Understanding the nature and variability of precipitation extremes is important for flood risk management. 20th century has seen more loss of life and property damage due to flooding than any other natural disaster in the United States. Changes in frequency, intensity, duration and timing of extremes can amplify the stress on societal systems. Climate extremes encompass large scale natural modes of variability of the climate system, which have dynamic linkages to the extreme events experienced at smaller space and time scales but could lead to major disasters those are persistent in time and spatial coverage. Here we present a comprehensive analysis of the extreme precipitation events in different seasons (DJF, MAM, JJA, SON) for the whole northern USA using observed daily precipitation data. The aim is to detect the spatio-temporal patterns of dominant modes of extreme precipitation magnitude using both linear (Principal Component Analysis) and non-linear (Maximum Variance Unfolding) techniques and understand the associated large scale atmospheric circulations (e.g. Pacific North American pattern) affecting the variance of such prototypes in the respective seasons. We also seek to identify those extreme precipitation patterns (single/multiple) identified by the PC and MVU analyses, which might be the cause of major flood events in the US and present the map of the regions at high risk. This would also help us to explore the predictability of magnitude and extent of flood risk.