Top left: covariance matrix obtained using all 29 features for each redshift bin, measured on the 500 maps from the simulation with the constraints on PC18 parameters, with smoothing θ_s = 2′. Top right: same as in the top left panel, but for the first redshift bin only. Bottom left: same as in the top right panel, but measured on 300 maps for smoothing θ_s = 4′. Bottom right: same as in the top right panel, but measured on 300 maps for smoothing θ_s = 6′. We notice that within the same simulation, the correlation between features at different redshift bins is quite small due to the adopted binning, while for a given redshift bin some features appear to be more correlated than others. The correlation between the graph statistics and the rest of the estimators further decreases with increasing smoothing scale.