Time Series Analysis for Predicting Tea Harvest Yield: A SARIMAX-Based Approach

Abstract

Precise prediction of tea yield is crucial for both agricultural planning and economic forecasting. Projection of the future trends, which rely on present and historical data, is the process termed 'forecasting'. Yield prediction is fundamental for research and development. By gathering the yield data over historic times, the researchers can make their work more valuable by predicting the yield patterns. This can help in analyzing the impact of changing environmental variables that can lead to a change in the yield prediction. To predict tea yield on the basis of historical yield and meteorological variables, this study proposed the optimal use of the Seasonal Autoregressive Integrated Moving Average with Exogenous Variable (SARIMAX) model, which was applied to the historical data from the years 1985-2022. The model integrates both the seasonality of tea production and external factors that influence the crop growth, such as rainfall, temperature, etc., which are known to influence crop growth. Two of the competing models, SARIMAX (1,1,1) and SARIMAX (1,0,0) (0,0,1,12), were applied and validated on statistical parameters log-likelihood, AIC, BIC, residual diagnostics, the Ljung-Box test, and the Q-statistic. The results showed that the hyperparameter-tuned model SARIMAX (1,0,0) (0,0,1,12) successfully captured both temporal and seasonal patterns. This model yielded a lower AIC (-553.70) and exhibited consistent residuals, normally distributed and free from autocorrelation. The results indicate the robustness of SARIMAX in yield predictions and also highlight its role in the planning and framing of agricultural policies.