Prediksi Wilayah Rawan Kebakaran Hutan di Provinsi Riau Menggunakan Metode Jaringan Saraf Tiruan

Pamungkas, Adhi Yoga Muris (2020) Prediksi Wilayah Rawan Kebakaran Hutan di Provinsi Riau Menggunakan Metode Jaringan Saraf Tiruan. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
05211640000073-Undergraduate_Theses.pdf - Accepted Version
Restricted to Repository staff only

Download (3MB) | Request a copy

Abstract

Kejadian kebakaran hutan di Indonesia memiliki dampak yang signifikan terhadap kualitas udara dan kesehatan di Indonesia. Kebakaran hutan dianggap sebagai ancaman potensial bagi pembangunan berkelanjutan karena memberikan dampak langsung terhadap keseimbangan ekosistem, peningkatan emisi karbon, dan peningkatan keanekaragaman hayati. Pada tahun 2019, dari total 857 ha luas lahan hutan yang terbakar dan tersebar di enam provinsi, 75,871 ha terjadi di Provinsi Riau. Sebagai upaya untuk mengurangi kemungkinan terjadinya lahan hutan yang terbakar, maka diperlukan model prediksi guna mengantisipasi terjadinya kebakaran hutan tersebut. Dalam tugas akhir ini, prediksi kemungkinan terjadinya potensi wilayah rawan kebakaran lahan hutan dibangun dan diimplementasikan menggunakan model prediksi berbasis jaringan saraf tiruan propagasi balik (JST-PB). Potensi wilayah rawan kebakaran diukur berdasarkan nilai confidence terjadinya titik panas sebagai variabel dependen dan beberapa nilai variabel lainnya, seperti brightness, tingkat radiasi, emisi karbon, suhu permukaan tanah, curah hujan, dan kelembapan tanah sebagai variabel independen. Nilai variabel yang diperlukan untuk pemodelan prediksi JST-PB diperoleh dari data yang berasal dari sensor MODIS yang terdapat pada satelit Terra dan Aqua, Tropical Rainfall Measuring Mission, Global Fire Emission Database, dan National Oceanic and Atmospheric Administration. Hasil uji coba menggunakan evaluasi berbasis 10-fold cross validation diperoleh model JST-PB terbaik dengan arsitektur yang terdiri dari empat hidden layer, fungsi aktivasi ReLU, dan optimizer adam. Kemudian, berdasarkan hasil analisis korelasi antara variabel independen terhadap variabel dependen, maka dengan menggunakan arsitektur JST-PB terbaik tersebut diperoleh dua skenario hasil uji coba pemodelan dengan hasil prediksi yang berbeda. Skenario terbaik dengan nilai SMAPE rata-rata sebesar 9,733% diperoleh untuk model dengan hanya melibatkan tiga variabel independen dengan tingkat korelasi sedang hingga hampir sempurna, yaitu brightness, tingkat radiasi, dan emisi karbon. Sedangkan skenario terbaik kedua dengan nilai SMAPE rata-rata sebesar 10,652% diperoleh untuk model yang melibatkan semua variabel independen, yaitu brightness, tingkat radiasi, emisi karbon, suhu permukaan tanah, curah hujan, dan kelembapan tanah. ======================================================================================================== Forest fires occurrence in Indonesia has a significant impact on air and health quality. Forest fires are considered as a potential threat to sustainable development because of its direct effect on ecosystem balance, increased carbon emission, and increased biodiversity. In 2019, a total of 857 ha area of forest was burned and spread across six provinces, including 75,871 ha located in Riau Province. In an effort to reduce the possibility of forest fires, a prediction model is necessary to anticipate the occurrence of these fires. In this Final Project, a prediction of the possibility of potential forest fire-prone areas is built and implemented using a prediction model based on a backpropagation neural network (BPNN). The potential of fire-prone areas is measured using the confidence value of hotspots as the dependent variable and other several variables, including brightness, fire radiation power, carbon emission, land surface temperature, rainfall, and soil moisture as independent variables. Variable values needed for BPNN modeling is obtained from data derived from the MODIS sensors found on Terra and Aqua satellite, Tropical Rainfall Measuring Mission, Global Fire Emission Database, and National Oceanic and Atmospheric Administration. The test results using 10-fold cross validation produced the best BPNN model consisting of four hidden layers, ReLU activation function, and Adam optimizer. Then, based on the result of correlation analysis between independent variables and the dependent variables, using the best BPNN model produced two modeling scenarios with the different predictive results. The best scenario with an average SMAPE value of 9,733% was obtained for the model involving only three independent variables with medium to almost perfect correlation level, namely brightness, fire radiation power, and carbon emission. The second best scenario with an average SMAPE value of 10,652% was obtained for a model that involves all independent variables, namely brightness, fire radiation power, carbon emission, land surface temperature, rainfall, and soil moisture.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: prediksi, jaringan saraf tiruan, kebakaran hutan, provinsi Riau, MODIS prediction, artificial neural network, forest fire, Riau, MODIS
Subjects: Q Science > QA Mathematics > QA76.9.D343 Data mining
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User: Adhi Yoga Muris Pamungkas
Date Deposited: 23 Aug 2020 13:49
Last Modified: 23 Aug 2020 13:49
URI: http://repository.its.ac.id/id/eprint/79267

Actions (login required)

View Item View Item