1 JESE 2 (1) (2022) Journal of Environmental and Science Education http://journal.unnes.ac.id/sju/index.php/jese Land Cover Mapping in Lake Rawa Pening Using Landsat 9 Imagery and Google Earth Engine Trida Ridho Fariz1*, Sapta Suhardono2, Habil Sultan1, Dwi Rahmawati1, Erma Zakiy Arifah1 1Environmental Science, Faculty of Mathematics and Natural Sciences, Universitas Negeri Semarang, Semarang, Indonesia 50229 2Center for Space and Remote Sensing Research (CSRSR), National Central University, Taiwan DOI: https://doi.org/10.15294/jese.v2i1.55851 Article Info ________________ Received 31 March 2022 Accepted 22 April 2021 Published 26 April 2021 ________________ Keywords: Land cover, Google Earth Engine, Machine learning, Landsat 9 ____________________ *Corresponding author: Trida Ridho Fariz Universitas Negeri Semarang E-mail: trida.ridho.fariz@mail.unnes.ac.id Abstract Lake Rawa Pening, in Semarang Regency, is one of the super lakes of revitalization priority. Lake revitalization is an activity to restore the natural function of the lake as a water reservoir through lake dredging, cleaning of invasive alien plants, and land use planning. This makes land use and land cover information in Lake Rawa Pening useful for formulating policy strategies related to revitalization. This study will discuss land cover mapping in Lake Rawa Pening. Mapping using Landsat 9 Imagery and machine learning on Google Earth Engine (GEE). Machine Learning used in this study is CART (Classification and Regression Tree) and RF (Random Forest). The research result shows that the land cover map with the best accuracy is obtained from machine learning RF with an overall accuracy of around 0.78, while CART machine learning is approximately 0.76. The overall accuracy values for CART and RF are not much different because they are both decision tree- based machine learning. This research needs to be developed using cloud masking, comparing image transformations, and comparing its predecessor data, namely Landsat 8. This is useful for providing representative land cover data as the basis for the policy of revitalizing Lake Rawa Pening. ©2022 Universitas Negeri Semarang p-ISSN 2797-0175 e-ISSN 2775-2518 mailto:trida.ridho.fariz@mail.unnes.ac.id Fariz, T.R. et al. / Journal of Environmental and Science Education 2 (2) 2022 2 INTRODUCTION Lake is a large natural water bodies surrounded by land. Indonesia has 840 large lakes and 735 small lakes (Haryani, 2013). Among these lakes, there are several lakes that have severe environmental damage that need to be revitalized. Lake Rawa Pening, in Semarang Regency is one of the super priority lakes handled by the Kementerian Pekerjaan Umum dan Perumahan Rakyat (Indonesian Ministry of Public Works and Housing) along with 5 other lakes. The revitalization of Lake Rawapening with Lake Toba, Lake Tondano, Lake Limboto, Lake Tempe and Lake Maninjau has been carried out in 2020 and will continue in 2021 (Tri, 2021). Lake revitalization is an activity to restore the natural function of the lake as a water reservoir through lake dredging, cleaning of invasive alien plants, making embankments and land use planning in watershed areas (Gobel & Koton, 2017; Mahmud et al, 2020; Suawa et al, 2021). Information on land use and land cover in Lake Rawa Pening is useful for formulating policy strategies related to revitalization. This is because land cover change affects the water quality of Lake Rawa Pening such as sedimentation (Apriliyana, 2015; Soewandita, 2017; Sanjoto et al, 2020). Land cover information can be extracted from satellite imagery by analytical techniques such as visual interpretation or digital interpretation. One platform that provides satellite imagery data and data processing is Google Earth Engine (GEE). GEE is a geo big data application which has access to large data and cloud-based data processing, so that data processing can be carried out without a super computer (Tamiminia et al, 2020, Fariz & Nurhidayati, 2020). GEE also provides machine learning that facilitates data processing such as Random Forest (RF), Classification and Regression Tree (CART) and so on (Arjasakusuma et al, 2020; Fariz et al, 2021). Therefore, this study discusses land cover mapping in Lake Rawa Pening using machine learning in GEE. For satellite imagery, the latest data release is Landsat 9 satellite imagery. The use of Landsat 9 satellite imagery distinguishes it from previous studies such as Hardini et al (2012), Heriza et al (2015) and Bangun et al (2021). METHOD This study is more precisely located in the Rawa Pening Catchment Area, where the area is an upstream part of the Rawapening Watershed sub-watershed. Administratively, the Rawapening catchment area is in Semarang Regency, Central Java Province, Indonesia (Figure 1). Geomorphologically, Rawa Pening is surrounded by mountains that function as water catchments area (Sanjoto et al, 2020). Figure 