Interactive comment on “ Ship-and island-based soundings from the 2016 El Niño Rapid Response ( ENRR ) field campaign ”

The overall quality of this paper is high and presents a unique and important data resource. I would recommend publications with a just a few minor revisions – the majority are covered in specific comments. The authors discuss in detail the needs for sounding reprocessing in light of the errors that can be introduced by incorrect ground observations. The paper would benefit if the authors quantified the uncertainties in variables at the varying processing levels.


Introduction
In early 2015, signs of a long-anticipated El Niño event began to appear in the Pacific Ocean. Both dynamical and statistical models were forecasting the continuation and likely strengthening of El Niño conditions in June 2015 (Climate Prediction 25 Center (CPC) and International Research Institute for Climate and Society (IRI), 2015a). By August 2015, CPC and IRI (2015b) were describing the event as "significant" with high confidence that a strong event would peak in the late fall or early winter. The National Oceanic and Atmospheric Administration (NOAA), aware of the potential for large impacts to weather in the United States, wanted to respond quickly to the emerging forecasting challenges. Thus, the El Niño Rapid Response (ENRR) program was instituted as a joint effort by both operational and research-oriented sectors within NOAA. Coordinated, 30 targeted field work was proposed as one element of ENRR by the Earth System Research Laboratory's (ESRL) Physical Sciences Division (PSD; Dole et al., 2017). The resulting ENRR Field Campaign, led by PSD but involving many partners within and outside of NOAA, had several observational and research components, including hundreds of atmospheric profiles of temperature, humidity and winds measured by radiosondes launched from Kiritimati (pronounced "Christmas") Island and the NOAA Ship Ronald H. Brown. This manuscript documents the collection of these soundings, their reprocessing with improved surface data, the initial quality control applied, and the resulting data sets.
2 Field sites and data collection 5

Kiritimati Island, Republic of Kiribati 1
The Line Islands lie south of Hawaii, extending from 6.4°N to 11.4°S (Fig. 1). Kiritimati, whose highest point is 13 m ASL and whose land area is 321 km 2 , is the largest of these islands (Scott, 1993). About 6,500 residents live in four villages on the northern and western portions of the island (Morate, 2016), while the southern portion is now designated as a bird sanctuary.
Physical contact with the outside world is limited; during most of 2016 there were 2 international passenger flights each week, 10 occurring during one round trip between Fiji and Hawaii which stops at Kiritimati in both directions. Most supplies are imported on monthly freight flights from Hawaii and during irregular visits by freighter ships. There is internet access on the island but it is slow, perhaps 0.26 Mbps downloading and 0.03 Mbps uploading on a good day. Some of the best internet connectivity on the island is at the Captain Cook Hotel. 15 The Captain Cook Hotel, located in the Main Camp neighbourhood of Banana village, was the center of our ENRR operations.
The hotel property is located on the northeast-facing coast (Fig. 2) and has hosted NOAA scientists in the past. The grounds are sandy and moderately vegetated, predominantly with short grasses and similar plants but also with some shrubs and coconut palms. The short beach runs steeply down to shallow water that extends for tens of meters to the edge of the reef; beyond that, the reef drops off sharply and the water is hundreds of meters deep. For nine weeks PSD maintained a staff of two observers 20 who occupied both sides of a two-unit bungalow. One of the bungalow's units also housed the computers and radiosonde equipment. Helium tanks were stored on the south corner of the large (13 m by 31 m) concrete pad outside the bungalow. The concrete pad was very slightly higher than the ground on which our bungalow and the surface station sat, but overall the area where observations were collected was flat and fairly level.

25
Site set-up began on 25 January 2016. A tripod was positioned north of the bungalow and east of the concrete pad, and outfitted with an R.M. Young 5103 anemometer, a tipping bucket rain gauge, a shielded Vaisala HMP45C temperature and relative humidity probe, and a box containing a Vaisala PT101B barometer and a Campbell Scientific CR23X datalogger. Above the box was a solar panel which charged the battery that provided power to the datalogger. This tripod was moved northeast about 10 m, to a location just above the beach, on 5 February 2016. More details of the surface instrumentation, as well as an explanation of the post-deployment processing of the collected surface observations, are found in Hartten et al. (2017a).
