We begin here a series of papers examining the chromospheric and coronal properties of solar active regions. This first paper describes an extensive data set of images from the Atmospheric Imaging Assembly on the Solar Dynamics Observatory curated for large-sample analysis of this topic. Based on (and constructed to coordinate with) the "Active Region Patches" as identified by the pipeline data analysis system for the Helioseismic and Magnetic Imager on the same mission (the "HARPs"), the "AIA Active Region Patches" (AARPs), described herein, comprise an unbiased multiwavelength set of FITS files downsampled spatially only by way of HARP-centered patch extractions (full spatial sampling is retained), and downsampled in the temporal domain but still able to describe both short-lived kinematics and longer-term trends. The AARPs database enables physics-informed parameterization and analysis using nonparametric discriminant analysis in Paper II of this series, and is validated for analysis using differential emission measure techniques. The AARP data set presently covers mid-2010 through 2018 December, is ≈9 TB in size, and is available through the Solar Data Analysis Center.
1.Introduction
Coronal magnetic topology, energetics, and dynamics are all believed to play key roles in triggering, powering, and enabling energetic events. However, large-sample studies of active regions that search for clues as to solar energetic-event productivity have long focused on characterizing photospheric active-region complexity through the analysis of continuum-image and magnetic field data, and the subsequent association of resulting descriptors with flare activity (Sawyer et al. 1986; Zirin & Liggett 1987; McIntosh 1990; Bornmann & Shaw 1994; McAteer et al. 2005; Leka & Barnes 2007; Korsós et al. 2014; Al-Ghraibah et al. 2015; Leka et al. 2018; Kontogiannis et al. 2019). In addition to descriptors based on the morphology of white-light images and the spatial distribution and character of the magnetic fields, such as the Zurich sunspot classification system, photospheric analysis of solar active-region flare productivity has also included plasma velocity and helicity patterns (Welsch et al. 2009; Park et al. 2018, 2021), wavelet and fractal analysis of photospheric images (Abramenko 2005; McAteer et al. 2005; Georgoulis 2012; Al-Ghraibah et al. 2015), and inferred subsurface plasma flows (Komm et al. 2011; Braun 2016).
The focus on photospheric magnetic fields and the drivers of their evolution makes physical sense, as these are ultimately the source of energy to power solar energetic events. The physical interpretation of the state of the photosphere for flare-productive active regions includes highly nonpotential magnetic fields that indicate stored magnetic energy, strong electric current systems and spatial gradients that provide pathways for magnetic reconnection, and emerging flux episodes that can destabilize the system (e.g., Zirin & Tanaka 1973; Canfield et al. 1975; Krall et al. 1982; Hagyard et al. 1984; Smith et al. 1996; Wang et al. 1996; see also Leka & Barnes 2003a, 2007, and references therein). These quantitatively interpretable characteristics provide the physics-based insight that can then guide or constrain numerical modeling and further understanding (recent examples include Threlfall et al. 2017). We intentionally cite some of the originating literature to highlight these physics-inspired investigations that led to, e.g., the Space-Weather HMI Active Region Patch ("SHARP" Bobra et al. 2014) parameters currently published as physical summaries of solar magnetic complexes as a metadata product from the Solar Dynamics Observatory (SDO; Pesnell et al. 2012) Helioseismic and Magnetic Imager (HMI; Scherrer et al. 2012; Schou et al. 2012; Centeno et al. 2014; Hoeksema et al. 2014).
In this series of papers we begin to do the same, quantitatively, for the chromosphere, transition region, and corona. The upper layers of the atmosphere are driven by, and respond to, the photospheric drivers. Case studies of the pre-event chromosphere, transition region, and corona have shown evidence of specific pre-event energization and kinematics, starting with enhanced hydrogen Balmer-series "Hα" emission in flare-imminent active regions (Sawyer et al. 1986; Zirin & Marquette 1991), an increase in chromospheric nonthermal velocities and high blueshifts (Harra et al. 2013; Cho et al. 2016; Seki et al. 2017; Woods et al. 2017), and very localized chromospheric heating (Li et al. 2005; Bamba et al. 2014). Coronal brightness and morphological signals include the energization and increased dynamic behavior of EUV structures ("crinkles"; Sterling & Moore 2001; Joshi et al. 2011; Sterling et al. 2011; Imada et al. 2014, and references therein) and coronal dimmings (Imada et al. 2014; Qiu & Cheng 2017; Zhang et al. 2017) in the hours prior to energetic events.
Large-sample studies of the corona that relate its physical state to energetic-event productivity have generally focused on modeling the coronal magnetic field and observationally inferring relevant topological characteristics, often for coronal mass ejections (e.g., Barnes & Leka 2006; Barnes 2007; Georgoulis & Rust 2007; Kusano et al. 2020; although see Aggarwal et al. 2018, regarding filament dynamics and eruptivity).
Thus far, investigations using the large-sample data available from the SDO Atmospheric Imaging Assembly (AIA; Lemen et al. 2012) in the context of energetic events have been carried out with a primary goal of forecasting, rather than physical understanding. To achieve this expressed objective, machine-learning tools have been invoked (Nishizuka et al. 2017; Jonas et al. 2018; Alipour et al. 2019), although these routinely subject the AIA data to significant degradation by spatial binning, which hampers the ability to discern any small-scale processes. The outcomes have focused on skill scores rather than physical interpretation, as "interpretable machine-learning" approaches have yet to be widely implemented in this context. In other words, on many levels, information about the physics of the corona has been lost or not pursued.
We focus here on the need to move away from case studies and toward the scientific objective of quantitatively characterizing the behavior of the chromosphere and corona using physically meaningful analysis. We ask parallel questions about the chromosphere and corona as were posed in previous works that centered on the photosphere (Leka & Barnes 2003a, 2003b; Barnes & Leka 2006; Leka & Barnes 2007): what are the broad physical differences between "flare-imminent" and "not-flare-imminent" active regions? For the photosphere, we focused on descriptors of the vector magnetic field and parameterizations that quantify, for example, the free magnetic energy available for energetic events. For the present objectives, we focus on both high-frequency kinematic analysis indicative of small-scale reconnection activity, and the hours-long trends in the brightness distributions indicative of growth, decay, or changing temperature distribution, or increasing or decreasing levels of small-scale activity. We design the AARP database to address these objectives with a statistically significant sample. As such, while we are not actively focusing on flare "prediction" per se, the results of these investigations may inform those activities eventually, we propose that the AARP database is designed such that it is appropriate for other, very different, scientific objectives as well.
