# Introduction This bundle contains Planetary Data System 4 (PDS4) versions of calibrated (L2C) data products from the Chinese Lunar Exploration Program (CLEP)'s Chang'e-1 (CE-1) and Chang'e-2 (CE-2) Microwave Radiometer (MRM) instruments, along with new products derived from those data. We produced these products with NASA support under grant #80NSSC20K1430. Note that NASA is generally disallowed from funding collaborations with Chinese entities[^1], so we conducted this effort without the participation of the China National Space Administration (CNSA). We thank NASA for this financial support. We also thank CNSA, the CE-1 and CE-2 archival teams, China’s Lunar and Planetary Data Release System (CLPDS), and the National Astronomical Observatory of China (NAOC) for making these data publicly available to the global scientific community. We chose to archive these products in the PDS for two major reasons: * Almost all scientists who use these data in time-series form want to analyze data taken across an entire mission. The NAOC-held L2C products, although written in a clean and consistent format, require laborious preprocessing before they can be used in these sorts of analyses. We have created tables that aggregate the MRM L2C data and add precomputed values for analytically crucial quantities like local time. We believe these products will improve time-to-value for scientists who wish to incorporate the MRM data into new investigations. * Many scientists prefer to work with data of this type in map-projected rather than time-series form, and the map-projected products we include in this bundle are significantly higher-quality, better-resolution, and more complete than the limited set of map-projected products made available by the NAOC. # Context ## Missions and Hosts The CE-1 and CE-2 missions comprised the first phase of CLEP’s Chang’e Project. They were intended to test navigation and communication systems, produce high-resolution maps of the lunar surface, conduct assorted geophysics and heliophysics investigations, and more generally enable surface operations planned for later phases of the Project. They each hosted a similar suite of surface-observing instruments: a visible-band stereo camera (CCD), a laser altimeter (LAM), X-ray and gamma-ray spectrometers (XRS and GRS), and a 4-channel microwave radiometer (MRM). They also each hosted a heliophysics package: a high-energy particle detector (HPD) and a solar wind ion detector (SWID). The CE-2 CCD’s design differed significantly from the CE-1 CCD’s, but all other instruments were identical in construction, subject to few if any modifications[^2]. In addition to these instruments, CE-1 also hosted a 32-band NUV-NIR imaging interferometer (IIM). (Huixian et al., 2005; Zuo et al., 2014). CE-1 entered lunar orbit on 2007-11-05. It ceased operation on 2009-03-01 upon planned impact with the lunar surface. (China National Space Administration, 2009; Wang et al., 2009). CE-2 entered lunar orbit on 2010-10-06. On determining that CE-2 would have much more fuel remaining by the end of its primary mission than initially expected, CNSA designed an extended mission that re-tasked CE-2 for deep space communications testing and small body observations. CE-2 exited lunar orbit on 2011-06-08 and subsequently performed several deep space operations. (NSSDCA, 2022; Zuo et al., 2014). These operations included a flyby of the asteroid Toutatis; however, Toutatis was only briefly within the effective range of the MRM, not long enough to collect clear Toutatis data (Li, 2013). CE-2 left communications range in 2014, and CNSA estimates that it may reenter communications range as early as 2027 (Jones, 2021). ## Geophysical Applications The MRM channels fall primarily into the super high frequency (SHF) or "centimeter" band (the 37 GHZ channel technically edges into the extremely high frequency (EHF) band). A handful of other planetary science investigations have taken SHF data, including Juno MWR, Rosetta MIRO, and Cassini RADAR. However, only the CE-1 and CE-2 MRMs have systematically imaged a large body's surface in this band (not counting coarse observations by ground-based radiotelescopes). SHF technologies are robust, cheap, and extremely mature; terrestrial applications range from aviation radar to short-range high-bandwidth communications (like WiFi) to ground-based weather radar to kitchen appliances. It is also extensively used in radioastronomy. Because water absorbs SHF very effectively, SHF radiometry has limited use in terrestrial geology. However, in the absence of water, passive SHF radiometers can essentially “see” into the top few meters of a body’s surface. Emissions in this band are principally due to blackbody radiation from variations in physical temperature, and because the depth of this “vision” is frequency-dependent (lower frequencies see deeper into the surface), a multiband radiometer can effectively measure subsurface temperature gradients. Effective measurement depth is affected by dielectric material properties as well as frequency, so regional variations in heating and cooling over a diurnal cycle can also give insight into compositional properties, particularly metallicity. See Siegler et al., 2020 and Siegler et al., 2023 for further discussion and examples of application. ## Antenna Characteristics The MRM has 4 channels, centered respectively at 3, 7.8, 19.35, and 37 GHz, with respective bandwidths of 100, 200, 500, and 500 MHz. Nominal radiometric resolution is 0.5 K for all channels. Angular full width at half maximum (FWHM) is ~13 degrees for channel 1 and ~10 degrees for the higher channels, although the main beams are platykurtic and slightly asymmetrical. The first side lobes are larger and highly asymmetric. Outer side lobes are not characterized in the literature. Please refer to Wang et al., 2010b for a more complete description of antenna patterns. Orbital height varies significantly within each data set. CE-1 mean orbital height was ~185 km with a standard deviation of ~32 km; CE-2’s was ~100 km with a standard deviation of ~10 km. This means that surface footprint size is not constant within either data set. However, at mean CE-2 orbital height (~100 km), the diameter of the surface footprint of the main beam at 50% of maximum response (i.e. FWHM) is ~23 km for channel 1 and ~17 km for the higher channels; at 10%, diameter is ~40 km for channel 1 and ~25 km for the higher channels. At mean CE-1 orbital height (~185 km), FWHM is ~45 km for channel 1 and ~30 km for the higher channels; diameter at 10% is ~70 km for channel 1 and ~45 km for the higher channels. (All of these figures assume perfect nadir pointing.) Various values for the MRM’s “spatial resolution” are quoted in the literature without derivation; they are usually close to FWHM at mean orbital height. Note, however, that it is not really meaningful to assign a single “spatial resolution” value to the MRM data. The effective resolution of the data set is much higher than the “resolution” of a sample considered in isolation, and is determined as much by sample density as by footprint size. Furthermore, both mean sample density and mean footprint size vary greatly between regions. ## Temporospatial Scope The CE-1 MRM L2C data set contains samples taken between 2007-11-27 and 2009-01-24. Most of these samples exhibit a regular observational cadence: 5 samples separated by 1.6-second intervals followed by 1 at a 3.6-second interval[^3]. The record has two gaps of ~3 months, several other gaps of 1-5 days, and 92 gaps of ~15 minutes. The data set contains ~6.3 million samples in total. The CE-2 MRM L2C data set contains samples taken between 2010-10-15 and 2011-05-20. They exhibit the same observational cadence as the CE-1 data. The record has several time gaps ranging in size from ~15 minutes to ~7 days. There is also a ~2-week interval during a major orbital pattern change in February 2011 during which the mission flagged all data as off-nominal. (The calibration for the mission-flagged data appears to us as possibly compromised, perhaps along with geopositioning.) The