FORECASTING SPI AND SPEI DROUGHT INDICES USING CLIMATE DATA AND LONG SHORT-TERM MEMORY (LSTM) NEURAL NETWORKS

Damayanti, Mellisa (2025) FORECASTING SPI AND SPEI DROUGHT INDICES USING CLIMATE DATA AND LONG SHORT-TERM MEMORY (LSTM) NEURAL NETWORKS. Undergraduate thesis, Universitas Pendidikan Ganesha.

[img] Text (COVER)
2115101037-COVER.pdf
Download (2MB)
[img] Text (ABSTRAK)
2115101037-ABSTRAK.pdf
Download (239kB)
[img] Text (BAB 1 PENDAHULUAN)
2115101037-BAB 1 PENDAHULUAN.pdf
Download (231kB)
[img] Text (BAB 2 KAJIAN TEORI)
2115101037-BAB 2 KAJIAN TEORI.pdf
Restricted to Repository staff only
Download (519kB) | Request a copy
[img] Text (BAB 3 METODOLOGI PENELITIAN)
2115101037-BAB 3 METODOLOGI PENELITIAN.pdf
Restricted to Repository staff only
Download (10MB) | Request a copy
[img] Text (BAB 4 HASIL DAN PEMBAHASAN)
2115101037-BAB 4 HASIL DAN PEMBAHASAN.pdf
Restricted to Repository staff only
Download (52MB) | Request a copy
[img] Text (BAB 5 PENUTUP)
2115101037-BAB 5 PENUTUP.pdf
Restricted to Repository staff only
Download (225kB) | Request a copy
[img] Text (DAFTAR PUSTAKA)
2115101037-DAFTAR PUSTAKA.pdf
Download (305kB)
[img] Text (LAMPIRAN)
2115101037-LAMPIRAN.pdf
Download (573kB)

Abstract

This research aims to develop a drought forecasting model using climate data. It is motivated by Northeast Thailand’s heavy reliance on natural rainfall, which makes crop production highly vulnerable to climate variability. Reliable drought forecasting is essential for early warning and preparedness. Therefore, this research employs a deep learning approach using Long Short-Term Memory (LSTM) neural networks to forecast two key drought indices: Standardized Precipitation Index (SPI-3) and Standardized Precipitation Evapotranspiration Index (SPEI-3) at a three-month scale. Historical monthly climate data spanning 40 years (1981–2020) were used to compute these drought indices. The forecasted model results were analyzed based on three drought characteristics: intensity, spatial variation, and category. The model’s capability to predict drought intensity was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Bias Error (MBE), and the Coefficient of Determination (R²). Both models demonstrated promising performance on the testing dataset, with R² values of 0.95 and 0.93 and RMSE values of 0.22 and 0.27 for SPI and SPEI, respectively. Spatial analysis was conducted to identify variations in drought patterns across different locations. Drought categories were assessed using the Receiver Operating Characteristic-based Area Under the Curve (ROC-AUC), yielding AUC values of 0.83 and 0.81. The model effectively identified drought patterns over time and distinguished common categories like Light, Moderate, and Severe. However, it struggled with rare categories like Extreme Drought, though predictions remained within closely related categories. Overall, the model shows promising potential for drought forecasting, offering valuable insights for farmers to anticipate dry conditions. This study highlights LSTM’s potential in drought prediction to support agricultural mitigation strategies. Future research could explore extending forecasts beyond one month and enhance the model’s ability to recognize the extreme drought category more effectively.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: Drought forecasting, Climate data, Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI), Long Short-Term Memory (LSTM)
Subjects: T Technology > T Technology (General)
Divisions: Fakultas Teknik dan Kejuruan > Jurusan Teknik Informatika > Program Studi Ilmu Komputer (S1)
Depositing User: Mellisa Damayanti
Date Deposited: 27 Mar 2025 03:40
Last Modified: 27 Mar 2025 03:40
URI: http://repo.undiksha.ac.id/id/eprint/23870

Actions (login required)

View Item View Item