Trend detection and prediction of rainfall using logistic machine learning and deep learning models in Bhopal region of Central India

View largeDownload slide

View largeDownload slide

Close modal

Climate change has made rainfall patterns more uneven and unpredictable. Recent advancements in machine learning and deep learning offer the capability to handle the complex, nonlinear nature of weather input parameters, leading to more reliable predictions. Therefore, in this study, trend detection and rainfall prediction using logistic machine learning and deep learning models have been carried out in the Bhopal region of central India. Trend analysis methods such as Mann–Kendall, Sen's slope, and Pettit test methods were applied to detect trends, estimate slope, and change points in weather parameters. The performance assessment of the logistic and deep learning model showed a higher F1 score on classification for the deep learning model (0.93) compared to the logistic model (0.56). The results revealed the greater capability of deep learning models for capturing the variations in rainfall compared to the logistic models. The sensitivity of the deep learning model was studied using gradients of the loss function (mean-square error) with respect to input variables. The gradient-based sensitivity measure revealed that rainfall was highly sensitive to RH_min, BSS, and RH_max. The deep learning model-based rainfall prediction may help with real-time decision-making for irrigation scheduling, planting, and harvesting, leading to the conservation of water and other resources.

• Rainfall prediction was conducted using logistic machine learning and deep learning models.

• Trend analysis used Mann–Kendall, Sen's slope, and Pettit tests.

• A significant increasing trend was found at maximum temperature, while minimum temperature showed no significant trend.

• Rainfall was significantly influenced by month, humidity, temperature, and wind velocity.

• The deep learning model outperformed the logistic model.