Rainfall Risk Modelling for Rice Farming Using Continuous Hidden Markov Models

Martal, David Vijanarco (School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)
Setiawaty, Berlian (School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)
Budiarti, Retno (School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)

Zero : Jurnal Sains, Matematika, dan Terapan
Vol 9, No 3 (2025): Zero: Jurnal Sains Matematika dan Terapan

Climate change has increased rainfall variability and unpredictability, significantly impacted agricultural productivity, and raised the risk of crop failure, particularly in rain-fed rice farming systems. This study models rainfall data from Tabanan, Bali, using a continuous-time Hidden Markov Model (HMM) to identify latent weather states and assess the associated risk of rice crop failure. The model assumes four hidden states, each generating rainfall observations following a Gamma distribution. Simulation results produced Mean Absolute Percentage Error (MAPE) values below 5% for training and testing sets, indicating strong model performance in replicating rainfall patterns. Risk analysis compared simulated rainfall with rice crop water requirements across three planting periods. The second planting period (July-October) exhibited the highest risk at 3.75%. Compared to other predictive models, HMM offers superior capability in capturing temporal rainfall structure and identifying critical transition phases, making it highly suitable for agricultural risk assessment and climate-adaptive planning.