The first Vaisala RS92-SGP radiosonde was launched at 1136 UTC on 26 January 2016. Typically, one observer was primarily responsible for filling the balloon. A flattened tarpaulin tube, held together by a strong Velcro pull-strip running its entire 5 length, was laid on the somewhat sheltered concrete porch outside the bungalow and the 350-gram balloon was laid inside it.
The balloon was connected to an analog regulator attached to one of the gas cans and the balloon was filled with approximately 600 PSI of helium, or until the balloon was about the same length as the tube. (We were unable to use lift weights to determine when the balloons were filled because we lacked anything resembling an inflation shed to keep the filling balloon protected from the wind and weather during the process.) The other observer was generally inside the bungalow, using the Vaisala 10 GC25 ground check set (Vaisala Oyj, 2006) and the DigiCORA sounding software (Vaisala Oyj, 2010b;version 3.64.1) to calibrate and initialize the radiosonde and set the sounding parameters. The DigiCORA software was directed to run in "research mode" so that it would continue to collect data during the downward trajectory of the radiosonde after the balloon burst. After the ground check was complete and the battery pack was connected, the radiosonde was usually carried outside and set on the surface station's instrument box, in the shade of the station's solar panel. This provided a well-ventilated 15 location for the radiosonde to equilibrate to outdoor conditions and acquire GPS lock. It also allowed for several minutes of independent pressure, temperature and humidity measurements co-located with the surface meteorological station. When it was time to launch, surface observations from the meteorological station were entered into the DigiCORA software; the radiosonde was attached to the balloon and then released from the concrete pad or the grassy space south of it and west of the bungalow. The coordinates (2.01°N, 157.40°W) adequately describe the location of both the surface station and the area from 20 which the radiosonde was launched.
Our goal was to collect 00 and 12 UTC soundings, so in accordance with WMO guidelines (World Meteorological Organization, 2013; see Part III Section 2.4.10) launches typically occurred 30-45 minutes before the nominal sounding time (i.e. 2315(i.e. -2330(i.e. and 1115(i.e. -1130. We monitored the launch from inside the bungalow, and relaunched if the radiosonde 25 did not reach 400 hPa (Office of the Federal Coordinator for Meteorology (OFCM), 2006). This only happened once, with the 7 February 2016 12 UTC sounding. The system had occasional trouble locking on to enough GPS satellites to get position readings, but typically only for short segments of a flight. There were also signal dropouts when the radiosonde was in the null directly over the site's telemetry antenna. After the balloon burst, we continued to monitor the radiosonde until we ceased receiving data. When the data collection finished, we used the DigiCORA software's simulation feature to run the radiosonde 30 transmissions through the software's "non-research" (operational) mode and created TEMP, PILOT, and BUFR messages from the operational-mode sounding data. After we ran a script to apply the proper WMO wrappers to the TEMP messages, the WMO, PILOT, BUFR, and research-mode text files were then compressed into a single file for transmission to Boulder , Colorado, via the internet access point at the main hotel building. Compression into one file was necessary because the internet was slow and tended to drop the connection after each transfer; even so, we were not always able to get the file to Boulder in the desired time window. Computer scripts in Boulder uncompressed the file and pushed the TEMP message to NOAA's National Center for Environmental Prediction (NCEP) Central Operations for transmission to the Global Telecommunication System (GTS). 5 While "CXI" is a common abbreviation for Kiritimati and is the International Air Transport Association (IATA) airport code for the local airport, we chose to use "CXENRR" as the station name to clearly distinguish these data from a variety of past meteorological data collections. The ICAO code assigned to Kiritimati is "PLCH", which is used for METAR and other data collected by the Kiribati Weather Service on Kiritimati. 10 There were two exceptions to the standard launch times. We did an extra launch on 2 March 2016 at 2017 UTC. This involved coordinated transmissions with several amateur radio antennas that were to operate at 3.505 -28.010 MHz near our bungalow for two weeks, in order to verify that there was no interference with our sounding data collection. The 00 UTC sounding on at the hotel. Over the course of our deployment, we successfully collected 125 soundings.

