Show simple item record

dc.contributor.advisorAidi, Muhammad Nur
dc.contributor.advisorSartono, Bagus
dc.contributor.advisorSyafitri, Utami Dyah
dc.contributor.advisorPranata, Bayu
dc.contributor.authorSirodj, Dwi Agustin Nuriani
dc.date.accessioned2026-11-02T06:13:14Z
dc.date.available2026-11-02T06:13:14Z
dc.date.issued2026
dc.identifier.urihttp://repository.ipb.ac.id/handle/123456789/180232
dc.description.abstractPendugaan nilai pada lokasi yang tidak teramati merupakan persoalan penting dalam analisis spasial. Ordinary kriging mempertimbangkan ketergantungan spasial melalui semivariogram, tetapi ketepatan prediksinya bergantung pada kemampuan model semivariogram merepresentasikan struktur spasial data. Model Gaussian, Eksponensial, dan Spherical memiliki bentuk teoritis tertentu sehingga kurang fleksibel ketika semivariogram empiris menunjukkan hubungan nonlinier atau pola tidak monoton. Permasalahan ini relevan pada kegempaan Sumatra yang dipengaruhi zona subduksi, Sumatran Fault System, dan Mentawai Fault System, sehingga membentuk struktur seismisitas yang heterogen. Dua parameter penting dalam karakterisasi seismisitas adalah a-value dan b-value pada hubungan Gutenberg–Richter. A-value menggambarkan tingkat aktivitas seismik suatu wilayah, sedangkan b-value merepresentasikan tingkat kerapuhan batuan. Kedua parameter tersebut hanya dapat dihitung pada wilayah yang memiliki informasi kejadian gempa yang memadai sehingga distribusi spasialnya tidak selalu tersedia secara kontinu. Oleh karena itu, interpolasi diperlukan untuk menduga a-value dan b-value pada lokasi yang tidak teramati agar diperoleh gambaran spasial parameter seismisitas yang lebih kontinu dan mampu menunjukkan keragaman karakteristik kegempaan antarwilayah. Pengembangan metode dilakukan melalui pemodelan semivariogram nonparametrik berbasis Support Vector Regression (SVR) yang diintegrasikan dengan Ordinary kriging. Keragaman lokal yang belum mampu dijelaskan oleh model utama, karena model tersebut lebih merepresentasikan struktur spasial global, selanjutnya dikoreksi melalui dua pendekatan, yaitu Hybrid SVR–Ordinary kriging–IDW Residual yang menggunakan residual bertanda dan Hybrid SVR–Ordinary kriging–IDW Absolut Residual yang menggunakan nilai absolut residual sebagai koreksi positif terhadap prediksi dasar. Koreksi Deutsch diterapkan pada tahap kriging untuk mengurangi pengaruh bobot negatif yang berlebihan dan menjaga kestabilan prediksi. Kinerja metode dievaluasi pada data empiris dan simulasi menggunakan Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Root Mean Square Error of Prediction (RMSEP), keragaman residual, dan Moran’s I. Penelitian ini bertujuan untuk (1) mengidentifikasi pola kejadian gempa, dan segmentasi seismik di Sumatra, (2) mengevaluasi kinerja SVR dalam pendugaan semivariogram nonparametrik yang diintegrasikan dengan Ordinary kriging serta membandingkannya dengan Inverse Distance Weighting (IDW) dan Ordinary kriging berbasis semivariogram parametrik, (3) mengembangkan dan mengevaluasi metode Hybrid SVR–Ordinary kriging–IDW dengan koreksi residual bertanda dan residual absolut, serta (4) mengevaluasi kinerja SVR–Ordinary kriging dan Hybrid SVR–Ordinary kriging–IDW pada data simulasi berdasarkan keragaman pola respons dan ukuran contoh. Data yang digunakan dalam penelitian ini terdiri atas data empiris dan data simulasi. Data empiris berasal dari katalog International Seismological Centre dan Badan Meteorologi, Klimatologi, dan Geofisika (BMKG) periode 2000–2023. Sementara itu, data simulasi dibangkitkan berdasarkan semivariogram teoritis Gaussian yang selanjutnya dimodifikasi untuk menghasilkan dua pola respons, yaitu tidak berkelompok dan berkelompok, dengan ukuran contoh 64, 225, dan 625 serta 100 ulangan untuk setiap skenario. Nearest Neighbor Index menunjukkan pola gempa di Sumatra mengelompok, sedangkan Density-Based Spatial Clustering of Applications with Noise (DBSCAN) mengidentifikasi dua kelompok utama, yaitu Sumatra selatan–tengah dan Sumatra utara. Semivariogram empiris a-value dan b-value menunjukkan pola tidak monoton, dengan SVR menghasilkan RMSE lebih rendah daripada model Gaussian. Namun, residual SVR–Ordinary Kriging masih menunjukkan autokorelasi spasial positif dan signifikan. Koreksi melalui Hybrid SVR–Ordinary Kriging–IDW Residual memberikan kinerja terbaik dengan residual yang lebih kecil dan Moran’s I mendekati