1. Study location in map Data collection and analysis uses the GEE platform via the code.earthengine.google.com page. The dataset used is USGS Landsat 9 Level 2, Collection 2, Tier 1 recording March 2022.This dataset contains atmospherically corrected surface reflectance from the data produced by the Landsat 9 OLI/TIRS sensors. The bands used are bands 1 to 7 which have the highest accuracy compared to the combination of all bands in the Landsat image (Yu et al, 2019; Fariz & Nurhidayati, 2020). The next stage is supervised classification using machine learning. Machine learning that is compared in this study is RF and CART. The land cover mapped in this study only consists of 6 classes, namely water bodies, forest and agroforestry areas, agricultural areas, open fields, built up and cloud cover. To support the supervised classification process using machine learning, several training samples are needed. The training sample used in this study is 300 Fariz, T.R. et al. / Journal of Environmental and Science Education 2 (2) 2022 3 points. In addition to the classification sample, a test sample was also taken which was used to test the accuracy of the classification results. The number of test sample points in this study is 180 points. RESULT AND DISCUSSION The land cover mapping process using GEE is very rapid because it does not require data download and cloud-based data processing. Based on the supervised classification results which is then performed an interpretation accuracy test shows that the land cover in the Rawa Pening’s water catchment area was classified into 6 (six) land cover classes consisting of water bodies, forest and agroforestry area, agricultural area, open field, built up and cloud cover. Cloud cover is included in the land cover classes because the date of imagery acquisition is during the rainy season with lots of cloud cover. After the land cover classes are determined, the next step is collecting training samples for each land cover class and selecting a machine learning algorithm. The configuration for each machine learning in this study is presented in Table 1. Tabel 1. Configuration for each machine learning in this study Machine learning Expression CART ee.Classifier.smileCart(6, 1) Random forest ee.Classifier.smileRandomForest(10, null, 1, 0.5, null, 0) The accuracy test result shows that the land cover map with the best accuracy is obtained from machine learning RF with an overall accuracy of around 0.78, while CART machine learning is approximately 0.76. The overall accuracy values for CART and RF are not much different because they are both decision tree- based machine learning. This decision tree is similar to the way humans determine a decision, where there is a split process in determining information (Patel & Prajapati, 2018). The decision tree on land cover mapping is made using a splitting class based on the reflectance value of the satellite image (Awaliyan & Sulistioadi, 2018). Therefore, the accuracy of mapping using machine learning can increase if more input data is used (Arjasakusuma et al., 2021). Farda’s research in 2017 became an excellent example because it uses many input data. The use of a lot of input data can be a solution from the mapping results in this study, namely by adding image transformations such as NDVI and NDWI (Loukika et al, 2021). The classification results have accuracy above 0.70, but there are still misclassifications, such as forest classified as water bodies (Figure 2). Future Work: Land Cover Mapping in Rawa Pening Misclassification of mapping results in this study is probably due to data quality factors. The quality of the data is the presence of clouds and cloud shadows, limiting the use of the obtained images and misclassification (Kalkan & Maktav, 2018). In our opinion, that is the cause of the emergence of misclassification of objects into water bodies or forest objects. The solution to that problem is using cloud masking with multitemporal approaches (Mateo-García et al., 2018). Fariz, T.R. et al. / Journal of Environmental and Science Education 2 (2) 2022 4 Figure 2. Land cover mapping results based on RF Figure 3. Comparison of Lake Rawa Pening from Landsat 8 and Landsat 9 The use of Landsat 9 in this study was also not fully explored, such as comparing band combinations or image transformations. This study needs to be developed by comparing band combinations or image transformations and corresponding with the previous dataset, namely Landsat 8. Landsat 9 carries the same instruments as the Landsat 8 satellite but with some improvement, namely a slightly better signal-to-noise ratio than Landsat 8 (USGS, 2022). There are differences in appearance between Landsat 8 and Landsat 9, where Landsat 9 imagery looks clearer (Figure 3). This study helps provide representative land cover data as the basis for the policy of revitalizing Lake Rawa Pening. CONCLUSION