NOAA Ship Ronald H. Brown
The NOAA Ship Ronald H. Brown, part of NOAA's Marine Operations Center -Atlantic (MOC-A) fleet, is a Global Class research vessel that has been in service since 1997. It was scheduled to perform routine servicing of the Tropical Atmosphere Ocean (TAO) array buoys along 140 °W and 125 °W in February and March 2016. These buoys' position within the area of 20 warm SST anomalies associated with the ongoing El Niño motivated a collaborative effort in which the ENRR Field Campaign was able to deploy three scientists and radiosonde equipment on the ship during the cruise. The objective of the ship-based ENRR sounding campaign was to collect atmospheric profiles of temperature, humidity and winds six times per day (at 00, 04, 08, 12, 16, and 20 UTC) at the beginning of the campaign, while the ship travelled from Hawaii southeast to the first scheduled TAO buoy position at (9° N, 140° W); eight times per day (at 00, 03, 06, 09, 12, 15, 18, and 21 UTC) while the ship 25 was near the anomalously warm water between 8° N and 8°S conducting the TAO maintenance (from 140° W to 125° W); and six times per day again during the final leg north to San Diego from (8° N, 125° W). With a few exceptions, we carried out this schedule, collecting 193 profiles from 16 February to 16 March (Fig. 3). This total includes 135 soundings between 8° N and 8° S, as well as three launches in mid-February that were coordinated with dropsondes deployed nearby by the NOAA Gulfstream-IV (G-IV; see Table 1). The G-IV conducted numerous research flights over the tropical Pacific as part of the 30 ENRR field campaign (Dole et al., 2017). During these three launches the shipboard ENRR staff were in close contact with scientists from the G-IV mission and adjusted their launch times as best they could in order to get nearly coincident launches from ship and plane.

General launch procedures
Launch preparation took place in two adjoining locations on the main deck: in the staging bay just forward of the fantail, and in an air-conditioned lab to port of that bay (Fig. 4). One observer ran the Vaisala RS92-SGP radiosondes through the standard 5 Vaisala ground check and initialization procedure in the small lab. After this, the radiosonde was placed outside in a safe place on the fantail to equilibrate to ambient conditions until just prior to launch. The radiosonde was always put on something that was elevated approximately 1 m above the fantail's deck, and in a location that was as shaded as possible. The equilibration time was several minutes and the radiosonde was determined to be equilibrated when the real-time data feed showed that both the temperature and humidity had stabilized to outside conditions. 10 While one observer prepared the radiosonde, another was in the staging bay filling the 350-gram latex balloon 2 with helium.
The staging bay's roll-down doors face aft, have a 5.5 m vertical clearance, and were open, as was a narrower tall doorway on the starboard side of the staging bay. The bay thus provided incomplete shelter from the wind, with the resulting localized turbulence preventing a direct and consistent measure of lift using weights. Instead, the amount of helium required to achieve 15 the optimum 5 m s -1 ascent rate was determined by trial and error based on ascent rate and helium use during previous launches, and in consultation with the ENRR observers on Kiritimati. Finally, the equilibrated radiosonde was brought to the staging bay and attached to the filled balloon; surface observations were entered into the DigiCORA software; there was a final check that the radiosonde's temperature, relative humidity, and pressure were working and that the GPS receiver was in contact with at least four GPS satellites; and then the pair was brought out onto the fantail for launch. 20 The fantail of the ship typically experiences a substantial amount of turbulence caused by high winds (gusting to 15-20 m s -1 free stream) passing through the ship's superstructure. Since the main purpose of this cruise was to exchange TAO buoys in order to restore data, the fantail area was also generally crowded with recovered and not-yet-deployed buoys. Both factors made launches challenging. During the day, the balloon launches had to be coordinated with deck operations on the fantail 25 involving the buoy maintenance. In several cases, this coordination resulted in late launches. At night, the ship was typically underway, which often increased the relative wind speed and local turbulence compared to daytime conditions. Ten to fifteen minutes prior to every launch the bridge was notified of our intent. If the ship was underway, we first checked the relative wind speed and direction to determine if we would ask the bridge to adjust the ship's course and speed to optimize a fantail launch. (Due to the ship's tight schedule, slowing the ship was not always possible.) The actual launch was from whatever position on the fantail that the operators estimated would maximize the likelihood of success.