In this first paper of the series, we describe the data preparation for, and full description of, the AIA Active Region Patch extractions (AARPs; Dissauer et al. 2022; see Section 2). We present global, cycle-scale trends across wavelengths in Section 3, plus some additional analysis aimed to motivate community use of the AARP data set. We leave the full discussion of physics-informed parameterization and results from nonparametric discriminant analysis to Paper II (Leka et al. 2023), and of a statistical analysis of the temperature and density of flare-imminent regions using differential emission measure (DEM) tools to Paper III (K. Dissauer et al. 2023, in preparation).
2.The AIA Active Region Patch (AARP) Database
In this section we describe the AIA AARP database and its construction (see Figure 1). The data cubes were designed to make available a statistically significant sample of solar active-region coronal and chromospheric imaging data with a database of tenable size (noting that the full-disk full-cadence AIA data set for our target period is over 25 PB). The resulting cubes are the basis for subsequent analysis (e.g., Paper II) and are available to the community (Dissauer et al. 2022).
In the present incarnation, we specifically select the cadence and duration of the cubes to match those we have produced for photospheric active-region investigations using the vector-field data from HMI (Leka et al. 2018); future studies may require different such particulars. The infrastructure to produce the present data set as well as others with different such options is available upon request, although it presently relies on specific data access and architecture (see Appendices B, C). Not everyone in the community has access to the hardware and software resources required to produce the AARP data set; we hope it will be of use to the community as a curated, ready-for-analysis resource.
The HMI Active Region Patch ("HARP"; Hoeksema et al. 2014; Bobra et al. 2014) definitions are used for the active-region identification system, and we intentionally include all HARPs. We do not downselect at the active-region level, either for size or activity. HARP designations are based on concentrations of magnetic activity, and the low-activity, small concentrations statistically dominate. As such, we avoid imposing a bias for later analysis, although any comparison of (for example) active versus less-active regions must statistically account for the uneven sample sizes.
2.1.AIA Data
The AARP database comprises coronal and chromospheric time-series patches that are extracted from the full-cadence full-size data of AIA (Lemen et al. 2012) on board NASA's SDO (Pesnell et al. 2012). AIA obtains 40962-pixel full-disk coronal and chromospheric images at 1
5 spatial resolution, sampled at 06 through multiple EUV filter bands at [94, 131, 171, 193, 211, 304, 335] Å with a 12 s cadence, and full-disk photospheric images at ultraviolet [1600, 1700] Å with a 24 s cadence. We embarked on this project to avoid copying all of the data, instead focusing on active regions specifically for the first "downselection" operation of the final database. However, that still required handling a significant amount of AIA data.
NWRA employs the Joint Science Operations Center ("JSOC") "NetDRMS" (Network Data Record Management System) / "RemoteSUMS"(Storage Unit Management System) system to be able to essentially mimic the availability of SDO (and other) data in a manner directly analogous to the host institution, Stanford University. Large data transfers are handled through the "JSOC Mirroring Daemon" (see Appendix B). The AIA metadata reside in the aia.lev1_euv_12s and aia.lev1_uv_24s series, while the images reside in the aia.lev1 series within the NetDRMS/RemoteSUMS system (the aia.lev1 series is largely transparent for data transfers through, e.g., the JSOC "LookData" facility). Details of setting up automatic targeted Java Mirroring Daemon (JMD) transfers and performing the extractions are provided in Appendix B. For small numbers of HARPs or customized extractions, the SolarSoft ssw_cutout_service.pro (Freeland & Handy 1998) is recommended to be used; however, as described in Appendix C, it was not the correct tool by which to develop the AARP database.
2.2.Downselecting in the Spatial Domain: Extracting the Active Region Patches
This data set is created to address the almost untenable requirements of large-sample coronal studies. The AARPs are first and foremost a downselection in the spatial domain, providing subarea extractions from the full-disk AIA image data that rely on the HARP (Hoeksema et al. 2014) metadata (hmi.Mharp_720s series) for target selection, pointing, and field-of-view (FOV) definition.
Because coronal structures extend in height from the solar surface, the extraction box is modified from the original HARP FOV to accommodate this projection effect both when a region is viewed face-on but also as it is viewed away from the disk center (see Figure 2). Hence, we expand the original HARP FOV by 20% in both x- and y-directions, limited to a maximum expansion of 25'' to avoid unwieldy increases in already-large HARPs. Additionally, we augment with an expansion toward the east or the west xex depending on the position of the HARP center with respect to the central meridian in the form of
where xcen is the center of the HARP in degrees. This expansion is limited to a maximum extension of 50'' and is done to accommodate the additional extension of loops viewed from the side as a region approaches the solar limb. Figure 2 shows two examples, a disk-center and a near-limb target, comparing the original HARP (cyan box) and resulting AARP FOVs. While loop tops originating from the target active region may still be cut off, the primary science this data set is designed for concerns relative (versus absolute) changes in AR behavior, and since all regions are treated consistently, this should impose no statistical concerns. In addition, automatically defining an optimal FOV for each individual active region is challenged by (i) AARPs frequently being in close proximity during periods of high activity, and (ii) an optically thin corona, such that there is arguably no optimal way to automatically disentangle all visible loops, their source active regions, and still ensure their tops are always included and correctly assigned.
All HARPs are included regardless of flaring activity levels, size of active regions, or observing angle. The originating HARP bounding-box information is used for the central-time (the "XX:48") image targets. The extracted higher-cadence 11-image time series is coaligned between the (nearly cotemporal) wavelengths and the central-time image using differential rotation to allow for DEM analysis (Figure 3; see also Appendix E) but no further tracking is needed.
2.3.Downselecting in the Temporal Domain
To reduce the data load while preserving the scientific requirements of the research performed in Paper II and further projects, we downsample the data in the temporal domain. The native cadence for AIA EUV images is 12 s, and for the UV images it is 24 s. The AIA Automatic Exposure Control (AEC) system is an observation mode designed to avoid image saturation during solar flares. When the AEC system is invoked, the AEC-regulated images for the EUV data are interleaved with normal-exposure (possibly saturated) images, reducing the effective cadence for consistent-exposure EUV data to 24 s. Native images in data numbers ("DN") or counts can be normalized by the exposure time to obtain consistent per pixel count-rate-based images. For reasons described below, however, we decided to avoid the AEC-regulated images and sample the AIA images at a 72 s cadence.
