Climate-Driven Insights: Predicting Black Pepper Yield and Quality with Long Short-term Memory Model

Abstract

In the context of climate variability, predicting agricultural output remains a pressing challenge, particularly for high-value crops like black pepper in Sri Lanka, a leading spice exporter. This study introduces a novel machine-learning approach to predict black pepper yield and quality, utilizing thirty years of detailed weather data from the Matale district. Employing Long Short-Term Memory (LSTM) networks, the complex dependencies between weather conditions—including rainfall, temperature, and humidity—and crop productivity are modelled. The analysis demonstrates that LSTM models can effectively forecast black pepper yield and quality by learning from historical weather patterns and corresponding crop performance data. The models achieved a mean absolute error of 18-20% for quality predictions and a mean squared error reflecting consistent model performance across different evaluations. By providing reliable yield and quality estimates, these models serve as valuable tools for farmers and policymakers to better plan and manage black pepper cultivation in response to anticipated climate conditions. Furthermore, the research highlights the potential for enhancing model accuracy by incorporating diverse regional data, thereby contributing to more resilient agricultural practices in the face of global climate change.