As at Kiritimati, the DigiCORA software was run in "research mode" and continued to collect data after the balloon burst, stopping only when transmission ceased because of splashdown or loss of signal. The sounding was then simulated in "non-5 research" (operational) mode. BUFR and TEMP messages were created from the operational mode data using the ship's assigned call sign, "WTEC", and appropriate WMO headers were applied to the TEMP messages. Then both the "research mode" and "non-research mode" sets of text sounding files as well as the BUFR and TEMP messages were uploaded to Boulder, from whence the TEMP messages were pushed to NOAA/NCEP's Central Operations for transmission to the GTS.

10
Conditions on the fantail resulted in a number of failed launches, for example when swirling winds drove the radiosonde into the superstructure or the ocean. In some cases, we successfully attempted a second launch; in other cases, the sounding was skipped. We had limited supplies of radiosondes, balloons, and helium, so many factors were taken into account in the decision to relaunch: the likelihood of a repeat failure, whether there was sufficient time remaining to collect and send the data before the next launch, and how critical data from this time period was. Of the 206 planned soundings, 13 scheduled launch times 15 (6%) were missed due to failures of the initial launch.

Special conditions and atypical soundings
Solar heating of the fantail deck often increased the air temperature where the radiosonde was equilibrated before daytime launches. We have compared the temperature from the corrected surface meteorology data set  to that recorded by the radiosondes on the fantail deck after equilibration but prior to launch. The surface meteorology was collected independently by the ship's standard complement of instruments; instrument details and post-processing are documented in 30 Hartten et al. (2017a). The differences between fantail conditions as measured by the radiosondes during their equilibration time and the contemporaneous post-processed surface data were +0.59° ± 0.6º C during the day (i.e. when solar zenith angle was less than 90º) compared to +0.04° ± 0.2º C when the sun was below the horizon. There is a corresponding difference in relative humidity (-3.61% ± 3.74% during the day compared to -0.01% ± 1.43% at night). We have made no attempt to change any sounding data based on this information, but are documenting it here for interested researchers.
During the cruise, the DigiCORA sounding system experienced problems tracking GPS satellites. Even though it had contact 5 at launch with at least the minimum four satellites required to calculate winds, the station would sometimes lose contact with all satellites after launch. After consulting with Vaisala we determined there was failing hardware in the GPS receiver card within the DigiCORA sounding processing subsystem (SPS220; Vaisala Oyj, 2010a, b). As a result, we collected no wind data in 26 of our soundings; 24 of the affected soundings (Table 2) were launched on 23-27 February 2016. A variety of other issues affected about 10% of the launches (Table 3). Ground checks were occasionally skipped during relaunches because of 10 timing issues; if the sonde didn't go up immediately, by the time the data were collected and formatted the window for uploading the data would be past. Lack of a ground check could affect the temperature and humidity measurements in the final sounding, but any pressure affects should be mitigated by the reprocessing described in Section 3.1. Sometimes visual observation indicated that the spool did not unwind properly or fully after launch. This could lead to unrepresentative measurements caused by the instruments "being too close to the radiation environment of the balloon or … encountering the 15 balloon's wake as it ascends" (OFCM, 2006), or to erroneous winds because the instruments were swinging on a shorter "pendulum" than expected by the software. On a few occasions the radiosonde passed through the ship's exhaust; this is discussed further in Section 3.1. Finally, some balloons were influenced by the flow through the moving ship and its cargo in ways that made the balloon drop perilously close to the water before moving away from the ship and then gaining altitude.