nol, sedangkan Hybrid Absolute Residual belum meningkatkan ketepatan karena cenderung menghasilkan prediksi berlebihan. Pada data simulasi, peningkatan ukuran contoh dari 64, 225, hingga 625 secara umum menurunkan RMSE dan MAE serta meningkatkan kestabilan prediksi. Gaussian–Ordinary kriging dan SVR–Ordinary kriging lebih kompetitif pada pola respons tidak berkelompok, khususnya pada ukuran contoh besar, sedangkan Hybrid SVR–Ordinary kriging–IDW Residual memberikan kinerja lebih baik pada pola respons berkelompok ketika residual model dasar masih menunjukkan ketergantungan spasial. Peta interpolasi menggunakan Hybrid SVR–Ordinary kriging–IDW Residual memperlihatkan a-value dan b-value yang relatif tinggi di Nias–Simeulue dan lebih rendah di Mentawai–Pagai–Enggano. Nilai rendah di Mentawai dapat berkaitan dengan penguncian dan akumulasi tegangan pada megathrust, yaitu zona patahan besar pada batas pertemuan lempeng di zona subduksi, serta mengindikasikan adanya seismic gap, yaitu segmen aktif secara tektonik yang menunjukkan aktivitas gempa relatif rendah dalam periode tertentu dibandingkan wilayah sekitarnya dan berpotensi masih menyimpan akumulasi tegangan. Interpretasi tersebut perlu mempertimbangkan kemungkinan slow slip event (SSE), yaitu pelepasan regangan secara perlahan dan aseismik pada bidang kontak antarlempeng, karena pelepasan regangan melalui SSE pada sebagian megathrust tidak berarti seluruh segmen telah melepaskan akumulasi tegangannya. Secara keseluruhan, semivariogram nonparametrik berbasis SVR lebih fleksibel dalam merepresentasikan struktur spasial kompleks. Integrasinya dengan Ordinary kriging dan koreksi residual bertanda melalui Hybrid SVR–Ordinary kriging–IDW mampu meningkatkan ketepatan prediksi dan mengurangi ketergantungan spasial residual, terutama pada pola respons berkelompok.
dc.description.abstractEstimation at unobserved locations is an important problem in spatial analysis. Ordinary Kriging accounts for spatial dependence through the semivariogram; however, its predictive accuracy depends on how well the semivariogram model represents the underlying spatial structure. Gaussian, exponential, and spherical models have predefined theoretical forms and may therefore lack flexibility when the empirical semivariogram exhibits nonlinear relationships or non-monotonic patterns. This issue is particularly relevant to seismicity in Sumatra, which is influenced by the subduction zone, the Sumatran Fault System, and the Mentawai Fault System, resulting in heterogeneous seismicity patterns. Two important parameters for characterizing seismicity are the a-value and b-value of the Gutenberg–Richter relationship. The a-value represents the level of seismic activity in a region, whereas the b-value reflects rock brittleness. These parameters can only be estimated in areas with sufficient earthquake observations; consequently, their spatial distributions are not continuously available. Spatial interpolation is therefore required to estimate a-values and b-values at unobserved locations, providing a more continuous representation of seismicity parameters and revealing spatial variations in seismic characteristics across regions. The methodological development employs a nonparametric semivariogram modeled using Support Vector Regression (SVR) and integrated with Ordinary Kriging. Local variations that remain unexplained by the primary model, which predominantly represents the global spatial structure, are subsequently corrected using two approaches: Hybrid SVR–Ordinary Kriging–IDW Residual, which interpolates signed residuals, and Hybrid SVR–Ordinary Kriging–IDW Absolute Residual, which uses absolute residuals as a positive correction to the baseline predictions. The Deutsch correction is applied at the kriging stage to mitigate the excessive influence of negative weights and improve prediction stability. Model performance is evaluated using empirical and simulated data based on Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Root Mean Square Error of Prediction (RMSEP), residual variability, and Moran’s I. This study aims to (1) identify earthquake occurrence patterns and seismic segmentation in Sumatra, (2) evaluate the performance of SVR in fitting a nonparametric semivariogram integrated with Ordinary Kriging and compare it with Inverse Distance Weighting (IDW) and Ordinary Kriging based on parametric semivariogram