Kajian Perbandingan Kinerja Algoritma Klasifikasi Non-Parametrik Pada Citra Landsat Pansharpened Untuk Analisis Kesesuaian Penggunaan Lahan Terhadap Tata Ruang Kabupaten Bangka
RUTH ADE PUTRI, Drs. Projo Danoedoro, M.Sc., Ph.D., Dr. Bowo Susilo, S.Si., M.T.
2026 | Tesis | S2 Penginderaan Jauh
Dinamika perubahan penggunaan lahan yang ekstensif di Kabupaten Bangka memerlukan pemantauan berkala, serta evaluasi kesesuaian terhadap RTRW agar sejalan dengan arah pembangunan wilayah yang berkelanjutan. Informasi penggunaan lahan berperan penting dalam mengidentifikasi pola perubahan spasial serta dampaknya terhadap perencanaan wilayah. Dalam konteks tersebut, penerapan teknik pansharpening pada citra penginderaan jauh dan algoritma klasifikasi data citra berpotensi meningkatkan ketelitian pemetaan penggunaan lahan pada skala yang lebih detail, namun kinerjanya perlu dievaluasi secara komprehensif. Penelitian ini bertujuan untuk (a) mengevaluasi dan membandingkan hasil penerapan berbagai teknik pansharpening pada citra Landsat 8 OLI untuk menentukan metode yang paling akurat dalam meningkatkan resolusi spasial dengan tetap mempertahankan kualitas spektral citra; (b) menentukan kinerja algoritma klasifikasi non-parametrik terbaik berdasarkan nilai akurasi hasil klasifikasi penggunaan lahan tahun 2017 dan 2024 di sebagian wilayah Kabupaten Bangka; (c) mengkaji perubahan spasial penggunaan lahan periode 2017–2024 dan menganalisis kesesuaian penggunaan lahan tahun 2024 terhadap RTRW Kabupaten Bangka.
Tujuan pertama dan kedua dicapai melalui proses penajaman citra menggunakan metode Gram–Schmidt (GS), NNDiffuse, dan Principal Component Analysis (PCA), yang dievaluasi secara visual kualitatif, serta kuantitatif menggunakan metrik Koefisien Korelasi (CC), Spectral Angle Mapping (SAM), Root Mean Square Error (RMSE), dan Erreur Relative Globale Adimensionnelle de Synthèse (ERGAS). Selanjutnya, citra pansharpened diklasifikasikan menggunakan algoritma non-parametrik yaitu Random Forest (RF), dan Multilayer Perceptron (MLP), serta Maximum Likelihood Classification (MLC) sebagai pembanding kinerja pada algoritma parametrik. Penilaian akurasi diproses berdasarkan Overall Accuracy (OA), Koefisien Kappa, dan F1-Score. Analisis perubahan penggunaan lahan dilakukan dengan pendekatan post-classification comparison menggunakan matriks transisi, sedangkan analisis kesesuaian penggunaan lahan tahun 2024 terhadap RTRW dilakukan melalui analisis spasial tumpang susun.
Hasil penelitian menunjukkan bahwa teknik NNDiffuse merupakan metode pansharpening terbaik dengan nilai CC tertinggi sebesar 0,9980 (2017) dan 0,9969 (2024), ERGAS terendah yaitu 5,6497 (2017) dan 4,5117 (2024), SAM terkecil yaitu 0,0396 radian (2017) dan 0,0547 radian (2024), dan RMSE <0>
The extensive dynamics of land use change in Bangka Regency require periodic monitoring and evaluation of land use suitability against the Regional Spatial Plan (RTRW) to ensure alignment with sustainable regional development objectives. Land use information plays a crucial role in identifying spatial change patterns and their impacts on regional planning. The application of pansharpening techniques and image classification algorithms has the potential to improve the accuracy of land use mapping at more detailed spatial scales; however, their performance needs to be comprehensively evaluated. This study aims to: (a) evaluate and compare the performance of various pansharpening techniques applied to Landsat 8 OLI imagery to determine the most accurate method for enhancing spatial resolution while preserving spectral quality; (b) determine the best-performing land use classification algorithms (Random Forest (RF), Multilayer Perceptron (MLP), and Maximum Likelihood Classifier (MLC) based on classification accuracy for the years 2017 and 2024 in part of Bangka Regency; and (c) analyze spatial land use changes between 2017 and 2024 and assess the suitability of land use in 2024 with respect to the Bangka Regency RTRW.
The first and second objectives were achieved through an image sharpening process using the Gram–Schmidt (GS), NNDiffuse, and Principal Component Analysis (PCA) methods. These methods were evaluated qualitatively through visual assessment and quantitatively using the Correlation Coefficient (CC), Spectral Angle Mapping (SAM), Root Mean Square Error (RMSE), and Erreur Relative Globale Adimensionnelle de Synthèse (ERGAS) metrics. Subsequently, the pansharpened images were classified using non-parametric algorithms (RF and MLP), with MLC employed as a performance benchmark for parametric algorithms. Classification accuracy was assessed based on Overall Accuracy (OA), the Kappa Coefficient, and the F1-Score. Land-use change analysis was conducted using a post-classification comparison approach with transition matrices, while the 2024 land-use suitability analysis against the RTRW was performed through spatial overlay analysis.
The results indicate that NNDiffuse is the best-performing pansharpening technique, achieving the highest CC values of 0,9980 (2017) and 0,9969 (2024), the lowest ERGAS values of 5,6497 (2017) and 4,5117 (2024), the smallest SAM values of 0,0396 radian (2017) and 0,0547 radian (2024), and RMSE values below 0,0200 for both image acquisition years. The integration of NNDiffuse pansharpened imagery with topographic variables using the MLP algorithm yielded the best classification performance, with OA values of 91,16% (2017) and 91,97% (2024), Kappa coefficients of 0,8849 and 0,8891, and F1-Scores of 85,52% and 88,96%, respectively. Land-use dynamics from 2017 to 2024 show significant net increases in the Mixed Gardens and Cropland class (161,71 km²), Plantations (113,68 km²), and Grassland (70,02 km²), while net decreases occurred in Shrub Forest (?155,49 km²) and Open-Pit Mining Land (?34,86 km²). The land-use suitability analysis for 2024 indicates that 55,53% of the study area falls within the “Suitable” category, 34,68% within the “Supporting” category, and 9,79% is classified as “Unsuitable”.
Kata Kunci : pansharpening, Random Forest, Multilayer Perceptron, perubahan penggunaan lahan, kesesuaian tata ruang
