlands of Ethiopia. Environ
growth. Published online at OurWorldInData.org. Syst Res 8(21): 1–30. doi.org/10.1186/s40068-019-
Retrieved on 26 Jun 2020 from https://ourworldindata. 0148-y
org/ world-population-growth
ROY DP, LEWIS PE, JUSTICE CO. 2002. Burned
area mapping using multi-temporal moderate spatial
resolution data—a bi-directional reflectance model-
based expectation approach. Remote Sens Environ
83: 263–286.
SANTILLAN JR, AMORA AM, MAKINANO-
SANTILLAN M, GINGO AL, MARQUESO JT.
2019. Analyzing the impacts of land cover change
to the hydrologic and hydraulic behaviours of the
Philippines’ third largest river basin. ISPRS-Ann
Photogramm Remote Sens Spat Inf Sci IV-3/W1:
41–48. doi.org/10.5194/isprs-annals-IV-3-W1-41-2019
SHIVAKUMAR BR, RAJASHEKARARADHYA SV.
2018. Investigation on Land Cover Mapping Capability
of Maximum Likelihood Classifier: A Case Study on
North Canara, India. Proc Comp Sci 143: 579–586.

734

You might also like