Google Earth Engine for Assessing Land Use and Land Cover

Change in Banyuwangi Regency

Abdul Holik

, Zulis Erwanto

and Siska Aprilia Hardiyanti

Department of Agribusiness, Politeknik Negeri Banyuwangi, Jl. Raya Jember Km 13 Kabat, Banyuwangi, Indonesia

Department of Civil Engineering, Politeknik Negeri Banyuwangi, Jl. Raya Jember Km 13 Kabat, Banyuwangi, Indonesia

Keywords: GEE, Land Use Land Cover, Remote Sensing, Supervised Classification, Banyuwangi.

Abstract: Land use and land cover (LULC) change needs regularly monitor because it causes a multiplayer effect. This

study examines the LULC in Banyuwangi regency between 2013 using Landsat-8 and 2021 using Sentinel-2

satellite images data. All processing data used Google Earth Engine (GEE) with supervised classification,

classified eight LULC characteristics. The results show that GEE can quickly process image data from the

acquisition process, cloud removal, classification to accuracy tests. LULC significant changes in paddy fields

(-3.76%) and forests (+3.08%), moderate changes in plantations (+0.44 %), water bodies (-0.39%), built-up

(+0.34%), and bare soil (+0.27%), while minor changes occur in field (-0.01%) and shrubs (+0.26%). Even

though forest land is increasing, the loss of paddy fields must be a serious concern to maintain the stability of

rice stocks.

1 INTRODUCTION

Remote sensing is a popular tool for mapping and

monitoring land use/land cover change. It is currently

the most reliable tool for monitoring various

spectrally sensitive changes of the earth (Alam, Bhat,

& Maheen, 2020). Several satellite images like

Landsat and Sentinel with band-combination are

widely used to detect the changes on a large scale.

Google earth engine (GEE) is a cloud-based

platform for geospatial data analysis, especially raster

data. GEE provides powerful computing capabilities

and provides many remote sensing data and various

auxiliary datasets (Qu, Chen, Li, Zhi, & Wang, 2021).

GEE can be accessed via an application programming

interface (API) over the web-based Internet. Users

can prototype, visualize results quickly (Gorelick et

al., 2017), and also analyze big geospatial data in a

sophisticated way without requiring access to

supercomputers or special coding skills (Tamiminia

et al., 2020). In addition, (Zurqani, Post, Mikhailova,

Schlautman, & Sharp, 2018) demonstrates the

advantages of using GEE and the public archive

https://orcid.org/0000-0001-9987-1161

https://orcid.org/0000-0001-7938-9116

https://orcid.org/0000-0001-6003-6791

database on its platform to track and monitor changes

in LULC over time.

The impacts caused by land cover changes are a

reduction of agricultural land (Wijaya & Susetyo,

2017), implications for groundwater quality (He, Li,

Wu, & Elumalai, 2019), sooner and later, theland is

converted into wasteland (Ramanamurthy &

Victorbabu, 2021). Monitoring of LULC to map

needs regular updates. It requires high spatial

resolution time-series data to understand better LULC

dynamics and their impact on biomass (Jr et al.,

2020).

Banyuwangi Regency is the largest regency and

the most significant contributor to staple foods in East

Java. In the last ten years, Banyuwangi has

experienced rapid development (international airport,

mining, railway factory, sugar industry factory,

building of five-star hotels and villas), so that land use

and land cover (LULC) change is inevitable. Rural

areas are experiencing profound social-economic

transitions driven by industrialization and

urbanization (Zhu et al., 2021) as well as

Banyuwangi. Practically, land cover change is

expected in the developing regions and part of the

416

Holik, A., Erwanto, Z. and Hardiyanti, S.

Google Earth Engine for Assessing Land Use and Land Cover Change in Banyuwangi Regency.

DOI: 10.5220/0010946800003260

In Proceedings of the 4th International Conference on Applied Science and Technology on Engineering Science (iCAST-ES 2021), pages 416-422

ISBN: 978-989-758-615-6; ISSN: 2975-8246

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

regional development indicators. Nevertheless, this

certainly impacts the carrying capacity of the

environment, especially in the agricultural sector. The

monitoring results are expected to be useful for

policymakers in Banyuwangi, especially to maintain

food stocks in a stable condition.

