Understanding the Microclimate predictors in coffee-pine agroforestry systems facilitates the prediction of microclimate, enabling the maximization of productivity and viability of the agroforestry systems. Coffee-pine agroforestry systems represent a VIABLE integrated system of agricultural productivity along with environmental sustainability. The models were tested in the dataset across different time frames of daily, weekly, and monthly. In this research, a number of ML approaches and models, including LSTM, DLinear, Transformer, ARIMA, and MLP, were employed in microclimate prediction with focus on variables including Temperature, Humidity, and Intensity. The focus of this study on the deployment of this many models, stems from the aim of harnessing the varied strengths of models on diverse classes of data and predictions. From the analysis, model DLinear always performs excellently over all time frames as compared to other models, where DLinear possesses high accuracy with a mean absolute error (MAE) of 0.43, while LTSM, ARIMA, MLP, and Transformer possess MAE with 1.12, 1.77, 2.27, and 4.74, respectively. These results further enrich the existing research on predicting geographical microclimate in agroforestry systems, providing evidence of the usefulness of various machine learning models in understanding and managing complex ecosystems. Given the range of these types of models, this ensures that a wide spectrum of the problem is addressed, which helps ensure the quality and correctness of the forecasts by using each model of its strengths to tackle a different area of the prediction process.
Nurwarsito et al. (Sun,) studied this question.