Due to the Ceylon tea industry having faced a range of economic, structural and climate conditions over the past few decades, a robust statistical approach is essential for short-term forecasting of tea yield and tea income. However, the existing literature indicates that very few analytical tools have been applied to forecast tea production and market trends in Sri Lanka. Therefore, the main purpose of this study is to introduce a novel Hybrid Grey Verhulst (HGV) Modeling framework to predict tea production in Sri Lanka. This proposed methodology supports evidence-based decision-making, effective risk management and the sustainable long-term development of the tea sector in Sri Lanka.
Accurate forecasting of tea production remains a significant challenge due to the presence of irregular and nonstationary monthly yield patterns, driven by a combination of environmental vulnerability, plant health conditions and fertilizer management practices. The proposed HGV Modeling Framework is composed of the Grey Verhulst Hybrid Method (HGV) for linear and the backpropagation neural network-based artificial neural network (ANN) approach for nonlinear domains, respectively. By combining empirical data with the HGV Modeling framework, this study aims to construct a reliable and adaptive predictive model that enhances the accuracy of tea yield forecasting under conditions of uncertainty and variability.
The proposed approach integrates field-level data collected from 12 randomly selected tea plots located in the upcountry plantations of the Haputale region, situated at an elevation of 1,695 meters above sea level in the Badulla District of Uva Province, Sri Lanka. For the model validation process, the dataset was divided into training (75%) and testing (25%) subsets from January 2013 to December 2022, which were used as the training dataset, while data from January 2023 to March 2024 were used as the testing dataset. The results indicated that the fitting and prediction error of the proposed HGV model had the best prediction performance among the other time series models, with the lowest root mean squared error values.
Accurately predicting tea yield is essential for ensuring efficient plantation management, resource allocation and long-term sustainability within the tea industry. In particular, due to constraints related to statistical assumptions and data availability, modern approaches such as Grey models and machine-learning-based hybrid models have been increasingly favored over traditional techniques.