The image parameterization for the statistical analysis in Paper II and later work uses moment analysis to describe the coronal and chromospheric behavior. During flare times, even for small flares, the higher moments jump dramatically—as expected in the presence of localized yet high-magnitude brightness changes (Figure 4, left four panels). Our scientific focus is not, generally, the flaring plasma itself but the broad active-region behavior. Avoiding the AEC images allows for times of flares to be included with less dramatic impact to the moment analysis. Additionally, the shorter exposures during flaring times lead to lower signal-to-noise ratios (S/Ns) for the larger active-region areas (Figure 4, right panels). Since these regions are our focus, ensuring a consistent S/N is key. Finally, given the format of the AARP FITS files (see Appendix A), avoiding the AEC data allows a simplified metadata format, as the exposure time is consistent for a particular wavelength.
When working with running-difference images, central to our analysis approaches, the native cadence presents two challenges: (i) for consistent-exposure sequential images the resulting S/N was very low, and (ii) when sequential images had varying exposure times, the resulting variation in the images' S/N contributed to even higher noise in the running-difference images. While the running-difference magnitudes increase as Δt, one must also balance this increase in S/N with the ability to evaluate and interpret the resulting images. Experimentation indicated that the full time resolution is unnecessary to capture the short-lived brightenings and kinematics that we wish to examine.
A coarser 72 s cadence for the AARP "high-cadence" bursts of images thus provides a good balance of consistent and acceptable S/N and data set volume reduction. Selecting a 72 s cadence is also a "common factor" between the EUV and UV native sampling, allowing for the data set to sample the different atmospheric layers at the same cadence.
The evolution of the corona over multiple hours is of high interest here. An interval covering ≈6 hr is motivated by numerical models of pre-event evolution and trigger formation (Ishiguro & Kusano 2017; Inoue et al. 2018), and provides sufficient data from which to quantify overall evolutionary trends within an active region. Still, 72 s sampling over 6 hr is a large data load. Hence we further downselect to ≈13 minutes of images at the 72 s cadence, centered hourly 15:48–21:48 UT (seven hourly "bursts" of images, over 6 hr, inclusive). The choice of timing is driven by an already-developed set of HMI vector-field time-series extracted data set (Leka et al. 2018). Centering the data on the "XX:48" times was necessary to avoid those HMI data affected by calibration sequences (taken at 00:00 and 18:00 International Atomic Time (TAI); see Hoeksema et al. 2014). Those studies also had a scientific requirement to minimize the statistical impacts on time-series analysis that were solely due to the spacecraft orbital velocity-induced artifacts (Hoeksema et al. 2014). Finally, we chose 11 images covering 13 m to effectively match the time period over which the input spectra are averaged for the hmi.B _ 720s series magnetograms. In this sense, we sample the upper atmosphere on a cadence appropriate to its physics (reconnection, heating, flows) but carefully pair the extractions to their arguably slower-evolving photospheric driver, at least at the HMI spatial resolution.
Thus, in summary, with the AARPs data set the detailed behavior of the lower solar atmosphere is sampled at a fair cadence (72 s) for 13 m every hour, repeatedly over a quarter of a day for a full 8.5 yr (2010 June–2018 December inclusive) over which we can query about active-region trends using a large-sample approach that enables robust statistical analysis.
2.4.Final Data-cube Preparation
All AIA data are processed through the SolarSoft (Freeland & Handy 1998) routine aia_prep.pro and a correction to account for time-dependent degradation of the instrument was applied using the following form:
where Aeff(tobs) is the effective area calculated at the calibration epoch for tobs, Aeff(t0) is the effective area at the first calibration epoch (i.e., at launch), p1, p2, p3 are the interpolation coefficients for the tobs epoch, and δ t is the difference between the start time of the epoch and tobs (Barnes et al. 2020).
All wavelengths and imaging data centered at each hour are coaligned and differentially rotated to the central time step (usually **:48 UT) using drot_map.pro. The preprocessed output data are therefore ready for running-difference analysis and DEM analysis (see Appendix E). Full FITS headers are generated with wavelength, HARP number, NOAA Active Region (AR) number (from the relevant keywords in the hmi.MHarp_720s series). The seven sets of hourly image sets are saved as extensions each with its own header recording times, number of valid images, and detailed pointing information as well as additional keywords imported from the hmi.MHarp_720s series such as LAT_FWT, LON_FWT and the World Coordinate System (WCS; Thompson 2006) keywords for the target 720s-based midtime. Appendix A contains examples of these AARP fits headers.
The final data product thus comprises a set of eight FITS files, one for each wavelength (by default: 94, 131, 171, 193, 211, 304, 335, and 1600 Å), for each HARP covering 6 hr inclusive (seven hourly samples 15:48–21:48 UT) of 13 minutes of data sampled at 72 s, resulting in 11 images hr−1. The full data set is summarized in Table 1, its archive size being only ≈9.5 TB, in contrast to the size of the full-disk AIA data set for this period, which is ≈25 PB. The power of the data set is its large, yet tenable sample size, designed to enable unbiased statistical analysis on spatially native-resolution images. We next present some topics for which models or case studies may now be tested on a statistically significant, cycle-covering sample.
Table 1.AARP Data Set Summary
| Date Range | HARP Range | NOAA AR Range | Total Number of Samples | Archive Size (TB) |
|---|---|---|---|---|
| 06/2010–12/2018 | 36–7331 | 11073–12731 | 256,976 | ≈9.5 |
Download table as: ASCIITypeset image
3.Analysis and Results
We present here a brief analysis to highlight the breadth of physics investigations that could be performed using the AARP database. We do not discuss flare- or event-related studies in depth, deferring this topic to Paper II (although see Section 3.4).
3.1.Active-region Coronal Behavior with Solar Cycle
With this extensive database we can begin to examine some subtleties of coronal behavior for active regions as a function of solar cycle. Generally speaking the AARPs will not include significant coronal hole areas, and thus will avoid that contamination and source of confusion with regards to interpretation.
In Figure 5 we present density histograms of total emission in AARPs for two AIA bands over most of Solar Cycle 24, as well as the same presentation for the mean emission in the same bands. We focus on 171 Å, which is sensitive to plasma at "quiet" coronal temperatures and transition region emission, and thus is often used to identify coronal magnetic structures such as coronal loops, and the 94 Å band, which is sensitive to hot plasma (Lemen et al. 2012), but which can contain contributions from cooler lines within the bandpass (O'Dwyer et al. 2010; Warren et al. 2012; Del Zanna 2013).
