This website contains the IGP-UHM AI model used in the paper by Rivera Tello et. al. (2023) “Explained predictions of extreme Eastern Pacific El Niño events”. The convolutional neural network (CNN) takes as input 3 consecutive months of four environmental variables (SST, SSH, UWND, VWND) which are known to contain strong signals of an El Niño event.
This interactive application lets you explore the model results as it was presented in the paper, along with a custom set of tools that can help with experimenting “on the fly”.
Furthermore, as shown in the original research, you can explore the relevance contours extracted from the model by using Layer-wise Relevance Propagation (LRP), an eXplainable AI method. This allows us to highlight areas where the model considers there is a high contribution to an extreme EP El Niño prediction.
Note:
Each initial condition is composed of 3 consecutive months. After selecting a month and a year, the model will load that month along with the previous two months for the input variables, e.g., selecting Jun 1997 as the initial condition uses Apr-May-Jun 1997 as input.
Reference: Rivera Tello, G. A., Takahashi, K., & Karamperidou, C. (2023). Explained predictions of strong eastern Pacific El Niño events using deep learning. Scientific Reports, 13(1).https://doi.org/10.1038/s41598-023-45739-3
Initial Conditions
Select the month and year for the prediction starting point. The model will generate 12-month forecasts based on these initial conditions.
Model Information
No Prediction Data
Select initial conditions and click "Run Model" to generate forecasts.