Fig. 2: Predictions of the interannual variability of the Amazon and Congo rivers based on observed and model-simulated sea surface temperatures compared with climatology.
From: Explainable deep learning for insights in El Niño and river flows
River flow ground truth observations (black) and predictions using different predictors from January 2003 to December 2005 for Amazon (a) and Congo (b) iver. The predictors are mean Niño 3.4 calculated from 32 Earth System Models (ESM) (ESM Mean Niño 3.4), Niño 3.4 calculated from each of 32 ESMs (ESM Niño 3.4), Niño 3.4 index from NOAA (Niño 3.4), Niño 3.4 calculated from 3 reanalysis (Reanalysis Mean Niño 3.4), Niño 3.4 calculated from each of 3 Reanalysis (Reanalysis Niño 3.4), Niño 3.4 anomaly (Niño 3.4 index calculated by NOAA from HadISST1), sea surface temperature (SST) from 32 ESMs (ESM SST, light purple) and SST from 3 reanalysis (Reanalysis SST, gray). Seasonality was subsequently added to the predictions of river flow anomaly based on Niño 3.4 anomaly to generate absolute river flow. The brown line is the historical average prediction result. For models using El Niño–Southern Oscillation (ENSO) index as predictor, we applied six models [linear regression, ridge regression, elastic net regression, random forest regression and deep neural network regression] and use their ensemble as the final prediction. The shaded areas are 1 standard deviation for ensemble methods and historical averaging.