The "total" emission (Σ(I*)) distribution tracks nicely with sunspot number, as expected, since it is an extensive parameter that scales with the size of the box, number of pixels, or more physically, the total amount of magnetic flux present. In both bands we see that the vast majority of AARPs have small total emission, a high density of points that itself tracks the sunspot number generally. In both bands there are also the less-frequent larger-total-emission regions whose distribution tracks the sunspot number even closer. None of this is unexpected.
What is interesting is that while for most "bumps" in total emission, i.e., those for which the eye sees a good correlation with a "bump" in sunspot number, there is good correspondence between the two passbands. However, there are apparently some large active regions for which a large total 171 Å is not matched with a correspondingly large total in 94 Å (e.g., late 2010, late 2017, early 2018) and vice versa (e.g., mid 2012, early 2015). Thus, while well correlated, the total emission displayed in these two channels may not be uniquely correlated, which prompts the questions "why or why not?" One immediate reason is the multithermal nature of both the AIA channels and active regions themselves. Clearly this extensive data set can begin to address the details of active-region coronal emission with, e.g., AR age, magnetic field morphology, and flaring history, and provide constraints for modeling efforts both with regards to intrafilter expected emission and variations with solar cycle. This is especially true since, as discussed in Sections 2.4 and Appendix E, the data are ready for not just hot-plasma isolation (Warren et al. 2012), but full DEM analysis.
The mean emission tells a slightly different story. Figure 5 shows that if there magnetic flux is present (as defined the HARPs and hence the AARPs), there is emission in 171 Å, or μ(I171) ≫ 0, essentially. The same is not quite as true for μ(I94) where the majority of AARPs display a low or near-zero mean emission. This is consistent with μ(I94) having sensitivity to hot coronal emission, which not all active regions contain (especially flare-quiet ones). However, μ(I171)/μ(I94) may not be perfectly constant with solar cycle. These simple findings are consistent with Schonfeld et al.'s (2017) Sun-as-a-star investigation showing that the cool/quiet-Sun corona varied little over the first half of Solar Cycle 24 while the hot component varies strongly with the rise in solar activity. A thorough investigation using the AARP database could provide further physical constraints especially with respect to the AR-based contributions to the two components and possibly with regards to "terminator" analysis (Leamon et al. 2022).
3.2.Active-region Coronal Behavior with Observing Angle
The coronal emission lines are generally optically thin. However, a few of the AIA bandpasses are centered on, or include, optically thick emission (Golding et al. 2017). Additionally, when densities become large such as during solar energetic events, some of the normally optically thin lines in the sampled bandpasses become optically thick (Thiemann et al. 2018). The differences in behavior can be seen when the mean emission levels of different lines are presented according to their location on the disk (Figure 6). The 94 Å channel generally displays a small mean emission that is flat with 
except at the very limb, where the optically thin properties afford a propensity to integrate over more emitting structures along the line of sight.
The He II 304 Å emission is optically thick but has challenging radiative transfer characteristics (Golding et al. 2017), and thus is expected to show a correlation with the observing angle 
(decreasing toward the limb). The density distribution of μ(I304) (Figure 6, right) does show a slight turnover beyond 
but it is extremely slight. There is no severe or obvious attenuation with the observing angle, but the behavior of He II 304 Å emission in active regions may not have yet been studied with large-sample data. There is the possibility of contamination within the AIA filter by optically thin emission that would mask the He II 304 Å behavior, and it also may be contaminated by the different behavior expected by different magnetic structures within the AARP FOVs (Mango et al. 1978; Worden et al. 1999). The detailed behavior of the corona as inferred from optically thin dominated versus optically thick dominated AIA filters for active regions can now, with the AARP database, be studied in detail with large-sample statistical analysis.
3.3.Coronal Behavior with Active-region Emergence/Decay
The relationships (in the spatial and temporal domains) between coronal emission and emerging magnetic flux should provide insights into the energy transfer to the upper atmosphere from the photosphere and below. Similarly, the relationships between coronal emission and decaying active regions should elucidate the final transfer of magnetic energy out of the Sun and the dominant dispersion and dissipative mechanisms at play. The AARPs provide the ability to examine these processes by sampling regions at all sizes and stages of evolution with a curated, large-sample data set.
In Figure 7 we show the transition region/flaring plasma sensitive μ(I131) and upper-photospheric μ(I1600) for two AARPs that are in significantly different stages of evolution as they traverse the disk. AARP 2291 underwent rapid growth starting near the disk center and eventually became NOAA ARs 11631 and 11632. It displays a rapid growth in the mean intensity in the Fe VIII, XXI 131 Å filter emission. AARP 6632 rotated on to the disk as NOAA AR 12556 and decayed to plage by disk center. Its 131 Å filter emission decays to essentially background. The comparison of the two regions in the μ(I1600), however, indicates that (as expected) this emission is optically thick (decreases systematically toward the limb) and is sensitive to the presence of plage rather than sunspots, with little difference between young and old plage.
Obviously we simply present a comparison here in order to encourage interest. We do not, for example, try to perform a larger-sample reexamination of the relationship between flux transport, flux decay, and active-region EUV radiance decay (Ugarte-Urra et al. 2017) but that is one study the AARPs could confirm with larger-sample statistics. Other questions are ready for analysis: Are there significant differences in coronal characteristics between old and new plage? Between emerging regions that will become flare productive and those that will stay quiet?
3.4.Coronal Behavior with Activity Level
The AARP data set was prepared specifically for a large-sample investigation into the coronal behavior as related to flaring activity. We defer our study details and results to Paper II (Leka et al. 2023) but here present a preliminary example of the directions available for study with the AARP data set.
We note that indeed, the AARPs are not extracted according to the time of flares, meaning that they are not designed for superposed epoch analysis (e.g., Mason & Hoeksema 2010; Reinard et al. 2010; Bobra & Couvidat 2015; Jonas et al. 2018). They are, however, extracted for quantitative interpretable analysis and not for flare prediction in and of itself, as has been the focus of many recent studies using large samples of AIA image data (e.g., Nishizuka et al. 2017; Jonas et al. 2018; Alipour et al. 2019).
In Figure 8 we show both the "direct" ("I193") and running-difference ("ΔI193") images, in the 193 Å band, for NOAA AR 11261 (HMI HARP 750, see Figure 4) on 2011 August 3. The quieter example is from the images centered at 21:48 UT, and the "active" example is taken from images centered at 19:48 UT during a C8.5 flare. The two "direct" images look fairly similar, even during a flare, with a small localized brightening being the only easily discernible difference. The running-difference images look very different. In the quiet example, there is a lot of small-scale fluctuation happening even outside the flare time. In the flare case, it is true that the saturated-emission area will show a lack of any dynamics, but do note that as the edges change, there are strong temporal gradients in this area. The expanding loops are also clearly visible in the running-difference image.
