Safira, Rizka Amelia Dwi (2024) Monitoring Pola Hidrologi Menggunakan Data Satelit Gayaberat GRACE dan GRACE-FO Untuk Identifikasi Potensi Bencana Hidrometeorologi di Indonesia. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6016231011-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6016231011-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (18MB) | Request a copy |
Abstract
Terrestrial water storage (TWS) merupakan indikator yang efektif dalam memantau bencana hidrometeorologi, seperti kekeringan dan banjir, serta dalam pengelolaan sumber daya air berkelanjutan untuk memahami dinamika air di Bumi. Satelit Gravity Recovery and Climate Experiment (GRACE) mengidentifikasi dinamika ini melalui perekaman anomali massa bumi sejak tahun 2002. Namun, keterbatasan resolusi spasial GRACE yang kasar (~150.000 km2) dan kekosongan data sebelum GRACE-FO beroperasi, menyebabkan diskontinuitas dan ketidakpastian dalam prediksi pola hidrologi, seperti untuk investigasi skala lokal, terutama untuk negara kepulauan seperti Indonesia. Penelitian ini mengaplikasikan algoritma extreme gradient boosting (XGBoost) dan random forest (RF) untuk mengisi kekosongan data GRACE/GRACE-FO serta meningkatkan resolusi spasialnya (dari 0,5° menjadi 0,25°) untuk mengamati variabilitas tren dan potensi banjir serta kekeringan di Jawa dan Kalimantan selama periode 2003-2022, dengan mengintegrasikan variabel hidro-klimatik. Banjir dan kekeringan menyumbang 97,12% dari seluruh kejadian bencana alam di Indonesia, khususnya di Pulau Jawa, sebagai Pulau dengan kepadatan penduduk tertinggi, dan di Kalimantan dengan kondisi geografis dan demografis berbeda serta perhatian sebagai lokasi pemindahan ibu kota. Untuk pemodelan gap-filling, tiga metrik statistik yang mengevaluasi testing set (20%) menghasilkan koefisien korelasi (CC), Nash–Sutcliffe efficiency (NSE), dan Root mean square error terskala (RMSE*) sebesar 0,907; 0,844; 0,395 (87,50% (CC ≥ 0,8), 74,31% (NSE ≥ 0,7), dan 70,14% (RMSE* ≤ 0,5) dari total piksel memiliki kinerja yang "Sangat Baik" untuk evaluasi skala grid) menggunakan algoritma XGBoost. Pada pemodelan downscaling, performa CC, NSE, dan RMSE* untuk keseluruhan testing set (20%) adalah 0,927; 0,858; dan 0,377. Korelasi antara hasil downscaling terhadap data observasi tinggi muka air di DAS di Jawa mencapai 0,686, dibandingkan dengan prediksi TWSA resolusi 0,5° sebesar 0,683, menunjukkan bahwa prediksi TWSA dengan resolusi lebih halus mampu menangkap detail spasial dan temporal lebih baik. Identifikasi potensi banjir dan kekeringan menggunakan flood potential index (FPI) dan water storage deficit index (WSDI) menunjukkan adanya 86 kekeringan hidrologis di wilayah kajian, sementara FPI mampu mendeteksi kejadian banjir Kalimantan Selatan pada Januari 2021, Jawa pada Maret 2008, dan Barito dan Mahakam pada Mei 2007, dengan FPI sebesar 0,754; 0,859; dan 0,789, yang merupakan banjir dengan durasi ≥ 15 hari. Perbedaan karakteristik lokasi geografis, luas basin, dan topografi di Pulau Jawa dan Kalimantan memberikan tren TWSA yang bervariasi dalam mengamati bencana hidrometeorologis dan strategi pengelolaan airnya.
========================================================================================================
Terrestrial water storage (TWS) is an effective indicator in monitoring hydrometeorological events, such as droughts and floods, and in sustainable water resources management to understand the dynamics of water on Earth. The Gravity Recovery and Climate Experiment (GRACE) satellite has been identifying these dynamics through recording Earth mass anomalies since 2002. However, the limitations of GRACE's coarse spatial resolution (~150,000 km2) and data gaps before GRACE-FO became operational, led to discontinuities and uncertainties in the prediction of hydrological patterns, such as for local-scale investigations, especially for archipelagic countries like Indonesia. This study applies extreme gradient boosting (XGBoost) and random forest (RF) algorithms to fill the GRACE/GRACE-FO data gaps and increase its spatial resolution (from 0.5° to 0.25°) to observe the variability of flood and drought trends and potential in Java and Kalimantan during the period 2003-2022, integrating hydro-climatic variables. Floods and droughts account for 97.12% of all natural disasters in Indonesia, especially in Java, the island with the highest population density, and in Kalimantan with different geographical and demographic conditions and attention as the location of the relocation of the capital city. For gap-filling modelling, three statistical metrics evaluating the testing set (20%) yielded a correlation coefficient (CC), Nash-Sutcliffe efficiency (NSE), and scaled root mean square error (RMSE*) of 0.907; 0.844; 0.395 (87.50% (CC ≥ 0.8), 74.31% (NSE ≥ 0.7), and 70.14% (RMSE* ≤ 0.5) of the total pixels have "Excellent" performance for grid-scale evaluation) using the XGBoost algorithm. In downscaling modelling, the performance of CC, NSE, and RMSE* for the whole testing set (20%) were 0.927; 0.858; and 0.377. The correlation between the downscaling results and observational data of water levels in watersheds in Java reached 0.686, compared to the 0.5° resolution TWSA prediction of 0.683, indicating that the finer resolution TWSA prediction was able to capture spatial and temporal details better. Identification of flood and drought potential using the flood potential index (FPI) and water storage deficit index (WSDI) indicated the presence of 86 hydrological droughts in the study area, while the FPI was able to detect flood events in South Kalimantan in January 2021, Java in March 2008, and Barito and Mahakam in May 2007, with FPIs of 0.754; 0.859; and 0.789, which were floods with a duration ≥ 15 days. The different characteristics of geographical location, basin size, and topography in Java and Kalimantan provide varied TWSA trends in observing hydrometeorological disasters and their water management strategies.
Actions (login required)
 |
View Item |