We have made no attempt to systematically identify these cases, but users of the data may encounter signs of this in the initial 20 seconds of some soundings.

Reprocessing with corrected surface values
The atmospheric soundings collected by these radiosondes are "anchored" by the independent surface pressure, temperature, humidity, and wind values fed to the DigiCORA system during the launch procedure. These surface data become the first level 25 of the sounding profiles and the DigiCORA software uses them as it begins to process the data transmitted from the radiosonde and create the profile. Therefore, the surface observations influence the profile, generally only quite close to the surface for temperature and humidity and approximately throughout the lower 300-400 m for the winds. Surface data are also used at one point in the pre-launch process, when surface pressure is entered during the ground check so the Vaisala GC25 can calibrate the radiosonde's barometer. 30 At both locations there were two types of issues with the surface data that were entered into the DigiCORA software in the field. First, it was obvious during the campaign that at least some surface instruments were incorrectly calibrated or not well positioned. Second, site staff did not have real-time access to averaged data but had to make their best subjective estimate of representative values to enter, often from on-screen values that were updating at 1Hz. Therefore, the first step in producing research-quality rawinsonde data sets was to correct the archived surface meteorology at both locations, as described in Hartten 5 et al. (2017a). The resulting Kiritimati data (Hartten et al., 2017c) are at two-minute resolution through 11 February 2016 04:05:43 UTC and at one-minute resolution thereafter, while the corrected surface data from the NOAA Ship Ronald H. Brown ) are all at one-minute resolution. With the exception of the pressure data, as explained below, we consider these corrected surface meteorology data appropriate for initializing the research-quality rawinsonde observations. 10 At CXENRR, the radiosonde ground check occurred within one of the bungalow units, which was typically cooled by an air conditioner set to 20°-25° C. The surface meteorology instrumentation was outside, about 29 m from the Vaisala GC25 until 5 February 2016 21-22 UTC when it was relocated to a position about 39 m away. The barometer height at the surface station's final location was 3.6 m ASL. As the campaign progressed the observers realized they were seeing pressure differences between the radiosonde at the ground-check set and the radiosonde when co-located with the surface station. We 15 attempted to quantify this during the post-deployment data processing by comparing the original station-recorded (not corrected) surface pressure with collocated pressure data from 101 radiosondes (Fig. 5). The mean difference was 0.155 hPa, with the radiosonde pressure lower than the station pressure in spite of the fact that the radiosonde pressure had been set equal to the station pressure during the ground check. This difference would correspond to a height difference of 1.29 m between the GC25 and the station barometer, much greater than the measured 3 difference of 0.03 to 0.09 m. We therefore attribute part 20 of the pressure difference to a height difference between the two locations, but the main part to the pressure from the airconditioning system that slightly raised the room pressure. As a result, when creating the CXENRR radiosonde data sets documented here, the corrected CXENRR surface pressure described in Hartten et al. (2017a) was decreased by 0.155 hPa before being used as a reference pressure during the ground-check phase of reprocessing each sounding.

25
A similarly subtle issue with surface pressure occurred on the ship. In the field, the pressure entered into the software during the ground check was from the radiosonde itself while it equilibrated on the fantail. We did this because it was the only available measurement at that moment, but it was inadequate because it was not independent and the radiosonde is re-calibrated during this process. The reprocessing used surface pressure based on measurements from the ship's barometer, a Vaisala PTB330 last calibrated on 29 December 2015 and positioned at 15.24 m on the port side of the bridge. We first adjusted the 30 pressure reported by the PTB330 to 3.8 m, the height of the radiosonde ground check set; these became the values in the corrected surface meteorology data set. As documented in Hartten et al. (2017a) flow distortion around the superstructure of the ship made pressures from the PTB330 biased slightly high when the port side of the ship was facing into the wind and slightly low when it was on the lee side. Since the ship passed southbound through the trade winds during the first part of the cruise and northbound back through the trade winds on the return, the bias was generally positive during the first half of the 5 campaign (0.15 ± 0.21 hPa) and negative during the second half (-0.17 ± 0.24 hPa). We developed a correction as a function of relative (with respect to the ship's heading) wind direction and applied it to the height-adjusted PTB330 pressures (i.e. to the WTEC corrected surface pressures described in Hartten et al. (2017a)). We used the resulting time series during the radiosonde reprocessing, both as calibration inputs for the radiosonde ground checks and as surface observations. When there was insufficient data available to make the bias correction, the average bias (1.1 hPa) observed in the radiosonde pressure 10 sensors during the campaign was used to calibrate the radiosonde sensor. For additional details on the handling of ship surface pressures, see Hartten et al. (2017a).