The images look similar to the eye, as the flare was not very large, but the images are quantitatively different as summarized by the moments (mean μ, standard deviation σ, skew ς, and kurtosis κ) presented in the table associated with Figure 8. In particular, the mean intensity at the two times is similar for both the direct and running-difference images (the latter both being fairly close to zero relative to the mean intensities) but the higher-order moments both for the direct and especially the running-difference are very different. Physically this implies enhanced localized brightening and kinematics on short timescales with little overall brightness enhancement or mean brightness increase, which would indicate (for example) significant differences in active-region-scale heating or density between the two time periods.
4.Summary and Conclusion
We present here the NWRA AARP database (Dissauer et al. 2022) that well samples the temporal evolution of solar active regions as captured in the E/UV with SDO/AIA over the majority of Solar Cycle 24 from 2010 June through 2018 December. The current version of the database includes daily, 7 hr samples of 13 minutes of images (centered hourly on "*:48 UT" from 15:48 to 21:48 UT) targeting all magnetic patches identified by SDO/HMI, resulting in a total sample size of 256,976 FITS files containing in total almost 20 million individual SDO/AIA images. We show a teaser of the science that could be performed using this data set, including analysis of coronal behavior related to the solar cycle and to the emergence and decay of solar active regions, as well as analysis of center-to-limb variations.
Crucially, the data set is prepared with attention to quantitative analysis methodology. The AARPs data set preserves the native spatial resolution of SDO/AIA (i.e., 15 sampled at 06), in contrast to other studies that downsample the full-disk images to, e.g., 512 × 512 (e.g., Galvez et al. 2019) or use solely an active region's total intensity or maximum intensity in a particular channel (e.g., Nishizuka et al. 2017). Keeping the full spatial resolution allows us to capture small-scale dynamics of the studied active regions (see, e.g., Figure 8), which are otherwise lost due to binning.
However, we note that the data set is (currently) limited to a selected time range (to match the NWRA database of HARP-based photospheric vector-field time-series data; Leka et al. 2018) although with an extended FOV from the original defining HARP bounding box to capture the projections of structures that extend in height. For our initial purpose of investigating unique coronal and chromospheric characteristics of flare-imminent active regions, this is a valid approach since both the short-term changes (over the course of 13 minutes every hour at a cadence of 72 s) as well as longer-term trends (7 hr of evolution per day per HARP) can be studied.
We stress that the AARP data set is ready for machine-learning applications or any other large-sample analysis. Importantly, it has been validated for DEM analysis (see Appendix E), which is a data product to be released in the near future. We note that this data set could be easily expanded to include additional times sampled over each day, even at a different cadence, such that both forecast-mode and superposed epoch mode analysis can be performed on every single flare that occurred since the start of the SDO mission. Although currently beyond the scope of this paper and not needed for the analysis presented in Paper II, such an expansion is quite feasible given the infrastructure NWRA has now developed.
The authors thank the referee for a thorough reading and insightful feedback that helped improve the paper. This work was made possible by funding from AFRL SBIR Phase-I contract FA9453-14-M-0170 (initial exploration), but primarily by NASA/GI Grant 80NSSC19K0285 with some additional support from NASA/GI Grant 80NSSC21K0738 and NSF/AGS-ST Grant 2154653. We extend our sincere thanks to Sam Freeland for his patience and help with the ssw_cutout_service.pro challenges.
Facility: SDO (HMI and AIA), GOES (XRS). -
Software: SolarSoft (Freeland & Handy 1998).
Appendix A: The AARP FITS File Format
AARP FITS files contain in total eight extensions. The first extension, i.e., extension 0, is the primary header and contains no data. A sample primary header is given in Table 2. The primary header is be followed by seven image extensions that contain 11 images per extension, from 15:48 through 21:48 UT. An example for an image extension header is given in Table 3.
Table 2.Sample FITS Header of the Primary Extension (Extension 0) and Its Keywords
| SIMPLE | INT | 1 | |
| BITPIX | LONG | 16 | |
| NAXIS | LONG | 0 | NUMBER OF DIMENSIONS |
| EXTEND | INT | 1 | FITS FILE CONTAINS EXTENSIONS |
| INSTRUME | STRING | 'AIA' | Instrument name |
| TELESCOP | STRING | 'SDO/AIA' | |
| WAVELNTH | LONG | 211 | [angstrom] Wavelength |
| HARPNUM | LONG | 377 | HMI Active Region Patch number |
| NOAA_AR | LONG | 11158 | NOAA AR number that best matches this HARP |
| NOAA_NUM | LONG | 1 | Number of NOAA ARs matching this HARP (0 allowed) |
| NOAA_ARS | STRING | '11158' | List of NOAA ARs matching this HARP |
| NSAMPS | LONG | 11 | Number of images in each extension |
| NTIMES | LONG | 7 | Total number of times in time series |
| T_START | STRING | '2011.02.15_15:48:00_TAI' | Time of first observation in timeseries |
| T_STOP | STRING | '2011.02.15_21:48:00_TAI' | Time of last observation in time series |
| TREC_00 | STRING | '2011.02.15_15:48:00_TAI' | Extension #1 time |
| TREC_01 | STRING | '2011.02.15_16:48:00_TAI' | Extension #2 time |
| TREC_02 | STRING | '2011.02.15_17:48:00_TAI' | Extension #3 time |
| TREC_03 | STRING | '2011.02.15_18:48:00_TAI' | Extension #4 time |
| TREC_04 | STRING | '2011.02.15_19:48:00_TAI' | Extension #5 time |
| TREC_05 | STRING | '2011.02.15_20:48:00_TAI' | Extension #6 time |
| TREC_06 | STRING | '2011.02.15_21:48:00_TAI' | Extension #7 time |
| NXMAX | LONG | 692 | Max NAXIS1 over all extensions |
| NYMAX | LONG | 425 | Max NAXIS2 over all extensions |
| NXMIN | LONG | 688 | Min NAXIS1 over all extensions |
| NYMIN | LONG | 421 | Min NAXIS2 over all extensions |
Note. The information contained includes the wavelength, the HARP number, the associated NOAA Active Region number, and the central time of each image extension, as well as the maximal dimensions of the image data cubes.