data set contains ~8.7 million samples in total, and ~7.5 million excluding the 2-week period of mission-flagged data. Both MRM data sets include samples of the entire lunar surface, and each MRM sampled most regions multiple times in adjacent orbital passes and in separate sets of passes at other times of day. Note, however, that because CE-1 held a roughly noon-midnight orbit throughout the observational phase of the mission, the CE-1 MRM data include few samples taken during the lunar “morning” or “evening,” particularly in lower latitudes. Conversely, because CE-2 changed its orbital pattern more dramatically over the course of its mission, moving from a noon-midnight orbit to a terminator orbit back to a noon-midnight orbit, the CE-2 MRM data include samples taken across the entire diurnal cycle (although sample density is greatest near midnight and noon). ## Calibration MRM calibration relied on constants determined during ground testing along with regular warm and cold calibration measurements taken in flight (see Wang et al., 2010b for details). The CE-2 calibration system was identical in basic architecture, but ground calibration was significantly more rigorous and the space-pointing cold horn was tilted up approximately 15 degrees to address concerns about possible contamination (Feng et al., 2013; also compare Wang et al., 2010a, 2010b). Both the CE-1 and CE-2 MRM data exhibit significant offsets in temperature between orbits (even adjacent orbits), and sometimes between groups of samples within the same orbit. The data are also quite “noisy”, especially the CE-1 data; what we interpret as "noise" includes sample-to-sample variations, a number of obviously “bad” orbits, and many outlier points within otherwise “good” orbits. Even with best-effort data cleaning, many uncertainties remain, and we recommend treating relative temperature values as more reliable in general than absolute values. Errors in calibration appear to be partly responsible for these offsets, noise, and inconsistencies. First, both MRM cold horns appear to have experienced contamination by emission from the lunar surface. Various approaches to addressing this phenomenon have been proposed (see for instance Hu et al., 2022, 2017; Tsang et al., 2014), but no complete “corrected” version of either data set has been published. Second, the CE-2 MRM exhibits changes in absolute calibration across the course of the mission (Siegler and Feng, 2017; St. Clair et al., 2022). Physical modeling suggests that, while both MRMs were subject to surface contamination effects, this CE-2 specific phenomenon likely had a separate physical cause (Hu and Keihm, 2021). Feng et al., 2020 postulated solar contamination of the cold horn (either via direct radiation or indirect physical heating) as the most plausible explanation. This is supported by a strong relationship between measured temperatures and the angle between the CE-2 orbital plane and the Sun-Moon axis (“orbit angle”). This relationship is clearest (and perhaps only present) for absolute orbit angles below ~45 degrees. This is to say that as CE-2 shifted away from a noon-midnight orbit towards a terminator orbit, its temperature measurements dropped, and they rose again as it shifted back towards a noon-midnight orbit. The peak effect size in channel 1 is ~13K (towards the middle of the large range of mission-flagged data in February 2011). This is far too large to be explained by diurnal effects. It also cannot be explained by sampling bias, as it appears across many different regions of the surface. This orbit angle-dependent effect has high statistical error. This is due in part to the confounding effects of latitudinal cooling, regional temperature variability, and the diurnal cycle; however, it also appears to have a complex relationship with orbiter attitude and Sun-relative orientation. For instance, it is asymmetric with respect to the sign of orbit angle (larger as the orbiter approached the terminator); i.e., the mean offset is larger at -15 degrees orbit angle than +15 degrees orbit angle. It is also worth noting that measurements from the final phase of mapping appear to be slightly *warmer* on average than measurements taken during early orbits with similar absolute orbit angles. Due to these complexities, we consider it unlikely that this calibration offset can be seriously investigated or addressed without data on orbiter attitude, which CNSA has never released. ## Georeferencing The MRM L2C tables include latitude, longitude, and orbital height values for each sample. Neither documentation or published literature specify a frame of reference for the lat/lon values, but comparison to other orbital data products suggests that they are referenced to MOON_ME or IAU_MOON. Because the millidegree offsets between MOON_ME and IAU_MOON are below even the best-case spatial resolution of the data, we have arbitrarily chosen to treat the coordinates as MOON_ME. In the published literature, the lat/lon values are treated both as (1) boresight intercept point coordinates and(2) directly nadir to the spacecraft. In other words, analyses of these data assume that the MRM was always perfectly nadir-facing, such that its boresight intercept point is also the orbiter’s subsurface point. However, because, both orbiters performed somewhat complex maneuvers, the MRM was almost certainly *not* directly nadir-facing for most measurements. Published literature and documentation do not describe how–or if–these georeferences were corrected for orbiter attitude, but we have observed quasi-periodic oscillations in the time-series data that are reminiscent of typical patterns in gyroscopic attitude determination and control systems (ADCS). These oscillations typically exhibit much higher amplitude past +/- 75 degrees latitude. This suggests the possibility that the orbiter performed regular corrective maneuvers as it passed the poles, and that attitude changes from these maneuvers propagated into measured temperatures, perhaps via changes in the MRM surface footprint or offsets in calibration or both. They are also much stronger in some orbits than others. They may be responsible in part for the higher “noisiness” of the CE-1 data, as CE-1’s greater average orbital height would have tended to increase the effects of pointing error. Unfortunately, CNSA has not released orbiter attitude data, so we are unable to investigate these possibilities in detail. However, we recommend exercising caution when using data taken past +/- 75 degrees latitude. We believe that the data striping and “jitter” visible in some portions of the map products may be related either to geopositioning/pointing errors or the calibration issues described above. There are also two effects that we believe may be due solely to geopositioning/pointing errors. First, there are small apparent spatial misalignments between some surface features in some time bins. Second, the effect of topography on observed temperatures appears to become much stronger at higher latitudes, much more so than can be explained by changes in effective local time due to topographic dependence of hour angle (especially in the lower channels). # Bundle Directory Structure ## /miscellaneous Root directory of the bundle's miscellaneous collection. This collection includes products that were used to help derive some products in the data collection, but that were not themselves derived from CE MRM data. The root directory contains two products used to help derive the products in this collection's "tbmod" subdirectory. ## /miscellaneous/tbmod Thermal brightness model maps. ## /browse Bundle's browse collection. ## /browse/[image type] Root directories per image type for browse images generated from MRM observational data, where "image type" is "datminus", "latshift", "stdev", or "temp". ## /browse/[image type]/[orbiter] Per-orbiter root directories for browse images of a particular type, where "orbiter" is "ce1" or "ce2". ## /browse/[image type]/[orbiter]/[channel] Per-channel directories containing browse images of a particular type generated using data