Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/172036
Title: Konstruksi Model Prediksi Kekeringan menggunakan Convolutional Long-Short Term Memory (ConvLSTM)
Other Titles: Construction of a Drought Prediction Model using Convolutional Long-Short Term Memory (ConvLSTM)
Authors: Faqih, Akhmad
Herlambang, Rizky Nugraha Putra
Issue Date: 2025
Publisher: IPB University
Abstract: Peningkatan risiko kekeringan dan banjir di sektor pertanian sangat terkait erat dengan variabilitas iklim, yang ditandai oleh pola curah hujan yang fluktuatif. Untuk mengatasi tantangan ini, dikembangkan model prediksi kekeringan menggunakan Convolutional Long Short-Term Memory (ConvLSTM), yang secara efektif menggabungkan kemampuan CNN dalam mengenali pola spasial dengan keunggulan LSTM dalam memodelkan urutan data temporal. Model ini dilatih menggunakan data historis curah hujan dan suhu permukaan laut sebagai prediktor utama, serta dioptimalkan melalui proses hyperparameter tuning dan cross validation. Hasilnya, model ConvLSTM menunjukkan peningkatan akurasi yang signifikan, dibuktikan dengan penurunan nilai MAE dan RMSE serta peningkatan r-squared. Evaluasi performa mencatat hasil terbaik pada issued time Desember, dengan korelasi prediksi dan data observasi mencapai 0,65. Temuan ini menegaskan potensi ConvLSTM dalam menyediakan prediksi kekeringan yang lebih akurat, sehingga memberikan implikasi penting bagi peningkatan ketahanan pangan dan pengelolaan risiko bencana.
The increased risk of drought and floods in the agricultural sector is closely linked to climate variability, which is characterized by highly fluctuating rainfall patterns. To address this challenge, a drought prediction model was developed using Convolutional Long Short-Term Memory (ConvLSTM), which effectively combines the advantages of CNN in recognizing spatial patterns with the strengths of LSTM in modeling temporal data sequences. This model was trained using historical rainfall and sea surface temperature data as the primary predictors and was optimized through Hyperparameter tuning and Cross validation. The results showed the ConvLSTM model demonstrated a significant improvement in accuracy, evidenced by a reduction in MAE and RMSE values and an increase in R-squared. Performance evaluation noted the best results at the December issued time, with a prediction and observation data correlation reaching 0.65. These findings affirm the potential of ConvLSTM in providing more accurate drought predictions, thus providing important implications for improving food security and disaster risk management.
URI: http://repository.ipb.ac.id/handle/123456789/172036
Appears in Collections:UT - Geophysics and Meteorology

Files in This Item:
File Description SizeFormat 
cover_G2401201080_be6d60ae3fc44571a92dc13d89c36554.pdfCover619.96 kBAdobe PDFView/Open
fulltext_G2401201080_bd232f6d2f6743b69081aeff68500dcd.pdf
  Restricted Access
Fulltext1.38 MBAdobe PDFView/Open
lampiran_G2401201080_ec0ce29f8c0c4ea28a18ce2b393b4468.pdf
  Restricted Access
Lampiran283.33 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.