A Three-Layer Distributed Architecture with Cloud-Based Predictive Modeling for Intelligent Greenhouse Monitoring and Forecasting in Semi-Arid Environments

Greenhouse agriculture in semi-arid regions faces persistent challenges from unpredictable thermal variability, frost events, and seasonal drought stress that are difficult to manage without anticipatory climate information. This paper presents the design, implementation, and validation of an end-to-end intelligent greenhouse monitoring and temperature forecasting system deployed in Donovan, San Luis Province, Argentina. The proposed platform integrates a three-layer IoT architecture with cloud-based statistical forecasting to support real-time decision making under semi-arid climatic conditions. The system integrates sensor nodes, a push-MQTT gateway co-located at the Universidad Nacional de San Luis to bypass regional API geo-restrictions, and a cloud application layer. Three forecasting strategies with a six-hour prediction horizon were evaluated: a univariate SARIMA baseline (Model 1), a SARIMAX model using four neighboring meteorological stations as individual exogenous regressors (Model 2), and a SARIMAX model employing a single correlation-weighted synthetic exogenous index (Model 3). The models were assessed using MAE, RMSE, and Diebold–Mariano statistical significance tests. The results show that directly incorporating multiple correlated exogenous variables does not improve forecast accuracy because of multicollinearity, whereas the proposed correlation-weighted synthetic index preserves the spatial predictive signal while reducing model complexity and achieving performance comparable to the baseline overall, with statistically significant improvement during the overnight block. Additional benchmarking against Random Forest, Support Vector Regression, Temporal Convolutional Networks, and Long Short-Term Memory models demonstrates that increasing model complexity does not necessarily translate into improved predictive performance for short-horizon greenhouse temperature forecasting. Together, the proposed IoT architecture and forecasting framework provide an operationally validated solution for intelligent greenhouse monitoring and predictive decision support in resource-constrained semi-arid environments.