from a particular orbiter, where "channel" is "t1", "t2", "t3", or "t4", corresponding respectively to the MRM channels centered at 3, 7.8, 19.35, and 37 GHz. ## /browse/tbmod Root directory for browse products generated from "tbmod" data products. As the "tbmod" products were generated from non-MRM observational data, this directory does not have per-orbiter subdirectories. ## /browse/tbmod/[channel] Per-channel directories containing browse images of modeled brightness temperatures, where "channel" is "t1", "t2", "t3", or "t4", corresponding respectively to the MRM channels centered at 3, 7.8, 19.35, and 37 GHz. ## /data Root directory for the bundle's primary data collection. Also contains two observational data products: ce1_mrm.fits/xml and ce2_mrm.fits/xml. These are binary tables which contain concatenated versions of, respectively, the CE-1 and CE-2 L2C MRM data, along with derived quantities necessary for standard corpus-level analyses of the data. ## /data/maps Map-projected CE MRM data products. ## /data_source Root directory for the bundle's data_source collection. ## /data_source/[orbiter] Per-orbiter root directories for the L2C source data products, where "orbiter" is "ce1" or "ce2". ## /data_source/[orbiter]/[YYYY][MM] Directories containing PDS4-labeled versions of the NAOC-archived L2C data files, split by 4-digit year and 0-padded 2-digit month. ## /document Root directory for the bundle's document collection. This collection has no subdirectories. # Map-projected Products ## Overview The bundle's data collection includes 4 types of map-projected products: "temp", "latshift", "datminus", and "tbmod". This section describes their contents and organization. More details on the process used to derive them are given in the "Deconvolved Map Product Processing" section below. Note that the "tbmod" maps are in the bundle's miscellaneous collection, as they do not use CE MRM observational data as an input. The others are in the bundle's data collection. ### Projection These products contain maps of the lunar surface in equirectangular projection at 32 pixels per degree, centered on 0 degrees latitude / 0 degrees longitude. All maps cover -180 to 180 degrees longitude. "temp" and "latshift" products cover -75 to 75 degrees latitude; "datminus" and "tbmod" products cover -70 to 70 degrees latitude. All maps are coregistered with one another aside from the aforementioned 10-degree latitude crop. All products also include longitude and latitude arrays corresponding to their x and y axes. Latitude and longitude values are given in the lunar mean Earth/polar axis reference frame (SPICE MOON_ME). ### Excluded Data We excluded some samples from the maps, including samples flagged by the mission, samples from a handful of orbits with errors in temperature or geopositioning, samples from a handful of extremely “noisy” orbits, samples with physically implausible temperatures or implied temperature gradients, and a few groups of samples with duplicate timestamps. Also, the CE-2 set has a large contiguous range of mission-flagged orbits that appear to have persistent calibration errors. We extended this range on both sides to exclude orbits in which those calibration errors appear to persist (although at lower intensity). See the “Calibration” section above for further discussion of this phenomenon. The concatenated data tables include quality flags that describe the specific rationales for sample exclusion. See the “Columns” subsection of the “Concatenated Data Tables” section below for a full description of these flags. ### Physical Product Structure Each map product consists of a FITS file and a PDS4 label file. Each product's individual maps, as well as its latitude and longitude values, are stored as binary arrays in separate HDUs of its FITS file. We chose this physical layout because storing multiple named array objects per file simplifies logical product structure and physical file access, and comes with almost no performance tradeoffs due to the affordances of the FITS format. ### Usage Notes * Users who want to use these maps with dedicated GIS software can convert them to GeoTIFF using the `gdal_translate` program included with [GDAL](https://gdal.org/en/latest/). The GDAL PDS4 driver will include the map projection specified in the .xml label in the output GeoTIFF. [Detailed instructions for use of the GDAL PDS4 driver are here.](https://gdal.org/en/latest/drivers/raster/pds4.html) * For example, `gdal_translate PDS4:ce1_t4_temp_32ppd.xml:1:3 ce2_t4_temp_4_6.tif` will produce a GeoTIFF from the third raster array in ce2_t4_temp_32ppd.fits. * For a list of which PDS4 object names correspond to which GDAL 'subdataset' number (the '3' in the '1:3' part of the `gdal_translate` command above) in a particular file, run `gdalinfo` on the .xml label file. * FITS normally uses bottom-to-top array order, but we wrote these maps top-to-bottom for compatibility with simple raster display tools like PDS4 Viewer and `pdr`. However, this means that users of FITS-specific display tools like `fv` or DS9 may see the maps flipped bottom-to-top (i.e. South up). If this occurs, simply mirror the maps about the x axis / equator. * Note that this has no effect on GDAL, because its PDS4 driver respects the `vertical_display_direction` keyword and will interpret the maps North up as intended. ### Data Objects All maps are stored as 2-D arrays of 16-bit MSB integers. Scaling factors and offsets are provided to convert them to physical units; standard FITS tools will automatically convert them to 32-bit floats in physical units on load. All map array values are in Kelvin, although their physical references vary; they are described in the "Map Product Types" section below. Latitude and longitude are stored as 1-D arrays of IEEE 32-bit MSB floats. Units are degrees. We chose to store them as 1-D arrays because, since the maps are in equirectangular projection, every grid cell in a particular column has the same longitude value, and every grid cell in a particular row has the same latitude value. This means that storing them as 2-D arrays increases file size but provides no additional information. Users who wish to produce traditional "backplanes" may simply repeat the latitude array across the x axis and the longitude array along the y axis. Each of these array objects also has an associated FITS header object. Because we included all metadata necessary for interpretation in the PDS4 labels, we left these headers fairly minimal. They primarily contain values that FITS software needs to correctly load the arrays and values required for compliance with the FITS standard. They also contain a handful of parameters used by our processing software. Because some of these values are not required for interpretation, we did not include them all in the PDS4 labels. However, they may be of interest to users who wish to run the software themselves. ### Time Binning Most scientific uses of the MRM data require knowledge of local time in order to account for or analyze brightness temperature variation over the diurnal cycle. For this reason, we defined an array of 2-hour local time bins and derived separate maps from data taken within each of these bins. (Maps in "tbmod" products contain brightness temperatures modeled at the center of the bin.) The array of local time bins starts at midnight. Time bins are described in HDU and PDS4 object names by their left and right edges (see the "HDU / PDS4 Object Names" section below). Note that while all map products define the same array of bins, products derived from CE-1 data do not actually include maps for the 0400-0600, 0600-0800, 1600-1800, or 1800-2000 bins. This is because CE-1 maintained a noon-midnight orbital pattern throughout the entire mapping phase of the mission and took too few observations in these intervals to permit production of meaningful maps. ### HDU / PDS4 object names #### Map HDUs `[maptype]_[binstart]_[binstop]` "maptype" may be "TEMP", "STDEV", "LATSHIFT", "DATMINUS", or "TBMOD". These are described in more detail below. "binstart" and "binend" are the left and right edges of the map's time bin. They may be given as either 1 or 2 digits. Examples: * An HDU named `TEMP_0_2` contains a TEMP map derived from data taken between midnight and 2 AM local time. * An HDU named `LATSHIFT_12_14` contains a LATSHIFT map derived from data taken between noon and 2 PM local time. #### Latitude/Longitude HDUs HDUs containing latitude and longitude arrays are simply named "LATITUDE" and "LONGITUDE". #### PRIMARY HDUs The FITS standard requires all FITS files to begin with an HDU named "PRIMARY". Because we wanted to give every PDS4 data object a meaningful name, and wanted these names to correspond directly to the HDU names (i.e. EXTNAME keyword card values) of the FITS files, we followed the common practice of placing a "stub" PRIMARY HDU at the beginning of each of these FITS files. These HDUs contain no data. They serve only to allow software tools to recognize the files as FITS. ## Map Product Filenames Filenames of "temp", "latshift", and "datminus" products follow the pattern: `[orbiter]_[channel]_[product type]_32ppd.