The prediction of yield for small-scale farms is helpful for food security as well as harvest management. Several studies have proven that image data and climatic data give yield estimation for small- and large-scale farms. Based on the growth pattern, we can estimate the yield more accurately. Crop development is influenced by essential parameters such as weather patterns and soil properties. In this work, climatic information is treated as time-series data and, together with soil attributes, is analyzed using deep learning models like RNN and LSTM for effective yield prediction. The combination of both provides yield estimation. The proposed model, LRNN, integrates RNN and LSTM networks to create a potent framework for sequential data modeling, efficiently capturing temporal dependencies and mitigating vanishing gradient issues. LRNN served as a standard for various deep learning and machine learning algorithms based on the selected parameters: Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). RNN with LSTM gave the least mean absolute percentage error compared to other machine learning algorithms overall. This study evaluates the yield prediction for different seed varieties of turmeric, scientifically known as Curcuma longa. The Rajendra Sonia variety yielded more than the other two varieties, approximately 36 tons per hectare. The Lakadong variety yielded less than the other two varieties, at 19.7 tons per hectare.

Crop Yield Deep Learning LSTM RNN MSE MAPE Turmeric