Download table as: ASCIITypeset image
Table 3.Sample FITS Header of an Image Extension
| XTENSION | STRING | 'IMAGE' | AIA DATA IMAGE EXTENSION |
| BITPIX | LONG | -32 | |
| NAXIS | LONG | 3 | NUMBER OF DIMENSIONS |
| NAXIS1 | LONG | 692 | X Pixels |
| NAXIS2 | LONG | 421 | Y Pixels |
| NAXIS3 | LONG | 11 | Samples in time |
| T_REC | STRING | '2011.02.15_15:48:00_TAI' | [TAI time] |
| T_IDX | LONG | 0 | Hours since timeseries start |
| VALID | LONG | 1 | Extension contains valid (1) or no (0) data |
| NIMVALID | LONG | 11 | Valid images out of 11 |
| T_IMG00 | STRING | '2011-02-15T15:42:02Z' | [ISO8601] time for image 00 |
| T_IMG01 | STRING | '2011-02-15T15:43:14Z' | [ISO8601] time for image 01 |
| T_IMG02 | STRING | '2011-02-15T15:44:26Z' | [ISO8601] time for image 02 |
| T_IMG03 | STRING | '2011-02-15T15:45:38Z' | [ISO8601] time for image 03 |
| T_IMG04 | STRING | '2011-02-15T15:46:50Z' | [ISO8601] time for image 04 |
| T_IMG05 | STRING | '2011-02-15T15:48:02Z' | [ISO8601] time for image 05 |
| T_IMG06 | STRING | '2011-02-15T15:49:14Z' | [ISO8601] time for image 06 |
| T_IMG07 | STRING | '2011-02-15T15:50:26Z' | [ISO8601] time for image 07 |
| T_IMG08 | STRING | '2011-02-15T15:51:38Z' | [ISO8601] time for image 08 |
| T_IMG09 | STRING | '2011-02-15T15:52:50Z' | [ISO8601] time for image 09 |
| T_IMG10 | STRING | '2011-02-15T15:54:02Z' | [ISO8601] time for image 10 |
| REFHELIO | STRING | 'W21S21' | Heliographic pointing string |
| HARPNUM | LONG | 377 | HMI Active Patch number |
| NOAA_AR | LONG | 11158 | NOAA AR number that best matches this HARP |
| NOAA_NUM | LONG | 1 | Number of NOAA ARs matching this HARP (0 allowed) |
| NOAA_ARS | STRING | '11158' | List of NOAA ARs matching this HARP |
| LAT_FWT | DOUBLE | -20.157000 | [deg] Stonyhurst LAT of flux-weighted cntr pix |
| LON_FWT | DOUBLE | 20.295200 | [deg] Stonyhurst LON of flux-weighted cntr pix |
| WAVELNTH | LONG | 211 | [angstrom] Wavelength |
| WAVEUNIT | STRING | 'angstrom' | Wavelength unit: angstrom |
| T_REF | STRING | '2011.02.15_15:48:00_TAI' | [time] Pointing reference time |
| DATE_OBS | STRING | '2011-02-15T15:48:00.62' | [ISO8601] Date when observation started |
| T_OBS | STRING | '2011-02-15T15:48:02.07Z' | [ISO8601] Observation time |
| EXPTIME | DOUBLE | 2.9012420 | [sec] Exposure duration: mean shutter open time |
| EXPSDEV | DOUBLE | 0.00020700000 | [sec] Exposure standard deviation |
| INT_TIME | DOUBLE | 3.1562500 | [sec] CCD integration duration |
| PIXLUNIT | STRING | 'DN' | Pixel intensity unit |
| DN_GAIN | DOUBLE | 18.300000 | [DN/electron] |
| AECDELAY | LONG | 1536 | AIA_IMG_AEC_DELAY |
| AECTYPE | LONG | 0 | AIA_IMG_AEC_TYPE |
| AECMODE | STRING | 'ON' | AIA_IMG_AEC_MODE |
| AIAWVLEN | LONG | 2 | AIA_IMG_WAVELENGTH |
| CDELT1 | DOUBLE | 0.60000000 | X Pixel size in CUNIT1 units |
| CDELT2 | DOUBLE | 0.60000000 | Y Pixel size in CUNIT1 units |
| CUNIT1 | STRING | 'arcsec' | X pixel units |
| CUNIT2 | STRING | 'arcsec' | Y pixel units |
| CRPIX1 | DOUBLE | -201.50000 | [pix] X location of Sun center in CCD |
| CRPIX2 | DOUBLE | 622.50000 | [pix] Y location of Sun center in CCD |
| CRVAL1 | DOUBLE | 0.0000000 | [arcsec] X origin---center of the solar disk |
| CRVAL2 | DOUBLE | 0.0000000 | [arcsec] y origin---center of the solar disk |
| CROTA2 | DOUBLE | 0.0000000 | [deg] Angle between satellite N and solar N |
| DSUN_OBS | DOUBLE | 1.4773886e+11 | [m] Distance from SDO to Sun center |
| RSUN_OBS | DOUBLE | 971.72022 | [arcsec] of Sun=arcsin(RSUN_REF/DSUN_OBS) |
| RSUN_REF | DOUBLE | 6.9600000e+08 | [m] Reference radius of Sun |
| INST_ROT | DOUBLE | 0.056433000 | [deg] Master pointing CCD rotation wrt SDO z -axis |
| IMSCL_MP | DOUBLE | 0.60075802 | [arcsec/pixel] Master pointing image scale |
| OBS_VR | DOUBLE | -1972.2185 | [m/s] Speed of observer in radial direction |
| OBS_VW | DOUBLE | 28105.801 | [m/s] Speed of observer in solar-W direction |
| OBS_VN | DOUBLE | -1553.5497 | [m/s] Speed of observer in solar-N direction |
| XCEN | DOUBLE | 328.80002 | [arcsec] Ref pixel pointing, arc sec W from Sun center |
| YCEN | DOUBLE | -246.90001 | [arcsec] Ref pixel pointing, arc sec N from Sun center |
| DEGRAD | STRING | 'degradation performed | |
| using calibration | |||
| version 10' | Degradation correction | ||
| PCOUNT | LONG | 0 | |
| GCOUNT | LONG | 1 | |
| CTYPE1 | STRING | HPLN-TAN | Type of X image coordinate axis |
| CTYPE2 | STRING | HPLT-TAN | Type of Y image coordinate axis |
Download table as: ASCIITypeset image
The FITS keyword values are either extracted from the original AIA keywords 3 or computed from the WCS coordinates upon FOV determination and extraction.