2 STUDY AREA

The study area covers about 3,600 square kilometres

with a coastline of about 175.8 km, including all

urban agglomeration which is located between the

latitude of 7°43' − 8°46' and longitude of 113°53' −

114°38'. The east, Banyuwangi is bordered by the

Bali strait and the west by Ijen mountain, which is

3,200 m high. While the south is bordered directly by

the Indian Ocean. The total population of

Banyuwangi regency increased from 1,559,088 in

2010 to 1,754,719 inhabitants in 2020 (BPS Badan

Pusat Statistik, 2020).

Topographically, the study area can be divided

into three distinct regions: the mountainous area

located to the west and northwest (Figure 1), north

and south dominated by dense forests, and

settlements are primarily in flat areas that tend to be

close to coastal areas and lowland areas. The average

slope level in the western and northern regions is 40°,

with a higher average rainfall when compared to other

parts of the region.

Figure 1: The topographic of Banyuwangi.

Lowlands stretch from south to north. Wherein

many rivers always flow in throughout the year. In

Banyuwangi, there are 35 watersheds, so that besides

being able to irrigate an extensive view of the rice

fields as well positive effect on the level of soil

fertility. The mountainous region is a plantation

product-producing area. Lowland produces primary

agricultural products, and areas around the beach line

produce various marine biota.

3 METHODS OF DATA

ACQUISITION AND ANALYSIS

Landsat images and Sentinel images were collected

from the official webpage of the Google Earth Engine

(GEE). LULC change monitoring requires data for

two different periods. Remote sensing approaches

typically use two or more date satellite imagery to

measure land use and land cover changes in any area

(Alam et al., 2020). In this study, collecting data was

in 2013 for Landsat-8 and 2021 for sentinel-2 images.

The spatial resolution of the image is 30*30 m to

10*10 m pixel size. The images were georeferenced

and projected as the Universal Transverse Mercator

(UTM) within Zone 50 S. Details of the satellite data

used in this study are presented in Table 1.

Table 1: Characteristic of the datasets.

Data

Spatial

resolution

Time

acquisition

GEE Image

Collection ID

Landsat-8 30 m 2013

LANDSAT/LC

08/C01/T1_TO

Sentinel-2 10 m 2021

COPERNICUS

/S2

In this analysis, LULC classification was

implemented based on eight classes, and a brief

description of major LULC classes is provided in

Table 2. This classification is based on the Indonesian

National Standard (SNI) 7645:2010. This standard

refers to land cover classification united nation – food

and agriculture organization LCCS-UNFAO and ISO

19144-1 geographic information system- part 1:

classification system structure and developed with

phenomena that exist in Indonesia.

Google Earth Engine for Assessing Land Use and Land Cover Change in Banyuwangi Regency

417

Table 2: LULC classification scheme.

LULC type Description value

paddy field the agricultural area that is

inundated with water either

with irrigation technology,

rainfed, low tide, or tidal,

which is characterized by a

bund pattern, by planting short-

lived food cro

add

field an area used for agricultural

activities with seasonal crops

on dr

lan

plantation land used for agricultural

activities without changing

crops for two years

shrubs the dry land area has been

overgrown with various

heterogeneous and

homogeneous natural

vegetation; the density level is

rare to dense. The area is

dominated by low vegetation

(natural). Bushes in Indonesia

are usually former forest areas

and usually do not show any

traces or build-up spots

anymore

forests forests that grow and develop

in dryland habitats which can

be lowland forests, hills,

mountains, or highland tropical

forests

built-up area areas where the natural or

semi-natural land cover has

been substituted with an

artificial land cover, which is

usually impermeable and

relatively permanent

water bodies all aquatic features, including

seas, reservoirs, coral reefs,

and seagrass beds

bare soil land without cover, whether

natural, semi-natural, or

artificial. according to

characteristics

4 RESULT AND DISCUSSION

4.1 Region of Interest

The first thing that must be done in the classification

process is to make a boundary of the study area. This

is intended so that the classification process can be

carried out correctly without and not going out of the

predetermined limits. ROI can be created by

uploading a shapefile file and saving it to a GEE’s

database. Then during the classification process, this

ROI can be recalled and used as a geometry variable.

