Analysis and Prediction of Tomato Growth Dynamics in an IoT Smart Agriculture Testbed

Plants' growth monitoring and modeling, supported by Internet of Things (IoT) and Deep Learning (DL) technologies, are key challenges in Smart Agriculture (SA) applications. In this work, we present a SA tomato testbed, featuring heterogeneous IoT sensors-including fruit and stem dendrometers-deployed to capture data related to the plants' growth dynamic. Data have been sampled every 10 min in the period July-September 2025, originating time series representing both plants dimensional variations and environmental parameters. Moreover, a growth analysis pipeline is proposed, considering Multiple Seasonal-Trend decomposition using LOESS (MSTL) combined with growth variation modeling to characterize trend, seasonal components, short-term dynamics and growing phases. Moreover, Recurrent Neural Networks (RNNs) are compared for predicting short-term growth variations, applying Permutation Feature Importance (PFI) to evaluate features importance to the best identified models. Experimental results on both fruit and stem size return: a non-linear increasing growth trend, followed by a seasonal deceleration; the presence of daily and weekly growth patterns in signals; and best short-term forecasting performance obtained through Long Short-Term Memory (LSTM) model, explaining up to 95% and 84% variance in fruit and stem growth variations, respectively.