After the surface data from both sites were corrected and the surface pressures subjected to the further adjustments described above, all soundings were reprocessed using the simulation feature of Vaisala's DigiCORA sounding software (Vaisala Oyj,15 2010a, b; version 3.64.1). The pressure, temperature, humidity, and winds, which would be entered as surface values during a real-time pre-launch process, instead came from the corrected surface meteorology data (plus the additional pressure corrections for CXENRR and for WTEC) at the time closest to launch. For WTEC soundings, initial position information came from the radiosonde, if available, valid at the listed time; otherwise, position information was from the corrected surface data set. The corrected surface winds for WTEC are based on true (earth-relative) winds. These were measured winds which 20 were adjusted by the onboard Science Computer System (SCS) to account for the ship's speed over ground, course over ground, and heading, but not for pitch, roll, and yaw. We viewed this as a more appropriate choice than ship-relative winds because ENRR's focus was the atmosphere's large-scale response to the strong El Niño and we therefore wanted soundings that were representative of the area in which the radiosondes were launched. Reprocessing was done using the standard "non-research" mode that is employed for operational synoptic soundings, and thus was only applied to the data collected as the balloons 25 ascended (Vaisala Oyj, 2010b). This mode corrects for solar radiation effects as well as for the pendulum motion of the instrument package below the balloon, and also performs some interpolation across small data gaps and some undefined filtering.
The effects of the interpolation and filtering in one particular situation can be seen in Fig. 6. The ship's log book identified 30 three soundings for which it appeared that the radiosonde passed through the ship's exhaust shortly after launch (Table 3); this feature is evident in the raw data, but there is no thermal signature in the final data files. We therefore performed a second simulation of those soundings in the "research" mode, which corrects for solar radiation and pendulum effects but otherwise "is designed to produce data that is edited as little as possible, because it is intended for research use only" (Vaisala Oyj, 2010b). Processing through the "research" mode leaves intact a temperature spike in the lowest 100 m of each sounding, but running the DigiCORA software in "non-research" mode removed the temperature spikes associated with the ship's exhaust.

Automated quality control
After the reprocessing, all soundings were subjected to automated quality control using ASPEN (Atmospheric Sounding 5 Processing ENvironment) software (University Corporation for Atmospheric Research (UCAR), 2016; version 3.3-265), which largely served to remove large random errors in the data. ASPEN The software was run with the "upsonde-1s" parameter settings, except for those settings related to GPS or satellites. ASPEN removed data that failed objective QC checks; smoothed the wind, pressure, temperature, and relative humidity; recomputed geopotential height; and computed the radiosonde ascent rate (Martin 2007). This post-processing with ASPEN satisfies the quality-control aspects of what Ciesielski et al. (2012)  10 define as a "Stage 2" or "Level 2" sounding data sets. While an assessment of the quantitative changes and uncertainties associated with the reprocessing described in section 3.1 and the quality control performed by ASPEN could be of interest to some segments of the user community, exploring that issue is beyond the scope of our objective. Such an analysis more naturally proceeds from the development of "Level 3" sounding data, which has been examined and corrected for systematic errors (bias). Ciesielski et al. (2012) includes a discussion of random and systematic errors in modern thermodynamic 15 soundings in that context, and contains citations to several relevant studies.
However, users need to be aware that automated quality control removes some but not all "bad" or "unrepresentative" data.
