The problem of long-range Indian monsoon forecasting is acute: official operational forecasts often miss the mark, and the El Niño–monsoon link has weakened in recent decades. Alongside this work in the archive are studies that also seek precursors in climate data: some predict El Niño a year and a half ahead, others analyze shifts in rainfall regimes or warn of drought using neural networks. This paper continues that line but uses a network approach to the global temperature field to directly predict rainfall amounts, not just monsoon onset or withdrawal.
The authors took NCEP/NCAR reanalysis surface temperature data from 1948 to 2019 and placed 726 nodes evenly covering the globe. For each pair of nodes, they computed the temporal correlation of temperature anomalies with lags up to 200 days. Links with the strongest positive strength (top 5%) formed a directed network. For each node, they counted the number of incoming links (in-degree) for each calendar year. They then identified the node whose in-degree correlates most strongly with the all-India rainfall index in the following year. That node turned out to be in the South Atlantic near 30°S, 30°W. The forecast was built using linear regression trained on 1950–1988 and tested on 1989–2018.
The key node in the South Atlantic showed a correlation of 0.49 with the all-India rainfall index over the training period. Independent validation on 1989–2018 yielded a correlation of 0.50 between forecast and observations, significantly better than official operational forecasts and comparable to the best models, but with a five-month lead instead of one month. The method also works for individual regions: for Central India the correlation was 0.43, and for East and Northeast India, after retuning the network to negative links, it was 0.46. The forecast for 2019 differed from the actual value by less than 2%. For 2020, 842.85 mm is predicted, about 5% below the long-term normal.
This work shows that the network structure of global temperature contains information about the future monsoon that traditional indices miss. Unlike neighboring studies that predict monsoon onset or individual extreme events, here the seasonal rainfall total is forecast directly with a long lead time. This fills a gap between predictability and practical forecasting for agriculture. In addition, a link to warming is found: stronger cross-equatorial links improve predictive skill, pointing to changes in the dynamics of the climate system itself.
The method can be extended to other monsoon regions and to forecasting extreme seasons. Further analysis of the physical mechanism, especially the role of Rossby waves and cross-equatorial flows, will help build confidence in the predictor. Combining the network approach with dynamical models for hybrid forecasts is also promising. As warming continues to reshape the link structure, regular recalibration of predictors will be needed.
Better long-range monsoon forecasts will directly affect planning in agriculture, water resource management, and mitigation of droughts and floods in India. The method can also be applied to other regions with seasonal rainfall tied to remote connections.
Test the predictor's robustness on independent data and in other climate models. Investigate the physical mechanism linking the South Atlantic node to the monsoon through model experiments.
The work addresses the fundamental problem of climate system predictability: how to extract a signal about the future state from a complex network of interactions. It also relates to the question of how global warming affects predictability and variability of monsoon systems. The network methodology opens a path to finding new predictors in nonlinear climate data, where traditional linear indices lose their power.
The key node in the South Atlantic has warmed by 2.5°C over 70 years, and over the same period its number of incoming links grew from 17 to 32, which improved the forecast.