Appendix B: Details on Data Acquisition for Remote-DRMS Sites
In order to produce the AARP data set, NWRA made extensive use of being a remote Storage Unit Management System/Data Record Management System (SUMS/DRMS) site for the SDO mission. As such, NWRA can query, access, and, in some aspects, "mirror" the SDO databases in a manner directly analogous to the host institution (Stanford University) through the Stanford Joint Science Operations Center (JSOC). Many institutions have invested in this capability. To produce this curated data set required numerous steps and considerations that we describe here, many of which are unique to the SUMS/DRMS system, but some were unexpected, hence worth providing to the community.
We did find this approach most tenable (versus using the JSOC interfaces including the ssw_cutout_service.pro; see Appendix C), in part because we could simultaneously process multiple HARP-based AARPs that were often present on the disk, reducing the input/ouput (I/O) load and processing time significantly.
Thus, NWRA subscribed to three AIA data series: aia.lev1 (the "data series"), aia.lev1_euv_12s, and aia.lev1_uv_24s (the "header series"). The first queries the actual image data, the latter two query the metadata or header information for EUV (94, 131, 171, 193, 211, 304, 335 Å) and UV (1600, 1700 Å) wavelengths respectively; the metadata alone, which come with the subscriptions, are ≈100 GB. In order to download and then access a complete image including its header information within the SUMS/DRMS system, both the data and header series must be queried; data within JSOC series are referred to by "prime keys" (often, but not exclusively, a reference time such as T_REC), then data are selected further by refining keyword searches. From servers hosting a remote-SUMS/DRMS system, querying the SUMS/DRMS database can be done using the command-line show_info, 4 which is callable from within other codes (e.g., bash, IDL, or Python) and the output used accordingly.
To construct an AARP data set, the needed image data are downloaded first then additional queries are constructed in order to access the correct entries in the SUMS/DRMS for analysis. Image data were batch-transferred using the JMD, written and implemented by the National Solar Observatories. 5 Using an example of the 13 m interval centered at 15:48:00 UT on 2011 January 1 for 94 Å, the following steps are taken:
1.
For 13 m of data at 72 s cadence (see Section 2.3), the "slot time" or first target image is 15:42:00. This somewhat generic timestamp is not accepted by the
aia.lev1series, so the prime keys are obtained by querying the
aia.lev1_
euv_
12series:
$
show_
info key="T_
REC,T_
OBS" "aia.lev1_
euv_
12s[2011-01-01T15:42:00/13m@72s][? QUALITY=0 ?][? EXPTIME>
1.8 and EXPTIME<
3.0 ?][? WAVELNTH=94 ?]">
T_
REC T_
OBS>
2011-01-01T15:42:02Z 2011-01-01T15:42:03.57Z>
2011-01-01T15:43:14Z 2011-01-01T15:43:15.57Z>
2011-01-01T15:44:26Z 2011-01-01T15:44:27.57Z...
since
T_
RECis required for the header series (
aia.lev1_
euv_
12), whereas
T_
OBSis needed for the data series (
aia.lev1). Note that we specify not just wavelength and exposure time (to avoid AEC data), but the data-quality requirement as well. In point of fact, we would query, e.g.,
2011.01.01_
15:42:00/13m@72sinitially and try, e.g.,
2011.01.01_
15:41:48/13m@72sif fewer than the expected number of records were present, or if the difference between
T_
REC, T_
OBSwas greater than 18 s. Note that the wavelength can be used in the header series as a second prime key, but not in the data series, so we generally adopted the practice of invoking the keyword request for all.
2.
Populate the local
aia.lev1series with the requested full-disk data. Setting up the JMD data-transfer request requires the
sunum(storage unit unique identifier that is associated with a given data series query) and
recnum(a record number corresponding to each record in the storage unit) available from the data series using the prime keys from above:
$
show_
info -rS key="FSN,T_
REC,T_
OBS" "aia.lev1[2011-01-01T15:42:03.57Z/13m@72s][? QUALITY=0 ?][?EXPTIME>
1.8 and EXPTIME<
3.0?][?WAVELNTH=94?]">
recnum sunum FSN T_
REC T_
OBS>
67456313 190140361 18250728 2011-01-01T15:42:04Z 2011-01-01T15:42:03.57Z>
67456361 190140436 18250776 2011-01-01T15:43:16Z 2011-01-01T15:43:15.57Z>
67456409 190140506 18250824 2011-01-01T15:44:28Z 2011-01-01T15:44:27.57Z...
Note here: there can be more than one record per
sunum(depending on the series), the returned
T_
RECdiffers from that returned from the header series, and
sunuminformation is not available from the header series.
The Filtergram Sequence Number ("FSN") integers become important later, but can be queried and saved at this step. For each imaging instrument of SDO (e.g., HMI or AIA) the FSN is a unique number for each original image produced by the instrument; we use them to validate the correspondence between the two series. The
aia.lev1series is not designed to accept multiple-entry requests ("
13m@72s"), and when it fails, the queries must be handled individually.
3.
We query for the metadata entries with the new
T_
RECto confirm the FSN number correspondence:
$
show_
info key="FSN, T_
REC,T_
OBS" "aia.lev1_
euv_
12s[2011.01.01T15:42:04Z/13m@72s][? QUALITY=0 ?][?EXPTIME>
1.8 and EXPTIME<
3.0?][?WAVELNTH=94?]">
FSN T_
REC T_
OBS>
18250728 2011-03-01T15:42:02Z 2011-03-01T15:42:03.57Z>
18250776 2011-03-01T15:43:14Z 2011-03-01T15:43:15.57Z>
18250824 2011-03-01T15:44:26Z 2011-03-01T15:44:27.57Z...
4.
Upon data transfer, we query the data series again, for the location of the files using the FSN numbers returned by the
aia_
lev1_
euv_
12s. Every remote-SUMS/DRMS system will have its own upper-level structure, but the
sunumis consistent between them and is integral to the location:
$
show_
info -qP 'aia.lev1[][18250728, 18250776, 18250824, ... ]' seg=image_
lev1>
/nwra/SUMS/SUM0/D190140361/S00000/image_
lev1.fits>
/nwra/SUMS/SUM0/D190140436/S00000/image_
lev1.fits>
/nwra/SUMS/SUM0/D190140506/S00000/image_
lev1.fits...
5.
We proceed to prepare the data by specifying wavelength, etc., keywords, and the file location (e.g.,
/nwra/SUMS/SUM0/D190140361/S00000). Invoking
drot_
map.pro, degradation, FOV determination and extraction, coalignment, etc., requires numerous additional keyword-queries through
show_
infoto the
hmi.Mharp_
720sseries. Along the processing, intermediate steps are saved as temporary
IDL"idl.sav" files for subsequent assembly into the AARP extraction FITS files.