For instance, RS92-SGP radiosondes contain two pulse-heated humidity sensors, both of which are reconditioned during the ground-check procedure (Vaisala Oyj, 2015). Data generally flows from them alternately. If the two sensors are for some 20 reason giving realistic but different readings, large step-like swings in humidity are possible. These swings may be very obvious (Fig. 7a) or be followed by failure of both sensors (Fig. 7b). Sometimes the problem is clearly intermittent (Fig. 7d), and other times it is harder to discern whether it is intermittent or pervasive (Fig. 7c). While these swings may be obvious to the human eye and look non-meteorological, they do not trigger anything in ASPEN's automated quality control and are not removed. Situations like these would be systematically identified in the more time-consuming and person-intensive process 25 of creating what Ciesielski et al. (2012) term "Level 3" and "Level 4" data sets.

Soundings from Kiritimati Island
The 125 rawinsonde soundings from Kiritimati span nominal times 26 January 2016 12 UTC to 28 March 2016 00 UTC. The 27 February 2016 00 UTC sounding launch was timed to be as coincident in time and space as possible with a dropsonde 30 launched from the NOAA G-IV during one of its near-equator missions (Table 1). Height coverage of the CXENRR soundings was very good (Table 4, Fig. 8). Ninety-four percent of the soundings reached the Cold-Point Tropopause (CPT), defined here as the lowest temperature in the profile (Selkirk 1993, Highwood andHoskins 1998) that occurs above 15 km and with at least 2 km of sounding data above the level with lowest temperature. The minimum height criteria was designed to prevent flagging the top of a short sounding; because it is more than 1 km below the mean CPT heights found at Truk (7.44° N, 151.83° E) by 5 Mehta et al. (2011) we consider it unlikely that we are falsely identifying a Lower Tropopause (LT, Mehta et al. 2011) or other inversion below the CPT. Near-equatorial zonal winds at 40 hPa are used to define the phase of the quasi-biennial oscillation (QBO, e.g. Baldwin et al. 2001), and 88% of the Kiritimati soundings reached that level. The Kiritimati soundings (Fig. 8,   Fig. 9a) clearly show the complex layers of easterlies and westerlies observed in the Singapore soundings during early 2016 (c.f. Fig. 2 in Newman et al. 2016) as well as downward propagating wave activity (reviewed in Baldwin et al. 2001). 10 For the observers on Kiritimati, the most notable atmospheric feature of our stay was the rainfall; our surface station recorded 0.938 m of rain, nearly as much as the long-term annual average observed at the Kiribati Meteorological Service's Kiritimati station. Most of that rain (69%) came during 6 major events that were associated with the Intertropical Convergence Zone (ITCZ). The driest soundings (e.g. 7 March 2016 at 00 UTC, shown in Fig. 10) had only 0.04 m of precipitable water vapour 15 and featured a deep very dry layer in the mid-troposphere with a temperature inversion at the base. This deep dry layer appears to be associated with the strong high pressure system located between Hawaii and the California/Baja coast; in similar synoptic settings, the sharply bounded dry upper-troposphere was also observed by dropsondes from the G-IV. At Kiritimati, profiles like this frequently contained incursions of marginally wetter air, such as seen in Fig. 10a just below 300 hPa and between 400-500 hPa. The wettest soundings were collected when the ITCZ was over or very near the island (e.g. 22 March 2016 at 20 12 UTC, Fig. 11) and measured as much as 0.08 m of precipitable water vapour, with saturated or nearly saturated conditions extending up to the middle of the troposphere. In this instance, with the high centered near 35° N, the ITCZ extended west of  (Table 1). Height coverage was also very good at this station (Table 4, Fig. 12), with 89% of the radiosondes reaching the CPT and 87% reaching 40 hPa. When viewed in a simple time-height section (Fig. 12), the ability to identify lowerstratospheric phenomena is complicated by the ship's movement into, through, and out of the near-equatorial zone. However, 30 the stacked easterlies/westerlies/easterlies observed at Kiritimati (Fig. 9a) and at Singapore (Newman et al. 2016) are clearly present in early-to mid-March 2016 (Fig. 9b), and fine-scale descending waves are evident in the lower stratosphere above about 20 km through much of the cruise (Fig. 12). The ship's soundings captured a wide variety of conditions, including extremely dry layers between 300 and 500 hPa (Fig. 13a, collected when the ship was north of the ITCZ and on the edge of the northern hemisphere subtropical high) and strongly capped marine boundary layers (Fig. 13b) in the northern parts of the cruise, as well as deep tropical moisture nearer the equator (Fig. 13c, d).

