Deep Investigation of Neutral Gas Origins (DINGO): Options for robust Deep Spectral Line Imaging in the SKA-Era

TL;DR

The paper evaluates three deep spectral-line imaging strategies for ASKAP-era HI surveys using the DINGO data: traditional visibility stacking, image stacking, and a novel uv-grid grid stacking. It demonstrates that image stacking underestimates HI flux and introduces non-physical artefacts, while grid stacking achieves flux recovery (~99%) comparable to the traditional method and avoids artefacts, at the cost of higher per-epoch memory but with efficient uv-grid compression (~7:1). The study provides a detailed resource analysis, showing grid stacking as the most viable approach for large-scale, IO-bound surveys in the SKA era. The findings support implementing uv-grid stacking for the full DINGO survey and advocate adopting this technique for future deep spectral-line programs to balance data quality, storage, and compute needs.

Abstract

The data storage requirements for deep spectral line observations with next-generation radio interferometers like the Australian Square Kilometre Array Pathfinder (ASKAP) and the Square Kilometre Array (SKA) are extremely challenging. The default strategy is to reduce data after each daily observation and stack the resulting images. Although this approach is computationally efficient, it risks propagating systematic errors and significantly degrades the final data quality. However, storage and computation requirements for a traditional way to image the entire deep dataset together are prohibitively expensive. We present an alternative \textit{uv}-grid stacking method and compare its scientific outcomes with both the traditional approach, which processes all data jointly and serves as the best-possible result, and the default image-stacking method. Our technique involves halting the standard imaging pipeline after the daily residual visibility grids are formed. These grids are then stacked and jointly deconvolved to combine many epochs of data. Using the traditional approach as a benchmark, we show that image-stacking recovers only 92\% of the true {\HI} flux. In contrast, our \textit{uv}-grid stacking method recovers 99\%, which is in excellent agreement with the traditional method within the noise limits. Furthermore, image-stacking introduces significant non-physical artefacts, such as negative bowls around strong sources, indicating poor deconvolution and a loss of physical information. Based on these findings, we intend to apply the \textit{uv}-grid stacking to the Deep Investigation of Neutral Gas Origins (DINGO) survey on ASKAP and strongly recommend this or a similar approach for future radio astronomy facilities.

Figures (5)

• Figure 1: ASKAP beam footprints used for DINGO observations over one tile of the G23 region. Two 36-beam patterns (blue for footprint A and red for footprint B, respectively) of each 6-by-6 beam configuration are interleaved to achieve the uniform sensitivity over the field. Beam IDs are overlaid on individual beams. For this study, we select two beams (dashed-line circles) from each footprint, surrounding a bright H i galaxy of NGC 7361 (denoted with a yellow star) already detected in this field by the HIPASS survey Koribalski:2004. The background is the optical all-sky image from Mellinger:2009.
• Figure 2: The schematic flowchart shows three deep imaging methods compared in this study using DINGO data.
• Figure 3: Comparison of the moment 0 images made with the three methods for NGC 7361. Shown on the diagonal are the moment 0 maps converted to column density for traditional visibility stacking (V), grid stacking (G) and image stacking (I), respectively. The lower off-diagonal plots are the differences between these three approaches and the upper off-diagonal ones are the ratios. Whilst the three approaches give very similar looking results, it is obvious that low level residuals remain in the image-stacking case, whereas the other two methods give very similar results. This is born out in the difference shown in V-G (visibility stacking minus grid stacking image) where there is a small ($\sim$2%) residual, wholly contained within the region of the galaxy with maximum flux. In comparison, the difference shown in V-I (visibility stacking minus image stacking) contains negative and positive features, and is not aligned well with the original image. The ratios reaffirm this result, in that the ratio of the V/G is smooth over the area with H i emission whereas the ratio of V/I shows a great deal of structure and a much greater range of values.
• Figure 4: Comparison of the H i spectra made with the three different methods (V, G and I) for NGC 7361 (top). The bottom panel shows the differences between three spectra. Grid-stacking residuals are close to a constant fraction of the V profile, which suggests the cleaning cutoff needs to be deeper in this method where there are no major cycles. Image-stacking results have more complex structure, which suggests there is no simple remedy.
• Figure 5: The comparison of derived physical parameters for detected sources using traditional visibility stacking (V), grid-stacking (G, blue circles) and image-stacking (I, red triangles). (a) Integrated flux: integrated flux from grid stacking (F_int$_{\rm G}$) and image stacking (F_int$_{\rm I}$) on y-axis versus the traditional visibility stacking (F_int$_{\rm V}$) on x-axis. (b) Integrated flux ratio: The ratio of the stacked integrated flux (Grid/Visibility in blue; Image/Visibility in red) over the traditional integrated flux. The median ratios are indicated by the dashed lines: 0.99 for grid-stacking (blue) and 0.92 for image-stacking (red). (c) The H i line width at 50% of the mean flux density across the spectrum ($W_{50}$) derived from grid-stacking and image-stacking (y-axis) against the traditional visibility stacking result (x-axis). Dashed lines show linear fits, with slopes of 1.00 (blue, V vs G) and 0.97 (red, V vs I). (d) The ratio of $W_{50}$ from grid-stacking and image-stacking $W_{50_{\rm G,I}}$/$W_{50_{\rm V}}$) over the traditional visibility stacking method. The median ratios are 1.00 (blue dashed line, G/V) and 1.02 (red dashed line, I/V). (e) Same as (c), but for the H i line width at 20% of the peak flux ($W_{20}$). Fitted slopes are 1.00 for both V vs G (blue) and V vs I (red). (f) Same as (d), but for the $W_{20}$ line width ratio ($W_{20_{\rm G,I}}$/$W_{20_{\rm V}}$). The median ratios are 1.00 (blue dashed line, G/V) and 1.00 (red dashed line, I/V).