[extension]` * "orbiter" may be "ce1" or "ce2". It indicates which orbiter's MRM data were used to produce the product. * "channel" may be "t1", "t2", "t3", or "t4", corresponding respectively to the MRM channels centered at 3, 7.8, 19.35, and 37 GHz. Literally, it is the name of the column that the deconvolution program(`run_deconv`) loaded from the corresponding binary source data table ( [orbiter]_mrm.fits) described in the "Source Data Products" section below.. * "product type" may be "temp", "latshift", or "datminus". * "extension" may be "xml" (the product's PDS4 label) or "fits" (the product's FITS file). Because "tbmod" products are derived from non-MRM observational data, their filenames do not include an orbiter code, and instead follow the pattern: `[channel]_tbmod_32ppd` "channel" here has the same basic meaning as in the MRM-derived map products. Literally, it indicates the frequency value given to the brightness temperature modeling program (`map_brightness`). *Note: the "32ppd" in these filenames does not imply that this bundle contains similar map products at other resolutions. We included this field partly for consistency with certain model inputs and partly because we may, in the future, choose to produce map products at other resolutions.* ## Relationships Between Derived Map Products Each orbiter/channel combination has three associated map products: a “temp” product, a “latshift” product, and a “datminus” product. Each channel also has an associated “tbmod” product. “temp” products are direct outputs of the deconvolution pipeline. They are derived from the time-series values in the concatenated data table of their corresponding orbiter. “latshift” products are derived from the “temp” product of the same orbiter and channel. “datminus” products are derived from the “temp” product of the same orbiter and channel and the “tbmod” product of the same channel. Although the source products for “temp”, “latshift”, and “datminus” FITS files are explicitly stated in their PDS4 labels, they can also be determined simply from their filenames: * `[orbiter]_[channel]_temp.fits` is derived from `[orbiter]_mrm.fits`. * `[orbiter]_[channel]_latshift.fits` is derived from `[orbiter]_ [channel]_temp.fits`. * `[orbiter]_[channel]_datminus.fits` is derived from `[orbiter]_ [channel]_temp.fits` and `[channel]_tbmod.fits`. The time suffixes of individual HDUs can then be used to match HDUs between these products. Examples: * The `LATSHIFT_0_2 HDU` of `ce1_t2_latshift_32ppd.fits` is derived from the `TEMP_0_2` HDU of `ce1_t2_temp_32ppd.fits`. * The `DATMINUS_12_14` HDU of `ce2_t1_datminus_32ppd.fits` is derived from the `TEMP_12_14` HDU of `ce2_t1_temp_32ppd.fits` and the `TBMOD_12_14` HDU of `t1_tbmod_32ppd.fits`. “tbmod” products are derived from non-MRM observational data. Specific source products are described in their PDS4 labels and this document’s reference section. ## Map Product Types ### temp Unlike the other map product types, "temp" products contain two distinct map types. TEMP maps contain lunar brightness temperatures derived by deconvolving MRM antenna temperatures. STDEV maps contain biased estimates of weighted standard deviation for the antenna temperature values used to derive the values in their corresponding TEMP maps (these values may be useful for both physical interpretation and data quality assessment). Each "temp" product contains either 19 (CE-1) or 27 (CE-2) HDUs: * PRIMARY (FITS stub) * one TEMP HDU for each 2-hour time bin with valid data (8 or 12 HDUs total) * one STDEV HDU for each 2-hour time-bin with valid data (8 or 12 HDUs total) * LATITUDE * LONGITUDE ### datminus The DATMINUS maps in these products contain model-subtracted derived brightness temperatures. Each is created by subtracting the corresponding TBMOD map of its matching “tbmod” product from the corresponding TEMP map of its matching “temp” product (after cropping that TEMP map to the latitude range of the TBMOD map). Their values can be interpreted as differences between measured temperatures at each grid cell and expected temperatures at each grid cell, where this expectation is set by a physical model that takes into account channel frequency, local time, latitude, thermal conductivity gradient, and albedo. Because this model does not account for dielectric loss, much of the regional and diurnal variability in these maps (especially at channels 1 and 2) is generally dependent on concentrations of the titanium-containing mineral ilmenite. Each "datminus" product contains either 11 (CE-1) or 15 (CE-2) HDUs: * PRIMARY (stub) * one DATMINUS HDU for each 2-hour time bin with valid data (8 or 12 HDUs total) * LATITUDE * LONGITUDE ### latshift The LATSHIFT maps in these products contain latitudinally adjusted derived brightness temperatures. Each is created by fitting a simple latitudinal model of temperature (`a + cos(lat) ** b`, with `a` and `b` as free variables) to the corresponding TEMP map of its matching “temp” product, then subtracting the fitted curve from the TEMP map. Their values are differences between measured temperatures at each grid cell and expected temperatures at each grid cell, where this expectation is set *only* by the latitudinal trend in data taken by this particular orbiter in this particular channel during this particular range of local times and incorporated into this particular TEMP map. We provided these maps in addition to the DATMINUS maps because there are many confounders in these data, including simple sample selection bias produced by offsets between different orbits, that make it difficult to model the joint dependence of temperature on LTST and latitude. By restricting the population of samples in the model, the LATSHIFT maps provide a stricter, although *strictly relative*, view of temperature variation. This means that some features may be apparent in these maps that are not apparent in the DATMINUS maps. Each "latshift" product contains either 11 (CE-1) or 15 (CE-2) HDUs: * PRIMARY * a LATSHIFT HDU for each 2-hour time bin with valid data (8 or 12 HDUs total) * LATITUDE * LONGITUDE ### tbmod The TBMOD maps in these products contain modeled brightness temperatures derived by interpolating Kaguya SELENE and LRO Diviner, LOLA and LROC data within the parameter space of a physical heatflow model. This process is described in more detail in the “Thermal Brightness Modeling” section below. Because they do not use CE MRM data as an input, there is only one “tbmod” product for each channel. Each “tbmod” product contains 15 HDUs: * PRIMARY * one TBMOD HDU for each 2-hour time bin (12 HDUs total) * LATITUDE * LONGITUDE # Source Data Products ## Overview All products in this bundle’s data collection are based on the contents of the NAOC-held datasets `CE1_MRM_2C_20071127161816_20090114192053_B` and `CE2_MRM_2C_20101015085002_20110520125150_A`. These datasets present the MRM L2C data as fixed-width ASCII tables, one table per orbit in which the MRM took observational data, each table in a separate file. These files have attached labels