Please use this identifier to cite or link to this item:
http://repository.ipb.ac.id/handle/123456789/119326| Title: | Deteksi Kerusakan dan Revegetasi Pasca Kebakaran Hutan dan Lahan dengan Pendekatan Decision-Tree Machine Learning: Studi Kasus di Jambi |
| Authors: | Jaya, I Nengah Surati Rizkiana, Ridha |
| Issue Date: | 15-Jun-2023 |
| Publisher: | IPB University |
| Abstract: | Informasi mengenai kondisi tutupan hutan dan lahan pasca kebakaran
merupakan salah satu bagian penting dalam kegiatan pemantauan setelah terjadinya
kebakaran hutan. Penelitian ini bertujuan untuk menentukan fitur, atribut, dan
parameter dalam mendeteksi kerusakan dan revegetasi akibat kebakaran berbasis
pada citra multiwaktu dengan citra sintetik menggunakan pendekatan Pohon
Keputusan (Decision Tree) Pembelajar Mesin (Machine Learning). Penelitian ini
menemukan bahwa parameter paling akurat adalah Gini Index pada K1 dengan
kombinasi pemangkasan (pruning), tanpa pra-pangkas (pre-pruning), kedalaman
pohon sebesar 60, pra-pangkas alternatif (pre-pruning alternative) sebesar 50,
sampling split dengan stratified sampling, dan ukuran daun sebesar 41 atau K2
dengan kombinasi tanpa pemangkasan (without pruning), tanpa pra-pangkas (pre
pruning), kedalaman pohon sebesar 80, pra-pangkas alternatif (pre-pruning
alternative) sebesar 30, sampling split dengan automatic sampling, dan ukuran
daun sebesar 100. Kajian ini memberikan overall akurasi (OA) sebesar 95,7% dan
kappa accuracy (KA) sebesar 94,6%. Dari tujuh citra indeks yang diuji dengan
pohon keputusan, ditemukan tiga indeks yang paling signifikan pada BNDVI (Blue
Normalized Differenced Vegetation Index), NBBI (Normalized Built-up & Bare
Land Index), dan NDWIG (Normalized Differenced Wetness Index, green-based). Information on post-fire forest and land cover conditions is important to monitoring activities after forest fires. The study’s objective is to determine features, attributes, and parameters in detecting damage and vegetation growth due to fires based on multi-temporal images with synthetic images using a Machine Learning approach. This study found that the most accurate parameters were the Gini Index at K1 with a combination of pruning, no pre-pruning, tree depth of 60, pre-pruning alternative of 50, sampling split with stratified sampling, and leaf size of 41 or K2 with a combination of no pruning, no pre-pruning, tree depth of 80, pre-pruning alternative by 30, sampling split by automatic sampling, and leaf size by 100. Overall accuracy (OA) gives a value of 95,7% and kappa accuracy (KA) of 94,6%. The classification results of the seven index images tested with decision trees obtained the three most significant indices on BNDVI (Blue Normalized Difference Vegetation Index), NBBI (Normalized Built-up &; Bare Land Index), and NDWIG (Normalized Differenced Wetness Index, green-based). |
| URI: | http://repository.ipb.ac.id/handle/123456789/119326 |
| Appears in Collections: | UT - Forest Management |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| Cover_Ridha Rizkiana (E14180080).pdf Restricted Access | Cover | 1.42 MB | Adobe PDF | View/Open |
| Lampiran_Ridha Rizkiana (E14180080).pdf Restricted Access | Lampiran | 343.04 kB | Adobe PDF | View/Open |
| Full Text_Ridha Rizkiana (E14180080).pdf Restricted Access | Full Text | 3.53 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.