Forecasting the Appearance Frequency of Rafflesia arnoldii in Bengkulu, Indonesia, Using Discrete-valued Time Series Modeling

[Purpose] This study aims to develop a probabilistic forecasting model to estimate the frequency of Rafflesia arnoldii appearances, a rare flower species endemic to Bengkulu Province, Indonesia. Motivated by the need for effective conservation and ecotourism planning, the study addresses a gap in the literature by applying discrete-valued time series models to ecological count data. [Design/methodology/approach] This study analyzes monthly data on the frequency of Rafflesia arnoldii appearances from January 2016 to December 2023. We applied and compared multiple discrete-valued time series models, including Poisson INAR, Negative Binomial INAR, Zero-Inflated INAR, INARMA, NB-INARMA, and INGARCH. The models were fitted using Maximum Likelihood Estimation and the EM algorithm. Model performance was evaluated using AIC, BIC, RMSE, and MAAPE, with post-estimation diagnostics conducted via Ljung-Box, Jarque-Bera, and Lagrange Multiplier tests. [Findings] The analysis found that the optimal models were INAR(1)-INGARCH(1,1) for South Bengkulu, and INAR(1)-INGARCH(1,1) for Kepahiang. These models effectively captured the discrete, overdispersed, and zero-inflated characteristics of the data, producing reliable short-term forecasts. The variation in model performance across regencies highlights underlying ecological differences and reinforces the need for region-specific forecasting strategies. [Research limitations/implications] Limitations include a focus on four regencies and the exclusion of exogenous environmental variables such as rainfall and temperature. Implications suggest that future studies should expand the data scope, incorporate environmental covariates, and explore spatio-temporal extensions to improve predictive accuracy. In addition, future extensions may explore integer-valued time series frameworks designed to accommodate non-stationary count data, including signed-thinning or generalized INAR formulations, to further enhance model flexibility and predictive performance. [Practical implications] The forecasting framework can inform the scheduling of conservation monitoring and guide ecotourism planning by anticipating peak flowering periods. This study’s primary relevance to Decision Sciences lies in providing a quantitative, data-driven framework for conservation resource allocation and ecotourism planning under uncertainty, transforming ecological data into actionable management strategies. [Originality/value] This study pioneers the application of discrete-valued time series models in forecasting the occurrences of a rare and sporadically flowering plant species. It contributes to the Decision Sciences literature by demonstrating how probabilistic modeling can be applied to biodiversity management under uncertainty.