Machine learning in GEE is very helpful in land cover rapid mapping, the highest accuracy for land cover mapping in Lake Rawa Pening is Fariz, T.R. et al. / Journal of Environmental and Science Education 2 (2) 2022 5 RF (78% overall accuracy). CART also has an overall accuracy not much different from RF, although the two mapping results still have misclassifications. This research needs to be developed using cloud masking, comparing image transformations and comparing its predecessor data, namely Landsat 8. This is useful for providing representative land cover data as the basis for the policy of revitalizing Lake Rawa Pening. REFERENCES Aprilliyana, D. (2015). Pengaruh perubahan penggunaan lahan sub das rawapening terhadap erosi dan sedimentasi Danau Rawapening. Jurnal Pembangunan Wilayah dan Kota, 11(1), 103-116. Arjasakusuma, S., Kusuma, S. S., Saringatin, S., Wicaksono, P., Mutaqin, B. W., & Rafif, R. (2021). Shoreline Dynamics in East Java Province, Indonesia, from 2000 to 2019 Using Multi-Sensor Remote Sensing Data. Land, 10(2), 100. Arjasakusuma, S., Swahyu Kusuma, S., Rafif, R., Saringatin, S., & Wicaksono, P. (2020). Combination of Landsat 8 OLI and Sentinel- 1 SAR Time-Series Data for Mapping Paddy Fields in Parts of West and Central Java Provinces, Indonesia. ISPRS International Journal of Geo-Information, 9(11), 663. Awaliyan, M. R., & Sulistioadi, Y. B. (2018). Klasifikasi Penutupan Lahan Pada Citra Satelit Sentinel-2A Dengan Metode Tree Algorithm. ULIN: Jurnal Hutan Tropis, 2(2). Bangun, Y., Simanjuntak, B. H., & Sutrisno, A. J. Analisis Perubahan Penggunaan Lahan Sawah dari Tahun 2008–2018. LaGeografia, 19(3), 302-315. Farda, N. M. (2017). Multi-temporal land use mapping of coastal wetlands area using machine learning in Google earth engine. IOP Conference Series: Earth and Environmental Science (Vol. 98, No. 1, p. 012042). IOP Publishing. Fariz, T. R., & Nurhidayati, E. (2020). Mapping Land Coverage in the Kapuas Watershed Using Machine Learning in Google Earth Engine. Journal of Applied Geospatial Information, 4(2), 390-395. Fariz, T. R., Daeni, F., & Sultan, H. (2021). Pemetaan Perubahan Penutup Lahan Di Sub-DAS Kreo Menggunakan Machine Learning Pada Google Earth Engine. Jurnal Sumberdaya Alam dan Lingkungan, 8(2), 85- 92. Gobel, E. Z., & Koton, Y. P. (2017). Pengelolaan Danau Limboto dalam Perspektif Implementasi Kebijakan Publik. Deepublish Hardini, A. R., Helmi, M., & Sasmito, B. (2012). Analisa Spasial Dinamika Morfometri Waduk Menggunakan Data Satelit Multi Temporal Di Waduk Rawa Pening Provinsi Jawa Tengah. Jurnal Geodesi Undip, 1(1). Haryani, G. S. (2013). Kondisi Danau di Indonesia dan Strategi Pengelolaannya. Prosiding Pertemuan Ilmiah Tahunan MLI I, 2, 1–19. Heriza, D., Sukmono, A., & Bashit, N. (2018). Analisis perubahan kualitas perairan Danau Rawa Pening periode 2013, 2015 dan 2017 dengan menggunakan data citra Landsat 8 multitemporal. Jurnal Geodesi Undip, 7(1), 79-89. Kalkan, K., & Maktav, M. D. (2018). A cloud removal algorithm to generate cloud and cloud shadow free images using information cloning. Journal of the Indian Society of Remote Sensing, 46(8), 1255-1264. Loukika, K. N., Keesara, V. R., & Sridhar, V. (2021). Analysis of Land Use and Land Cover Using Machine Learning Algorithms on Google Earth Engine for Munneru River Basin, India. Sustainability, 13(24), 13758. Mahmud, S. L., Achmad, N., & Panigoro, H. S. (2020). Revitalisasi Danau Limboto dengan Pengerukan Endapan di Danau: Pemodelan, Analisis, dan Simulasinya. Jambura Journal of Biomathematics (JJBM), 1(1), 31-40. Patel, H. H., Prajapati, P. (2018). Study and analysis of decision tree based classification algorithms. International Journal of Computer Sciences and Engineering, 6(10), 74-78. Sanjoto, T. B., Sidiq, W. A. B. N., & Nugraha, S. B. (2020). Land cover change analysis to sedimentation rate of Rawapening Lake. GEOMATE Journal, 18(70), 294-301. Soewandita, H. (2017). Studi Ekologi Lahan Koridor Sungai dan Status Kualitas Penggunaan Lahan di Wilayah DAS Rawapening. Jurnal Alami: Jurnal Teknologi Reduksi Risiko Bencana, 1(1), 33-42. Suawa, P. J., Pioh, N. R., & Waworundeng, W. (2021). Manajemen Pengelolaan Dana Revitalisasi Danau Tondano Oleh Pemerintah Kabupaten Minahasa (Studi Fariz, T.R. et al. / Journal of Environmental and Science Education 2 (2) 2022 6 Kasus Di Balai Wilayah Sungai Sulawesi). GOVERNANCE, 1(2). 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 Tri, T. 2021. Kembalikan Fungsi Tampungan Air, Kementerian PUPR Lanjutkan Penanganan 8 Danau Kritis di Indonesia. Accessed from pu.go.id. USGS. (2022). What sensors does the Landsat 9 satellite carry? Accessed from Yu, Z., Di, L., Yang, R., Tang, J., Lin, L., Zhang, C., ... & Sun, Z. (2019). Selection of Landsat 8 OLI Band Combinations for Land Use and Land Cover Classification. In 2019 8th International Conference on Agro- Geoinformatics (Agro-Geoinformatics) (pp. 1- 5). IEEE