The Power of TimeLLM for Greenhouse Microclimate Mapping

The monitoring and estimation of climatic conditions in a greenhouse environment is an extremely important activity for any farmer. Through such activity, the well-being and growth of plants can be ensured by preventing diseases or biological stress, while wise decisions can be made about the use and consumption of resources, like water. Prediction and forecasting of such conditions are usually performed via tools following in the family of time series models. This kind of model has been represented for a long while by Recurrent Neural Networks (RNNs), and their improved variants, and only recently by transformer architectures. In particular, due to the spread of Large Language Models (LLMs) and their recognized capabilities in capturing complex interactions between consecutive tokens, some attempts have been made to use them for time series forecasting. In this paper, we explore the ability of a model, called Time-LLM, to capture and model the microclimate inside a greenhouse. Some experiments with a real-world dataset and a comparison with other techniques are also provided.