The script used to call the uploaded data is as

follows:

var shp =

ee.FeatureCollection('users/abdu

lholik/adm_bwi')

Map.addLayer(shp)

shp is the variable name, while the script in brackets

is a link to the storage location. ROI can be shown by

the following Figure 2.

Figure 2: Region of Interest (ROI).

4.2 Image Acquisition

The image acquisition process can be quickly done by

looking at the Earth Engine Data Catalog script

(https://developers.google.com). It depends on the

type of satellite used. Different satellite, then the code

name will also be different. The acquired image data

can be directly adjusted to the time and condition of

the cloud cover, where cloud cover can be cleared all

at once.

The script used for Landsat-8 in this process is as

follows:

var maskL8 = function(image) {

var qa = image.select('BQA');

var mask = qa.bitwiseAnd(1 <<

4).eq(0);

return image.updateMask(mask);}

var composite =

ee.ImageCollection('LAND

SAT/LC08/C01/T1_TOA')

.filterDate('2013-04-

01','2013-12-31')

.map(maskL8)

.median()

.clip(shp);

While script for sentinel-2 is as follows:

function maskS2clouds(image) {

var qa = image.select('QA60');

var cloudBitMask = 1 << 10;

var cirrusBitMask = 1 << 11;

iCAST-ES 2021 - International Conference on Applied Science and Technology on Engineering Science

418

var mask =

qa.bitwiseAnd(cloudBitMask

).eq(0).and(qa.bitwiseAnd(

cirrusBitMask).eq(0));retu

image.updateMask(mask).divide(10000

);}

var composite =

ee.ImageCollection('COPERN

ICUS/S2_SR')

.filterDate('2021-01-01', '2021-

06-30')

.filter(ee.Filter.lt('CLOUDY_PIX

EL_PERCENTAGE',20))

.map(maskS2clouds)

.median()

.clip(shp);

To visualize the image data, the script is written as

follows:

var RGBTrue =

composite.select(['B4', 'B3',

'B2']);

var RGBparam = {min: 0, max: 0.3,};

Map.addLayer(RGBTrue, RGBparam,

'TRUE');

The results are shown in Figure 3. The true colours

displayed are the composite colours of band 4, band

3, and band 2 for landsat-8 and band 5, band 4, and

band 3 for Sentinel-2.

Figure 3: True Color composite band.

4.3 Training Area

LULC classes are earned by creating a training area.

The training area is an area created to represent the

value of each class. It is created by adding a geometry

variable and filling it with a name and value

(according to Table 2) for each training area as Figure

Figure 4: Properties of the training area.

4.4 LULC Classes

After the training area has been created, the next step

is to run the classification process through guided

classification. In this study, the algorithm used in the

classification process uses a random forest. The

Random Forest (RF) classification method is used

most often and shows minimum and maximum

overall accuracy compared to other classifiers

(Tamiminia et al., 2020). The classification results are

shown by the following Figure 5 and the script used

in this step is:

var aoi =

paddy_field.merge(field).merge(plan

tation).merge(shurbs).merge(forest)

.merge(built_up_area).merge(water_b

odies).merge(bare_soil);

var bands =

['B1','B2','B3','B4','B5','B6','B7'

];

var training =

composite.select(bands).sampleRegio

ns({ collection: aoi, properties:

['lc'], scale: 30});

var classifier =

ee.Classifier.smileRandomForest(100

).train({features: training,

Google Earth Engine for Assessing Land Use and Land Cover Change in Banyuwangi Regency

419

classProperty:'lc',inputProperties:

bands});

var classified =

composite.select(bands).classify(cl

assifier);

Map.addLayer(classified,{min: 1, max:

8,palette:['#98ff00','#c4980e','#46

a37a','#40ab36','#048711','#ff0000'

,'#68d9ff','#ffb868']},'landcover';

Figure 5: LULC classification in 2013 and 2021.