Data Availability and Use 5
NOAA's National Center for Environmental Information (NCEI) is the archival location for these ENRR data sets. The DOIs are as follows: Level 2 rawinsonde data from Kiritimati (doi:https://doi.org/10.7289/V55Q4T5K), Level 2 rawinsonde data from the NOAA Ship Ronald H. Brown (doi:https://doi.org/10.7289/V5X63K15). The data sets, together with sample code to read them and skew-T plots extending to 100 hPa, are also available from NOAA/ESRL/PSD at https://www.esrl.noaa.gov/psd/enso/rapid_response/data_pub/. Users of these data must cite the appropriate DOI and 10 reference the data as indicated below: Each data set is comprised of many separate files, one for each sounding. The files are in the NASA Ames Format for Data 20 Exchange ("NASA-Ames Format") which is written in plain (ascii) text, making it highly portable to different computers and easily accessible to users who have limited computer resources. Detailed explanations of the format are found in Gaines and Hipskind (1998) and at the Centre for Environmental Data Analysis (CEDA)British Atmospheric Data Centre (BADC) website (Centre for Environmental Data AnalysisBritish Atmospheric Data Centre, 2002). NASA-Ames Format requires certain metadata in various header sections and provides for additional optional headers; we have endeavored to take advantage of 25 this aspect to make each file self-describing. It also requires that the total number of header lines be the first number in the first line of the file, and that the data following the header lines be space-delimited. Our files use the last header line to provide rudimentary column headers for the data that follow, to make it easier to import the data into spreadsheets or to quickly examine them by eye. Time is the independent variable in each file and is given in seconds since launch time. Users working with the data from Kiritimati Island, especially those interacting with residents and local meteorological data collections, may want to 30 be aware that in terms of civil time Kiritimati is in the Line Island Time (LINT) time zone (UTC+14).

Conclusions
The ENRR Field Campaign, conducted in January-March 2016 during a very strong El Niño event, included a wide variety of special intensive observations. Among those were frequent radiosonde launches from Kiritimati Island (2. 0°N, 157.4°E) and from the NOAA Ship Ronald H. Brown during its cruise along the near-equatorial TAO buoy transects at 140° W and 125° W. One radiosonde from Kiritimati and three from the ship were launched in coordination with dropsonde releases from 5 the NOAA Gulfstream-IV (G-IV; see Table 1), which conducted numerous research flights over the tropical Pacific as part of the ENRR field campaign (Dole et al., 2017).). This resulted in four pairs of radiosonde and dropsonde soundings, each pair separated in (horizontal) launch location by only 28-54 km and launched within a 5-8 minute window.
We have reprocessed 125 soundings from Kiritimati and 193 from the ship, using corrected surface pressure for the radiosonde 10 calibration and corrected surface meteorology as the first point of the sounding. We then subjected them to automated quality control via NCAR's ASPEN software. ASPEN removed and flagged values that failed objective checks, smoothed the wind, pressure, temperature, and relative humidity profiles; recomputed geopotential height; and computed the radiosonde ascent rate. We have written the resulting data to NASA-Ames format (ASCII) files, incorporating rich metadata in the header together with mandated information. The results are what Ciesielski et al. (2012) define as a "Stage 2" or "Level 2" sounding 15 data sets: data in a uniform, digital, easily read format which have been processed with automated quality-control checks to remove bad values. While both sites also collected the data reported by the radiosondes as they fell after balloon burst, only the data collected during radiosonde ascent have been reprocessed and included in these data sets.