The Autoregressive Moving Average (ARIMA) model has also been applied successfully for forecasting maize production in Tanzania, as noted in. Nonetheless, ARIMA is inherently limited to univariate time series data and does not adequately address spatial dependencies. To overcome these limitations, machine learning approaches, particularly deep learning techniques such as Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks, have been applied for spatiotemporal data analysis. Though the CNN-LTSM hybrid model performed fairly well with spatiotemporal data, it failed to capture stationarity and handle the underlying linear structure in the model, leading to higher error metrics. Moreover, developed an ARIMA-CNN-LSTM model to forecast carbon pricing, effectively capturing both linear and non-linear data characteristics, and achieving lower RMSE compared to benchmark models. However, a key drawback of the ARIMA-CNN-LSTM model is the lengthy training time due to the complexity of the LSTM component. In light of these changes, we propose the development of a novel hybrid model, termed the ARIMA-CNN-GRU (ACG)-hybrid model, which seeks to bridge the gaps identified in conventional and existing hybrid deep learning models. This model leverages the strengths of the ARIMA as a foundational model to handle the underlying linear structure in time series data while integrating CNN and Gated Recurrent Unit (GRU) architectures to address complexity and nonlinear residual patterns in unstructured spatiotemporal datasets.