The data transfers and referencing were indeed time consuming to automate, as these series are constructed with different structure than, e.g., the HMI series. However, having the full-disk data locally was key while we formulated the spatial extraction details (see Section 2.2) and the sensitivity corrections (see Section 2.1), because we were not required to re-request full-disk data repeatedly as the respective approaches were refined.
Appendix C: Notes on Using the SolarSoftware Cutout Service
Our initial attempt to construct an AARP time-series HARP-congruent data set over 2014–2015 relied on JSOC and its online interfaces including through the Solar SoftWare (SSW) IDL package (Freeland & Handy 1998). We used the ssw_cutout_service.pro tool written by Sam Freeland at LMSAL. This approach is advantageous for many studies, as the AARP-required data can be downloaded and assembled without the effort and resource allocation necessary to subscribe to the voluminous DRMS series, instead utilizing LMSAL's own subscription. Further, the LMSAL-generated SSW routines are equipped to deal with the subtleties of AIA data processing, such as how to set the cadence to avoid AEC frames.
We automated various steps required to implement the cutout service. Drivers were written specifically to
1.
Get HMI metadata, using
show_
infoto select desired times, and patch size/location coordinates:
SSWIDL>
fovxa=abs(crsize1*cos(crota2)-crsize2*sin(crota2))*cdelt1 SSWIDL>
fovya=abs(crsize2*cos(crota2)+crsize1*sin(crota2))*cdelt12.
Parse the AIA metadata, and select start/end times to avoid AEC images, getting specified URLs:
SSWIDL>
ssw_
jsoc_
time2data, t0, t1, index, urls, /urls_
only, wave=wave, ds='aia.lev1_
euv_
12s'3.
Queue AIA cutout exports corresponding to HARPs, for the given day using
ssw_
cutout_
service, invoking the option of an email when the job is complete:
SSWIDL>
ssw_
cutout_
service, t0n, t1n, query, stat, ref_
time=tref, fovx=fovxa, fovy=fovya, wave=wavelengths, ref_
helio=ref_
helio, instrument='aia',aec=0, cadence=cadencestr, description=descr, max_
frames=10000, email=email, /RICE4.
Wait for the exports to complete, and move exported FITS files into place;
5.
Read in the FITS files for each HARP/wavelength/time;
6.
Call
aia_
prepto perform alignment/adjustments on each data cube:
SSWIDL>
aia_
prep, indexIn, dataIn, index, data, /cutout, index_
ref=indexIn[refidx]7.
Save the final processed data cubes to FITS (and IDL.sav) files (one per HARP per day).
All of the above steps were automated, initiated by specifying a date. The 94, 131, 171, 193, 211, 304, and 335 Å filter data can be processed together, but the different cadence and data series means the 1600 (UV) must be processed separately.
To produce the full AARP database, however, we came to a few situations that nudged us to the approach in Appendix B. The main failure point stemmed from the (lack of) robustness of data retrieval and the needed information not being available to easily detect and validate failures. Data-transfer throttling—which can occur on either end of the transfer, and it is not always easy to assess where it occurs—plus "standard-issue network intermittency" often created problems accessing records at the query step. When records then came up absent, it was not straightforward to determine exactly which were missing.
Additionally, the cutout service proved to be slower than expected, primarily due to the need to request data by individual HARPs and individual days, which necessitated repeated full-disk I/O at the host institution when, for example, there were multiple HARPs on the same day and time step. We found significant speed up by internally accessing the full-disk data and performing cutouts once for all targets per day and time step.
The ssw_cutout_service.pro works very well for requests that focus on limited time periods or a small number of targets. It allows for fully customized FOVs, significant flexibility on cadence and filter selection, and provides up-to-date preprocessing. Some challenges that were faced during this exploration were beyond the purview of ssw_cutout_service.pro (e.g., the too-generic JSOC email content). Thus, while the exploration into using the ssw_cutout_service.pro involved significant investment, it also provided good education and preparation for the switch to relying upon NWRA's remote-DRMS/SUMS subscription (see Appendix B).
NWRA will share ssw_cutout_service.pro-related driver codes with the community upon request.
Appendix D: Pathology
The AARP database (Dissauer et al. 2022) follows the HARP definitions (e.g., Hoeksema et al. 2014). By default, all HARPs that were identified between 2010 June and 2018 December are included. However, the HARPs include "active pixel" maps within the bounding boxes that are unique to each HARP, even if two (or more) HARP bounding boxes overlap. We do not include the masks (such as those in the hmi.Mharp_720s bitmap.fits segment) as they would only refer, roughly, to the footpoints of the coronal structures. Hence, some AARPs may spatially overlap, or be fully included in one another. We do not remove or try to disentangle these situations at all, but do note that generally the "subsuming" situation is rare and generally for which one of the AARPs is very small.
However, we did find pathological HARPs that needed to be excluded, resulting primarily from periods when the HMI magnetograms were corrupted in such a way as to flag large sections of limb as "active pixels," which then propagated to an HARP definition even though there were otherwise no magnetic concentrations. Of note, the following HARP numbers were removed from the AARP database:
In addition, HARP 4225 was also initially defined with a corrupted bounding box but in fact there were two NOAA active regions within, although they took up only a small fraction of the original bounding box. We redefined an HARP-appropriate bounding box according to the active pixel masks (hmi.Mharp_720s bitmap.fits segment; see Hoeksema et al. (2014)), and then extended that FOV as per described in Section 2.2 for inclusion in the AARP database.
Appendix E: Differential Emission Measure Validation
The AARP data set is corrected for instrument degradation. In general, we use the DEM code by Cheung et al. 2015 to calculate emission measure, temperature, and density maps (Paper III). By definition, this particular DEM code handles instrument degradation by correcting the temperature response functions returned from aia_get_response.pro using the image observation times.
Here, we demonstrate that the DEM calculation using the AARP data set as input, and turning "off" the degradation correction, provides similar results as using the original JSOC output where the degradation has not yet been performed and allowing the DEM code to make these corrections (Figure 9). The differences are displayed as both spatial maps and a correlation plot, and do not exceed the [−0.1, 0.1] range in a pixel-by-pixel comparison. This error range is in agreement with uncertainties for the sparse solution of Cheung et al. (2015) when compared to a "ground truth" (see their Figures 2 and 3).