written in valid Parameter Value Language (PVL). Considered as data products, they are generally compliant with version 3.8 of the PDS Standards. The CE-1 dataset contains 1691 such files; the CE-2 dataset contains 2402. The Chang’e Project’s processing level definitions are similar to NASA EOSDIS definitions. Like EOSDIS L2 products — and most other L2 products across the Chang’e Project data corpus — the MRM L2C products consist of georeferenced time-series samples in physical units. We expect that most users interested in working with the MRM data at time-series level will prefer to use the concatenated tables in this bundle’s data collection. However, to help ensure the reproducibility of our work (as well as a large body of existing scientific literature), we have included PDS4-labeled versions of these source products in this bundle’s data_source collection. We have omitted only a handful of “empty” products from the CE-1 dataset that contain a PVL label but no table. As of writing, we are not aware of any other copy of these data in a PDS-equivalent archive. Because the fixed-width tables are consistently formatted and the attached PVL labels are clearly separated from the tables, these files require no physical modification to be described as valid PDS4 Product_Observational objects with associated Header and Table_Character objects. We have therefore chosen to interfere with them as little as possible. Although no checksums are available, all data files in the data_source collection are, to the best of our knowledge, byte-level equivalent copies of the L2C files held by the NAOC. We have also retained their original filenames, including the unconventional “.2C” extension. (We have, however, organized them into separate volumes by month and year to make navigation easier.) ## Source Data Product Filenames All source data filenames follow the pattern: `[orbiter]_BMYK_MRM-L_SCI_P_[start time]_[stop time]_[orbit number]_ [letter].2C` * “orbiter” may be “CE1” or “CE2”. * “start time” and “stop time” provide nominal time bounds for the product in UTC time scale, expressed as YYYYMMDDHHMMSS. For instance, “20071127182559” is equivalent to “2007-11-27T18:25:59Z”. Note that these are not fixed to the first and last timestamps in the data tables (although they are usually within a few seconds), and should likely be understood as time bounds for the mission-designated orbit. * “orbit number” is the mission-designated orbit number as a 0-padded 4-digit number. * “letter” is always “B” for CE-1 products and “A” for CE-2 products. Our PDS4 labels for these products share these filename stems (converted to lowercase in accordance with PDS4 conventions). ## Table Contents and Format CE-1 and CE-2 table formats are semantically equivalent, although column names and boundaries differ slightly. (Please refer to their PDS4 or PVL labels for details.) The columns contain: 1. Time UTC, expressed in ISO 8601 format to millisecond precision 2. Channel 1 brightness temperature in Kelvin, given to two decimal places 3. Channel 2 brightness temperature in Kelvin, given to two decimal places 4. Channel 3 brightness temperature in Kelvin, given to two decimal places 5. Channel 4 brightness temperature in Kelvin, given to two decimal places 6. Signed solar incidence angle in degrees, given to four decimal places 7. Solar azimuth angle in degrees, given to four decimal places 8. Latitude in degrees, given to four decimal places 9. Longitude in degrees, given to four decimal places 10. Orbital height in kilometers, given to six decimal places 11. Quality state, given as an encoded string (possibly a malformatted hexadecimal integer) in CE-1 tables and a 0-padded 2-digit decimal integer in CE-2 tables. The exact meanings of specific quality state values are not included in documentation, but any value of this field other than “0X000000”(for CE-1) or “00” (for CE-2) appears to indicate severely off-nominal data. There are short missing time spans in some tables and large time gaps between some tables, but most of the samples are separated by only 1.6 or 3.6 seconds (mean of ~2 seconds). Most tables have about 4000 samples / rows. # Concatenated Data Tables ## Rationale Each NAOC-released L2C file contains data from a single orbit, but most applications for the time-series MRM data require analysis of data from many orbits. This means that scientists who wish to work with these data must write their own preprocessing code to load millions of rows from thousands of separate text files and transform them into a format suitable for their specific application. Most applications also require scientists to write additional custom code to derive various geometric quantities. These coding tasks present a barrier to use and run the risk of propagating unique preprocessing errors into downstream code. We have spent a lot of effort on these chores in support of our own investigations. To remove this burden from future researchers, we have provided two preprocessed tables, one per mission, in a tabular format suitable for most anticipated uses of the time-series data. ## Concatenated Table Filenames There are exactly two products of this type: ce1_mrm.fits/xml, which contains data from the CE-1 MRM, and ce2_mrm.fits/xml, which contains data from the CE-2 MRM. ## Concatenated Table Contents Each concatenated table product consists of a PDS4-labeled FITS file. The TABLE HDU of each FITS file contains a binary table of annotated MRM data. Each row of these tables corresponds to one row from an L2C ASCII table (described in the “Source Data Products” section above). Equivalently, each row corresponds to a single MRM sample. The CE-2 table has ~8.7 million rows; the CE-1 table has ~6.3 million. They exclude only a handful of rows from the source data tables that contain special constants or out-of-bounds values (about 50 in total). Note that these tables discard the solar azimuth and incidence fields from the source tables due to the presence of some implausible values. They replace them with SPICE-computed solar position vectors and local true solar time (LTST) at the boresight intercept point. They also do not directly copy the quality flag columns, as there is no public documentation on the meaning of the different flags; they instead propagate the presence of any nonzero quality flag value as an element of the FLAG column bitmask. In addition to LTST and solar position vectors, the tables contain several other quantities that are not directly copied from the source files: surface normal vectors at boresight intercept point, time expressed as seconds since J2000 (for compatibility with SPICE), and a quality flag bitmask. A complete list of columns follows. Note that all integer and float values are stored in MSB order. ### Columns 1. ORBIT (2-byte unsigned integer, unitless): Mission-specified orbit number, taken from names of source files. 2. UTC (23-byte UTF-8 string, unitless) ISO 8601-formatted time string, taken from source tables. 3. ET (8-byte float, seconds): Ephemeris Time (offset in seconds from J2000, commonly used in the SPICE ecosystem). Derived from UTC and NAIF ephemerides. 4. LTST (4-byte float, unitless): Local true solar time, normalized to 0-1, derived from ET, LON, and NAIF ephemerides. 5. T1-T4: (4-byte float, K, 4 columns total): antenna temperature at channels 1-4, taken from source tables. 6. LAT (4-byte float, degrees): Selenodetic latitude of boresight position in MOON_ME, taken from source tables. 7. LON (4-byte float, degrees): Selenodetic longitude of boresight position in MOON_ME, taken from source tables and converted from 0-360 to -180-180 representation. 8. D (4-byte float, km): Distance from orbiter to boresight intercept point, taken from source tables. 9. BX, BY, BZ (8-byte float, km, 3 columns total): X, Y, and Z components of boresight intercept point relative to Sun body center in ecliptic coordinates (ECLIPJ2000), derived from LAT, LON, ET, and NAIF ephemerides. 10. CX, CY, CZ (8-byte float, km, 3 columns total): like BX/BY/BZ, but for orbiter position rather than boresight intercept point. 