The accurate prior information of the south asian monsoon helps the government and farmers to mitigate agricultural losses and proper planning of water resources. However, the forecasting of south asian monsoon is a challenging task due to the involvement of complex nonlinear dynamics and its variability over time. In this work, we developed a method to predict the mean seasonal, intraseasonal and meteorological region-wise south asian monsoon using evolving climate networks combined with machine learning. The climate networks are constructed using daily surface air temperature data (SATA) from1948 to 2009.The SAT captures the underlying dynamics betweenthe ocean and the atmosphere due to heat exchange and other local processes. We used the various network measures i.e, local degree, local clustering coefficient, and average link distance measures as predictors for machine learning algorithmn. We implemented the modern machine learning regression technique, i.e., extra trees regressor, to predict mean seasonal, intraseasonal and region-wise south asian monsoon. Based on our testing results, Our new method can forecast the mean seasonal monsoon in average 45 days in advance with median deviation of 3.6%. For intraseasonal monsoon months median deviation vary from 2.78 to 7.92 with lead-time from 45 to 135 days. In case of meteorolgical subdivison region, the median deviation vary from 5 to 20 %. The fore-casting results show the method is competitive with the other methods used for forecasting south Asian monsoon.