models, (3) develop and evaluate Hybrid SVR–Ordinary Kriging–IDW with signed and absolute residual corrections, and (4) evaluate the performance of SVR–Ordinary Kriging and Hybrid SVR–Ordinary Kriging–IDW using simulated data under varying response patterns and sample sizes. The data consist of empirical and simulated datasets. The empirical data were obtained from the International Seismological Centre and the Indonesian Agency for Meteorology, Climatology, and Geophysics (BMKG) earthquake catalogs for the 2000–2023 period. The simulated data were generated from a theoretical Gaussian semivariogram and subsequently modified to produce two response patterns, namely unclustered and clustered patterns, with sample sizes of 64, 225, and 625 and 100 replications for each scenario. The Nearest Neighbor Index revealed a clustered spatial distribution of earthquakes in Sumatra, while Density-Based Spatial Clustering of Applications with Noise (DBSCAN) identified two major clusters in south–central and northern Sumatra. The empirical semivariograms of the a-value and b-value exhibited non-monotonic patterns, for which SVR achieved lower RMSE values than the Gaussian model. However, significant positive spatial autocorrelation remained in the SVR–Ordinary Kriging residuals, indicating unresolved local spatial structure. The Hybrid SVR–Ordinary Kriging–IDW Residual method effectively reduced this residual dependence, yielding smaller residual errors and Moran’s I values close to zero. In contrast, the Hybrid Absolute Residual approach did not improve predictive accuracy because its positive-only correction tended to produce overprediction. For the simulated data, increasing the sample size from 64 to 225 and 625 generally reduces RMSE and MAE and improves prediction stability. Gaussian–Ordinary Kriging and SVR–Ordinary Kriging are more competitive for unclustered response patterns, particularly at larger sample sizes, whereas Hybrid SVR–Ordinary Kriging–IDW Residual performs better for clustered response patterns when the residuals of the baseline model retain spatial dependence. The interpolation maps generated using the Hybrid SVR–Ordinary Kriging–IDW Residual method show relatively high a-values and b-values in the Nias–Simeulue region and lower values in Mentawai–Pagai–Enggano. The low values observed in Mentawai may be associated with locking and stress accumulation along the megathrust, a major fault zone at the convergent plate boundary within a subduction zone, and may indicate a seismic gap, i.e., a tectonically active segment exhibiting relatively low seismic activity over a certain period compared with surrounding regions and potentially retaining accumulated stress. This interpretation should also consider the occurrence of slow slip events (SSEs), which involve slow, aseismic strain release along the plate interface, as strain release through SSEs along part of the megathrust does not necessarily imply that the entire segment has released its accumulated stress. Overall, the SVR-based nonparametric semivariogram provides greater flexibility in representing complex spatial structures. Its integration with Ordinary Kriging and signed residual correction through the Hybrid SVR–Ordinary Kriging–IDW framework improves predictive accuracy and reduces residual spatial dependence, particularly for clustered response patterns.
dc.description.sponsorshipBalai Pembiayaan Pendidikan Tinggi (BPPT), Kementerian Pendidikan Tinggi, Sains, dan Teknologi atas dana pendidikan melalui Beasiswa Pendidikan Indonesia Tahun 2022 berdasarkan Kontrak Nomor 02858/J5.2.3./BPI.06/10/2022.
dc.language.isoid
dc.publisherIPB Universityid
dc.titlePengembangan Interpolasi Spasial Hybrid Berbasis Semivariogram Nonparametrik Melalui SVR–Ordinary Kriging–IDW dengan Koreksi Residual untuk Parameter Seismisitas Sumatraid
dc.title.alternativeEnhancing Hybrid Spatial Interpolation Based on a Nonparametric Semivariogram using SVR–Ordinary Kriging–IDW with Residual Correction for Seismicity Parameters in Sumatra
dc.typeDisertasi
dc.subject.keywordInterpolasi spasialid
dc.subject.keywordordinary krigingid
dc.subject.keywordParameter Seismisitasid
dc.subject.keywordSemivariogram Nonparametrikid
dc.subject.keywordSupport Vector Regressionid
dc.subtypeDissertations


Files in this item

Thumbnail
Thumbnail
Thumbnail

This item appears in the following Collection(s)

Show simple item record