11. MX, MY, MZ (8-byte float, km, 3 columns total): like BX/BY/BZ, but for Moon body center rather than boresight intercept point. 12. NX, NY, NZ (4-byte float, unitless, 3 columns total): X, Y, and Z components of unit normal vector at boresight intercept point in body-fixed (MOON_ME) coordinates. 13. FLAG (2-byte unsigned integer, unitless): Quality flags expressed as a bitmask. These are our own flags and are not directly propagated from the source data. For most analytic purposes, we recommend discarding all samples with a nonzero value in this field, with the possible exception of 0b1000000. Individual (summable) values denote: * 0b1: Sample flagged as bad by data providers. Note that this value is not used in the CE-1 table. This is because there are very few flagged rows in the CE-1 L2C source files (~30, as opposed to ~1.5 million in CE-2), and all of these rows contain invalid special constant values, so we did not include them in the concatenated table. * 0b10: Physically implausible temperatures (< 34K) at one or more channels. * 0b100: Physically implausible offset (> 75 K) between channel temperatures. * 0b1000: Sample is from an orbit we have manually flagged due to large ranges of implausible temperature or geopositioning data. * 0b10000: Sample is from an orbit almost completely flagged by the original data providers or with very few provided samples (the surviving samples in these orbits are uniformly suspicious). * 0b100000: Sample's timestamp duplicates the timestamp of another sample in the dataset (the data may or may not be duplicated, but it is unclear what the orbiter was actually doing at that point). * 0b1000000: Sample is from an orbit range we have flagged on suspicion of secular offsets in calibration that make its temperature values poorly comparable with other portions of the data set. Not used in CE-1 table. See the “Calibration” section above for details on the CE-2 intermission calibration offsets that led us to apply this flag. * 0b10000000: Sample is from an orbit with implausibly low or high latitudinal variation in channel 1 temperature (after excluding data with prior flags). * 0b100000000: Sample is from an orbit with suspiciously large numbers of individual outliers in channel 1 temperature across a variety of latitudinal ranges (after excluding data with prior flags). ## PRIMARY HDUs The concatenated table FITS files each begin with a “stub” PRIMARY HDU for compliance with the FITS standard. These HDUs contain no data and no metadata relevant to interpretation. they are present only to make the files valid FITS. # Deconvolved Map Product Processing ## Rationale There is no single, widely-used map-projected version of the MRM data. The NAOC holds a set of equirectangular maps of the CE-1 data at 1 pixel / degree. (The dataset title is CE1-MRM-SDP-V1.0. See Zheng et al., 2012 for a discussion of these products (the dataset itself is cited as GSDSA 2008 in this document’s bibliography). However, no similar maps are available for CE-2, and, in practice, most scientific investigations that approach the MRM data in map-projected form begin by creating ad hoc “basemaps” directly from the L2C data. These are generally equirectangular projections produced by creating a lat/lon grid, selecting a temperature channel, and defining the brightness temperature of each grid cell as the mean of the temperature values at that channel for all samples whose nominal boresight intercept coordinates fall within that grid cell (potentially after selecting a subset of and/or applying an inline correction to the samples). ## On Binning and Averaging Although this "binning and averaging" (BAA) technique is very common across remote sensing disciplines, it can introduce unpredictable distortions, and there are several reasons that BAA is a questionable mapmaking technique for the CE MRM data in particular. (These observations are also applicable, at least in part, to any other spatially-referenced time-series data in which individual samples have spatial extent on the scale of the spatial frequency of the data, the spatial gradient of the measured quantity, or the size of the grid cells.) BAA makes the following implicit assumptions: 1. A sample provides information about the grid cell in which its georeference falls, and no other cells. Equivalently, no sample whose georeference falls outside a grid cell provides information about that cell. 2. Every sample whose georeference falls within a particular grid cell provides the same amount of information about that grid cell. These are not good assumptions for the CE MRM unless you are making *extremely* coarse maps. This is because the MRM antenna patterns are large enough that radiative emissions from a large swath of the lunar surface contribute to each antenna temperature measurement. This means that every sample provides information about a large swath of the lunar surface. Because lunar brightness temperatures can vary significantly on this spatial scale, it also means that looking at a single antenna temperature measurement cannot tell you whether the MRM “saw” a wide or narrow range of temperatures within that swath. Applying BAA to these data discards much of the evidence they provide and overweights the rest. Samples whose footprints cover a grid cell but whose georeferences do not fall within that cell do not contribute to the derived value for that cell. This sharply reduces the *apparent* resolution and coverage of the data: a particular map may have many “holes” in extremely well-measured areas simply because no sample’s georeference fell within those cells. This means that insignificant differences in orbital swath patterns–down to tens of meters in some cases–can make dozens of kilometers of the surface appear unmeasured. It also reduces the *effective* resolution of the data, because the presence of a feature within a particular grid cell may be apparent only from its aggregate contributions to multiple samples whose georeferences fall outside the cell (which BAA specifically cannot relate to that particular cell). The fact that BAA discards the contributions of samples to the derived values of adjacent cells also means that it overweights the contributions of samples to their assigned cells. This tends to amplify noise, because it includes outlier sample values in only one cell instead of distributing them across multiple cells. It can also arbitrarily shift features into adjacent cells–or even fragment them into multiple adjacent cells in different directions–when they were measured principally “from” those cells. These effects are exacerbated by the fact that they vary unpredictably, in both strength and location, depending on the chosen grid. They tend to be worse at higher grid resolutions, but because they depend on essentially arbitrary assignment of samples to cells, even a very small change in resolution can create artifacts in different places, and some lower resolutions might have worse distortion than some higher resolutions. In equirectangular projection, these effects become worse at higher latitudes, because each grid cell represents a smaller total area, meaning that BAA discards more information from each sample. And finally, they have complex mission-phase-dependent variability due to the fact that the size of the MRM footprint varies with the orbiter’s distance from the surface. Although we have used BAA previously, we believe it is an inferior approach. It is something like making a photomosaic by taking a series of orbital images, arbitrarily cropping some of them, and then blurring remaining overlapping regions together rather than attempting to coregister them. The CE MRM data have more “real” resolution than is apparent under BAA. Data anomalies make it uncertain, but based on coregistration with known lunar features, we believe meaningful resolution down to 32 pixels/degree is available in densely-sampled regions of the surface, especially in CE-2 channels 2, 3, and 4. (By "meaningful resolution", we refer to feature centroids and gradients; many features are still "blurred" relative to data from instruments