The simplest way is to apply dropna over the dimensions separately (which will remove the nan coordinates)

era_land.dropna(dim='long',how='all').dropna(dim='lat',how='all')