We propose a new detection strategy for gravitational waves (GWs) below few Hertz based on a correlated array of atom interferometers (AIs).Our proposal allows to reject the Newtonian Noise (NN) which limits all ground based GW detectors below few Hertz, including previous atom interferometry-based concepts.Using an array of long baseline AI gradiometers yields several estimations of the NN, whose effect can thus be reduced via statistical averaging.Considering the km baseline of current optical detectors, a NN rejection of factor 2 could be achieved, and tested with existing AI array geometries.Exploiting the correlation properties of the gravity acceleration noise, we show that a 10-fold or more NN rejection is possible with a dedicated configuration.Considering a conservative NN model and the current developments in cold atom technology, we show that strain sensitivities below $1\times 10^{-19}/\sqrt{Hz}$ in the $ 0.3-3 \ Hz$ frequency band can be within reach, with a peak sensitivity of $3\times 10^{-23} /\sqrt{Hz}$ at $2 \ Hz$.Our proposed configuration could extend the observation window of current detectors by a decade and fill the gap betweenground-based and space-based instruments.