with higher per-sample resolution.) Also, although this resolution is not consistently present across the MRM data sets, the method we present here is, at worst, a strictly-correct upsampling technique (if performed to adequate orbiter-referenced angular resolution). It introduces fewer distortions into comparisons between the MRM data and high-resolution data from other instruments than generic resampling methods. ## Deconvolution Method Consider each antenna temperature value as a surface integral over the product of two functions: one function represents the brightness temperature of the Moon, and the other represents the MRM antenna pattern with its center shifted to the boresight intercept point. In other words, each antenna temperature value is a convolution of lunar brightness temperature and antenna response. We know the antenna response curve, so we can reconstruct the brightness temperature of the Moon by deconvolving georeferenced antenna response from antenna temperature. Because the brightness temperature function cannot be written in closed form, there is no way to solve this with clever symbolic manipulation. However, it *is* possible to back out brightness temperature by numerically approximating the integrals. To do so, we perform the following steps (ignoring many implementation details): 1. Define an equirectangular latitude / longitude grid on the lunar surface. 2. Define three arrays referenced to this grid: W (weight), WT (weighted temperature) and WS (weighted squared temperature). Set all elements of these arrays to 0. 3. Define the antenna pickup pattern in instrument-relative elevation / azimuth coordinates, with high resolution relative to orbital height and latitude / longitude resolution. 4. Then, at each sample: 1. using boresight intercept coordinates and orbital height as input variables, apply a reverse orthographic projection to the antenna pickup pattern grid to transform it into latitude/longitude coordinates. 2. Compute a binned sum of the projected antenna pattern elements on the latitude / longitude grid. Call that array A. Call the temperature value of the sample V. Then: * W = W + A * WT = WT + A * V * WS = WS + A * V^2 5. When you are done with all the samples, define the output quantities like this: * WT / W is the derived brightness temperature at each grid cell.(These are the TEMP HDUs in our “temp” deconvolved map products.) * sqrt(WS * W - WT^2) is the standard deviation of measured temperature at each grid cell. (These are the STDEV HDUs in our “temp” deconvolved map products.) ### Notes * We chose an equirectangular map projection because we wanted to make comparisons to existing equirectangular maps based on data from other instruments, but this basic method is independent of the choice of map projection. To modify the algorithm to work with other map projections, you could either change the coordinate system transformation function in step 2, add an additional step between 2 and 3 to convert lat/lon into the coordinates of your map projection, or simply deal with the inconveniences of counting on an irregular grid in step 3. * We considered only the antenna main beams, not their sidelobes. This is because the MRM sidelobes are highly asymmetrical and no spacecraft attitude data are available, so we believed attempting to model the sidelobes would simply cause spatial ‘smearing’ of unknown and variable direction and magnitude. * The method we describe here approximates a deconvolution as a massive weighted average. As such, it is perfectly reasonable to think of BAA as a special case of this method that treats each measured value as an integral over the product of the measured quantity and a Dirac delta function shifted to the center of the nearest grid cell. This allows you to skip the complicated parts by setting almost everything to 0 or 1. * This is formally similar to "super-res" techniques in visible-band imaging. In this case, however, we do not consider it an "image enhancement" technique, but rather a literally correct--even brute-force--approach to the problem. Each measurement always already provides information about an extended area. * We are not aware of prior work with the CE MRM data that takes this approach. We suspect that other investigators may have considered it but found it computationally intractable. Although the procedure is conceptually simple, the MRM data sets include millions of data points, and approximating the surface integrals to useful precision without introducing discontinuities requires performing expensive operations on multiple arrays with millions of elements at each of those millions of points. We discarded our first, straightforward implementation because processing all of the data with it would have broken our timeline or our budget or both. To create a practical implementation of this procedure, we had to perform intensive, multifaceted software development efforts, ranging from finicky optimization of application-specific calculations to developing a general-purpose dynamic numerical approximation library (St. Clair, 2024). For full implementation details, please refer to the deconvolution source code (St. Clair et al., 2024). ### Related Work There are at least two distinct existing approaches to dealing with the wide footprint of the MRM data. One is a deconvolution approach of a very different character that applies a maximum entropy-based image reconstruction technique to some unspecified “original map” of the CE-2 data (Xing et al., 2015). We suspect our approach is more valid; however, as the authors did not fully specify their inputs, and made neither their code nor their data products available, we have not been able to meaningfully assess their work. Another approach, which is applicable only to analyses that fuse the MRM data with other observational or model data, borrows a common technique from imaging camera data processing and convolves the comparison data with MRM antenna patterns. However, time-series radiometer data is very little like imaging camera data. Because the “image” is built up from many point samples, convolving comparison data with the radiometer antenna pattern makes comparison of feature extents more realistic, but leaves in all the noise and loss effects associated with BAA. We have used this technique in some previous projects, but now consider it inferior to “correct” deconvolution. # Thermal Brightness Modeling This section provides a basic description of how we produced the modeled thermal brightness temperatures used in our “tbmod” and “datminus” maps. It is a variation of the technique described in Siegler et al., 2023. We first create a forward model of physical temperature using an iterative, self-equilibrated heat flow model (the physical assumptions are described in Hayne and Aharonson, 2015; Hayne et al., 2017 describes the immediate precursor to our software implementation of the model). This model takes latitude, albedo, and h-parameter (a metric of density / thermal conductivity change with respect to depth) as inputs and returns physical temperature by “depth” (really a tuple of strictly covariant depth-related quantities) and local time. To do so, we define sets of latitude, albedo, and h-parameter values that span the range of our external inputs and desired map bounds, then execute one instance of this model for each element of their Cartesian product (~7500 total for the products in this bundle). The outputs of this set of model executions form a discrete, bounded 5-dimensional parameter space in which each element has an associated scalar representing expected physical temperature. We then map a simple brightness temperature model across the latitude, albedo, local time, and h-parameter axes of this space. This model takes temperature gradient with depth, material :dielectric parameters, and emission frequency as inputs and returns brightness temperature. We execute this model four times (once per MRM channel) for each element of these axes—the model essentially “collapses” all elements along the depth axis at each of these points into a scalar brightness