4.5 Accuracy Test

The accuracy test is carried out in almost the same

way as making a classification, namely making a

sample training area. This training area sample is then

combined with the classification results for the

validation process. The accuracy test script is shown

as follows, while the accuracy test matrix is shown in

Table 3 and 4.

var uji_aoi =

accuracy_test_paddy_field.merge(acc

uracy_test_field).merge(accuray_tes

t_plantation)

.merge(accuracy_test_shurbs).merge(acc

uracy_test_forest).merge(accuracy_t

est_built_up_area).merge(accuracy_t

est_water_bodies).merge(accuracy_te

st_bare_soil);

var validasi =

classified.sampleRegions({collectio

n: uji_aoi,properties:

['lc'],scale: 30,});

print(validasi);

var akurasi =

validasi.errorMatrix('lc',

'classification');

print('Confusion matrix', akurasi);

print('Overall accuracy: ',

akurasi.accuracy());

var class_areas =

ee.Image.pixelArea().divide(1000*10

00).addBands(classified)

.reduceRegion({

reducer:

ee.Reducer.sum().group({groupField:

1,groupName: 'code',}),geometry:

shp, maxPixels : 5000000000,scale:

30,})

.get('groups');

print(class_areas);

Table 3: Accuracy of LULC maps obtained from 2013

Landsat-8 satellite data.

Matrix

Sample area test

1 2 3 4 5 6 7 8 Total

Class image

1 234 0 0 0 36 1 0 0 271

2 0 0 0 0 3 0 0 0 3

3 2 0 33 0 53 0 0 0 88

4 2 0 1 0 0 0 0 0 3

5 0 0 2 0 6434 16 0 0 6452

6 1 0 0 0 4 34 0 0 39

7 0 0 0 0 0 0 3 0 3

8 1 0 0 0 1 0 0 0 2

Total 240 0 36 0 6531 51 3 0 6861

Sum of diagonal = 6738

Total = 6861

Overall accuracy = 98.2%

iCAST-ES 2021 - International Conference on Applied Science and Technology on Engineering Science

420

Table 4: Accuracy of LULC maps obtained from 2021

Sentinel-2 satellite data

Matrix

Sample area test

1 2 3 4 5 6 7 8 Total

Class image

1 352 0 1 0 2 0 1 0 356

2 0 0 0 0 0 0 0 0 0

3 1 0 216 0 0 0 0 0 217

4 0 0 0 0 0 0 0 0 0

5 73 12 310 8 35032 1 1 1 35438

6 0 0 0 0 0 169 0 2 171

7 0 0 0 0 0 0 47 0 47

8 0 0 0 0 0 0 0 122 122

Total 426 12 257 8 35034 170 49 125 36351

Sum of diagonal = 35938

Total = 36351

Overall accuracy = 98.8%

4.6 Distribution of LULC Change

The results of the study indicate that all land classes

have changed. The most significant changes occurred

in paddy fields and forests. Medium changes occur in

plantations, water bodies, built-up land, and open

land. While minor changes occurred in fields and

shrubs, as shown in Table 5.

Paddy fields decreased drastically due to several

things: First, changing paddy fields into built-up land.

The existence of new settlements caused by the

increase in population and the development of several

industrial sectors contributed to converting paddy

fields into built-up land. Second, the change of paddy

fields into forests because farmers convert rice plants

into wood plants, namely Sengo (Albizia Chinensis).

Third, the conversion of paddy fields into plantations.

Many farmers in southern Banyuwangi have turned

their fields into dragon fruit plantations (pitaya:

Hylocereus undatus).

The subsequent reduction in the class category is

water bodies and fields. Many water bodies are

located around the coast. The decrease in water

bodies is caused by many ponds being abandoned

because their permits have expired. Then switch

functions into mangrove forests, and some of them

turn into built-up land. The number of fields is

reduced because farmers change their crop

commodities into wood plants, so it is included in the

forest class.

Table 5: The distribution of land cover classification for

each year.

Land use

classes

The coverage area for each year

2013 2021

change 2013-

2021

% km

paddy

field

514.59 14.24 378.71 10.48 -135.88 -3.759

field 0.59 0.02 0.41 0.01 -0.18 -0.005

plantation 76.48 2.12 92.38 2.56 +15.90 +0.440

shrubs 0.27 0.01 1.36 0.04 +1.09 +0.030

forests 2890.74 79.98 3001.95 83.06 +111.21 +3.077

built-up 101.69 2.81 114.04 3.16 +12.35 +0.342

water

odies

29.8 0.82 15.72 0.43 -14.08 -0.390

bare soil 0.18 0.00 9.75 0.27 +9.57 +0.265

Total 3614.34 100 3614.32 100

5 CONCLUSIONS

The study results conclude that the GEE Platform can

perform high-speed spatial analysis for large areas

and large data sets without searching and

downloading satellite data for desktop processing.

