The dropsonde data has been run through correction code that addresses the dry bias (see UCAR tech doc). Note that there are more missing RH values than previously. This is because the software error write out 1%RH for values below a threshold. When correcting these very low values, the  algorithm had no way of knowing the correct humidity measurement and set those values to missing. It may be possible to interpolate between known RH values to establish an upper level to the RH where the values were missing.