Soil moisture can be used to provide predictive information important for precision farming, irrigation management, and pollution monitoring. Nonetheless, existing approaches often struggle to handle non-linear temporal relationships, environmental uncertainties, and poor real-time integration of sensor data, thereby degrading prediction performance. To overcome these problems, in this article, we introduce JaNarX, a soil moisture prediction framework (SMPF) based on IoT and the NARX model, optimised with the Jackal Apis Optimisation (JAO) algorithm. The approach utilises time-series radar satellite variables, meteorological terms, and continuous in situ sensor data to better capture dynamic changes of soil moisture. The Wazihub Soil Moisture Dataset (WSMD), which is the aggregation of multi-sensor environmental data measured in in-field conditions from real agricultural fields in Senegal, was used for training and validation. JAO was used to speed up convergence and tune NARX hyperparameters for solving the local minima problem,and increase the generalisation performance of the model. Experimental results in MAE, MSE, and RMSE evaluations had MAE = 1.53, MSE = 2.89, and RMSE = 1.70, which outperformed the state-of-the-art base-models such as SVM, LSTM, GLM XGBR, and traditional NARX network. The simulation results confirm that the proposed IOT-based, JAO-optimised NARX structure can achieve more accurate prediction than the traditional method and has good stability and computational efficiency. This paper presents a scalable, high-accuracy model for the prediction of soil moisture, suitable to enable real-time planning decisions for precision irrigation and sustainable water management in agriculture.