The land class can also be appropriately classified

using random forest classification with a very

satisfactory level of accuracy.

Significant land changes occur in paddy fields

and forests. Even though forest land is increasing, the

reduction of lowland rice land must be a severe

concern for the government to secure rice as the

leading staple food for the people of Banyuwangi. In

the future, monitoring of land change must continue

to be carried out by assessing the impact of land

change on environmental carrying capacity and land

criticality in the Banyuwangi Regency.

ACKNOWLEDGEMENTS

We would like to gratefully acknowledge the funding

support from Politeknik Negeri Banyuwangi through

the research master plan scheme.

REFERENCES

Alam, A., Bhat, M. S., & Maheen, M. (2020). Using

Landsat satellite data for assessing the land use and land

cover change in Kashmir valley. GeoJournal, 85(6),

Google Earth Engine for Assessing Land Use and Land Cover Change in Banyuwangi Regency

421

1529–1543. https://doi.org/10.1007/s10708-019-

10037-x

BPS Badan Pusat Statistik. (2020). Banyuwangi dalam

Angka. Retrieved from www.banyuwangikab.bps.go.id

Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S.,

Thau, D., & Moore, R. (2017). Google Earth Engine:

Planetary-scale geospatial analysis for everyone.

Remote Sensing of Environment, 202, 18–27.

https://doi.org/10.1016/j.rse.2017.06.031

He, S., Li, P., Wu, J., & Elumalai, V. (2019). Human and

Ecological Risk Assessment : An International

Groundwater quality under land use / land cover

changes : A temporal study from 2005 to 2015 in.

Human and Ecological Risk Assessment: An

International Journal, 0(0), 1–27.

https://doi.org/10.1080/10807039.2019.1684186

Jr, C. M. S., Shimbo, J. Z., Rosa, M. R., Parente, L. L.,

Alencar, A. A., Rudor, B. F. T., … Oliveira, S. W. De.

(2020). Reconstructing Three Decades of Land Use and

Land Cover Changes in Brazilian Biomes with Landsat

Archive and Earth Engine. Remote Sensing, 12(2735).

Qu, L., Chen, Z., Li, M., Zhi, J., & Wang, H. (2021).

Accuracy Improvements to Pixel-Based and Object-

Based LULC Classification with Auxiliary Datasets

from Google Earth Engine. Remote Sensing, 13(453).

Ramanamurthy, B. V, & Victorbabu, N. (2021). Land Use

Land Cover ( LULC ) classification with wasteland

demarcation using Remote sensing and GIS

Techniques. In IOP Conference Series: Materials

Science and Engineering. International Conference on

Sustainable Systems and Structures (p. 012035). IOP

Publishing. https://doi.org/10.1088/1757-

899X/1025/1/012035

Tamiminia, H., Salehi, B., Mahdianpari, M., Quackenbush,

L., Adeli, S., & Brisco, B. (2020). Google Earth Engine

for geo-big data applications: A meta-analysis and

systematic review. ISPRS Journal of Photogrammetry

and Remote Sensing, 164, 152–170.

https://doi.org/10.1016/j.isprsjprs.2020.04.001

Wijaya, A., & Susetyo, C. (2017). Analisis Perubahan

Penggunaan Lahan di Kota Pekalongan Tahun 2003,

2009, dan 2016. Jurnal Teknik ITS, 6(2), 417–420.

Zhu, C., Lin, Y., Zhang, J., Gan, M., Xu, H., Li, W., …

Wang, K. (2021). Exploring the relationship between

rural transition and agricultural eco-environment using

a coupling analysis : A case study of Zhejiang.

Ecological Indicators, 127.

https://doi.org/10.1016/j.ecolind.2021.107733

Zurqani, H. A., Post, C. J., Mikhailova, E. A., Schlautman,

M. A., & Sharp, J. L. (2018). Geospatial analysis of

land use change in the Savannah River Basin using

Google Earth Engine. International Journal of Applied

Earth Observation and Geoinformation, 69, 175–185.

https://doi.org/10.1016/j.jag.2017.12.006

iCAST-ES 2021 - International Conference on Applied Science and Technology on Engineering Science

422