<p>The present study examined three key aspects of the sunshine duration -global solar radiation model over North India: calibration of Angstrom coefficients for both daily and monthly time scales, stability analysis of the Angstrom coefficients, and lastly, implying the data-driven artificial and convolutional neural network (ANN and CNN) regressions for higher accuracy and identification of non-linear relationship between the two variables. The average values of the Angstrom coefficients for the region were determined to be a = 0.26, b = 0.45 with coefficient of variation (R²) = 0.80 for monthly and a = 0.29, b = 0.43 with R² = 0.79 for daily sunshine-radiation data, respectively. A significant (p &lt; 0.01) linear relationship exists between sunshine duration and global radiation, with interchangeable Angstrom coefficients at two time scales. The daily Angstrom coefficients were steadier than those calibrated with monthly data with different data lengths. Also, the time-variation of Angstrom coefficients was insignificant. ANN and CNN regression models showed better performance, improving R² from 1–6%. Further, ANN regression plots for monthly data values also exhibited pronounced non-linear behaviour for Lucknow and Varanasi, and exhibited the usual non-linear relationship for daily values.&#xa0;However, for most stations, the improvement of the ANN model over the conventional A-P model was insignificant. Therefore, for practical applications, the linear A-P model should be considered due to its simplicity and reasonable accuracy.</p>