Sandpiper model predicts rainfall

AI can predict rainfall intensity better than several widely used forecasting models in tests using historical weather data from India. The new model reported in the International Journal of Mobile Communications shows that combining different forms of AI, along with advanced data-cleaning and optimisation techniques, can make rainfall prediction more accurate and reliable, particularly when expressed in practical categories such as light, moderate, or heavy rain.

The system uses a deep convolutional spiking neural network to identify spatial patterns in weather maps. The spiking aspect of the neural network was inspired by how brain cells communicate using short electrical pulses over time. Before the network training step, the researchers cleaned the data using a method called anisotropic diffusion Kuwahara filtering. This process reduces noise, random errors, while preserving important patterns. This is important in weather datasets, which often contain missing or uneven measurements.

The new model was evaluated using the India Rainfall Analysis dataset, which contains historical records from selected regions. Instead of predicting exact rainfall amounts, the system classifies conditions into rainfall categories. This type of classification is often more useful in practice, because decisions in agriculture, water management, and disaster response are frequently based on thresholds rather than precise measurements.

In the performance tests, the system worked better than established AI methods such as machine learning tools, like recurrent neural networks and gradient-boosting models. The new system raised fewer false alarms and did not miss major rainfall events, as was a problem with earlier models.

The team has improved the model using the sandpiper optimisation algorithm. This additional tweak models the behaviour of foraging waders (shorebirds) known as sandpipers. In machine learning terms, this additional tweak helps the model reduce prediction errors by optimising its internal settings.

Amanullah, M., Ananthajothi, K. and Agoramoorthy, M. (2026) ‘Optimised deep convolutional spiking neural network for accurate long-term and short-term rainfall forecasting in climate prediction systems’, Int. J. Mobile Communications, Vol. 27, No. 3, pp.300–315.