temperature. Note that analytic difficulties precluded us from creating dielectric property estimates that were compatible with the deconvolution process, so we treated dielectric properties as constant, using mean values for the Moon as a whole. (Note that this differs from the analyses in Siegler et al. 2020 and Siegler et al. 2023, which included fit values of the localized loss tangent to remove the effect of varying composition. It is also why titanium content pops out so clearly in some of our products.) Finally, we ingest derived maps of h-parameter, albedo, slope, and azimuth based on observational data from Kaguya SELENE and LRO Diviner, LOLA, and LROC then coregister them to the intended scale and lat/lon bounds of our output maps. For each element of our array of local time bins, we adjust effective local time using the slope/azimuth data (this essentially treats hour angle as “real” local time); then, for each of the MRM channel center frequencies, we produce derived temperature values by interpolating latitude, h-parameter, albedo, local time, and frequency within the parameter space produced by the brightness temperature model. These are the TBMOD maps in the “tbmod” products. Our “datminus” maps are produced by subtracting the values in these maps from the TEMP maps in the “temp” products. (The slope and azimuth maps are available from the PDS LOLA node and are referenced in the "tbmod" product labels. The albedo and h-parameter maps are, respectively, wac_hapke_604nm_70N70S_64ppd.fits and hpar_global_128ppd.fits, and are both included in this bundle's miscellaneous collection.) # Bibliography China National Space Administration, 2009. China’s lunar probe Chang’e-1 impacts moon [WWW Document]. Feng et al., 2013. Data processing and result analysis of CE-2 MRM. Earth SciJ China Univ Geosci 38, 898–906. Feng, J., Siegler, M.A., Hayne, P.O., 2020. New Constraints on Thermal and Dielectric Properties of Lunar Regolith from LRO Diviner and CE‐2 Microwave Radiometer. J. Geophys. Res. Planets 125, e2019JE006130. https://doi.org/10.1029/2019JE006130 GSDSA, 2008. CE1-MRM-SDP-V1.0. https://doi.org/10.12350/CLPDS.GRAS.CE1.MRM-03.vA Hayne, P.O., Aharonson, O., 2015. Thermal stability of ice on Ceres with rough topography. J. Geophys. Res. Planets 120, 1567–1584. https://doi.org/10.1002/2015JE004887 Hayne, P.O., Bandfield, J.L., Siegler, M.A., Vasavada, A.R., Ghent, R.R., Williams, J., Greenhagen, B.T., Aharonson, O., Elder, C.M., Lucey, P.G., Paige, D.A., 2017. Global Regolith Thermophysical Properties of the Moon From the Diviner Lunar Radiometer Experiment. J. Geophys. Res. Planets 122, 2371–2400. https://doi.org/10.1002/2017JE005387 Hu, G., Keihm, S.J., Wang, Z., 2022. An In-Flight Recalibration for Chang’E-1 and E-2 Microwave Radiometer Datasets Based on Highland Thermophysical Models. IEEE Trans. Geosci. Remote Sens. 60, 1–21. https://doi.org/10.1109/TGRS.2021.3125714 Hu, G.-P., Chan, K.L., Zheng, Y.-C., Tsang, K.T., Xu, A.-A., 2017. Comparison and evaluation of the Chang’E microwave radiometer data based on theoretical computation of brightness temperatures at the Apollo 15 and 17 sites. Icarus 294, 72–80. https://doi.org/10.1016/j.icarus.2017.04.009 Hu, G.P., Keihm, S.J., 2021. Effect of the Lunar Radiation on the Cold Sky Horn Antennas of the Chang’E-1 and -2 Microwave Radiometers. IEEE Geosci. Remote Sens. Lett. 18, 1781–1785. https://doi.org/10.1109/LGRS.2020.3007606 Huixian, S., Shuwu, D., Jianfeng, Y., Ji, W., Jingshan, J., 2005. Scientific objectives and payloads of Chang’E-1 lunar satellite. J. Earth Syst. Sci. 114, 789–794. https://doi.org/10.1007/BF02715964 Jones, A., 2021. China to launch a pair of spacecraft towards the edge of the solar system. SpaceNews. Li, C. & Li, H, 2013. Chang’e 2 Flyby of Toutatis (presentation). SBAG 8 (2013). https://www.lpi.usra.edu/sbag/meetings/jan2013/presentations/sbag8_presentations/TUES_0930_CE_Toutatis.pdf NSSDCA, 2022. Chang’e 2 [WWW Document]. NASA Space Sci. Data Coord. Arch. Siegler, M.A., Feng, J., 2017. Microwave Remote Sensing of Lunar Subsurface Temperatures: Reconciling Chang’e MRM and LRO Diviner 1705. Siegler, M.A., Feng, J., Lehman-Franco, K., Andrews-Hanna, J.C., Economos, R.C., Clair, M.St., Million, C., Head, J.W., Glotch, T.D., White, M.N., 2023. Remote detection of a lunar granitic batholith at Compton–Belkovich. Nature 620, 116–121. https://doi.org/10.1038/s41586-023-06183-5 Siegler, M.A., Feng, J., Lucey, P.G., Ghent, R.R., Hayne, P.O., White, M.N., 2020. Lunar Titanium and Frequency‐Dependent Microwave Loss Tangent as Constrained by the Chang’E‐2 MRM and LRO Diviner Lunar Radiometers. J. Geophys. Res. Planets 125, e2020JE006405. https://doi.org/10.1029/2020JE006405 St. Clair, M., 2024. quickseries. https://doi.org/10.5281/ZENODO.12702826 St. Clair, M., Brown, S., Million, C., Feng, J., Siegler, M., 2024. ce-mrm-processing. https://doi.org/10.5281/ZENODO.12709958 St. Clair, M., Million, C.C., Ianno, A., Feng, J., Siegler, M., 2022. Approaches to Production of Intermediate Data Products for Characterizing Systemic Anomalies in the Chang’e-2 Microwave Radiometer Data, in: 53rd Lunar and Planetary Science Conference. Presented at the Lunar and Planetary Science Conference. Tsang, K.T., Fu, S., Gu, J., Zhou, M., Chan, K.L., Zheng, Y.C., 2014. A statistical learning approach to Chang’E microwave radiometer data calibration, in: 2014 11th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD). Presented at the 2014 11th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD), pp. 374–378. https://doi.org/10.1109/FSKD.2014.6980863 Wang, M., Shi, X., Jian, N., Yan, R., Ping, J., 2009. Real time monitoring of the Chang’E-1 lunar orbit insertion, in: 2009 15th Asia-Pacific Conference on Communications. Presented at the 2009 15th Asia-Pacific Conference on Communications, pp. 442–445. https://doi.org/10.1109/APCC.2009.5375596 Wang, Z., Li, Y., Zhang, D., Jingshan, J., 2010a. Prelaunch Calibration of Chang’E-2 Lunar Microwave Radiometer, in: 2010 International Conference on Microwave and Millimeter Wave Technology. pp. 1551–1554. Wang, Z., Li, Y., Zhang, X., JingShan, J., Xu, C., Zhang, D., Zhang, W., 2010b. Calibration and brightness temperature algorithm of CE-1 Lunar Microwave Sounder (CELMS). Sci. China Earth Sci. 53, 1392–1406. https://doi.org/10.1007/s11430-010-4008-x Xing et al., 2015. The deconvolution of lunar brightness temperature based on the maximum entropy method using Chang’e-2 microwave data. Res. Astron. Astrophys. 15., pp. 293-304. Zheng, Y.C., Tsang, K.T., Chan, K.L., Zou, Y.L., Zhang, F., Ouyang, Z.Y., 2012. First microwave map of the Moon with Chang’E-1 data: The role of local time in global imaging. Icarus 219, 194–210. https://doi.org/10.1016/j.icarus.2012.02.017 Zuo, W., Li, C., Zhang, Z., 2014. Scientific data and their release of Chang’E-1 and Chang’E-2. Chin. J. Geochem. 33, 24–44. https://doi.org/10.1007/s11631-014-0657-3 # Footnotes [^1]: "Pursuant to The Department of Defense and Full-Year Appropriation Act, Public Law 112-10, Section 1340(a); The Consolidated and Further Continuing Appropriation Act of 2012, Public Law 112-55, Section 539; and future-year appropriations (hereinafter, "the Acts"), NASA is restricted from using funds appropriated in the Acts to enter into or fund any grant or cooperative agreement of any kind to participate, collaborate, or coordinate bilaterally with China or any Chinese-owned company, at the prime recipient level or at any subrecipient level, whether the bilateral involvement is funded or performed under a no-exchange of funds arrangement." [^2]: Most CE-2 instruments were in fact flight spares from CE-1, including the MRM. Therefore, unless otherwise specified, all technical specifications given in this document for the MRM apply equally to both CE-1 and CE-2. [^3]: The MRM’s observational cadence is clearly visible in some of the deconvolved maps, particularly in the CE-2 data at lower latitudes. Because the spatial offset “jumps” every sixth sample, the measured temperature gradient also “jumps” every sixth sample, creating evenly-spaced “blocks” along each orbital track.