Tables 2–5 show, on a parameter by parameter scale, a comparison of sensor characteristics. The RS41-SG uses a platinum temperature resistor while a band-gap temperature sensor is used in the Windsond S1H2. The silicon band-gap temperature sensor is a type of thermometer or temperature detector commonly employed in electronic devices. It has good stability in extreme environmental conditions due to the integral stability of crystalline silicon. Silicon band-gap temperature sensors operate on the principle that the forward voltage of a silicon diode is temperature dependent. Band-gap technology has the advantage of being low-cost, accurate, and reliable, as well as providing highly consistent measurements and having a positive temperature coefficient with a very low drift over time (Burlet et al., 2015).

Both sensors have the same resolution but the S1H2 has a smaller operational range. The platinum wire temperature sensor of the RS41-SG is both more accurate and has a faster response time than the band-gap sensor (Table 2; Vaisala, 2014; SparvEmbedded, 2016).

Both sondes use a thin film capacitor to make humidity measurements. These sensors provide a high accuracy, excellent long-term stability and negligible hysteresis. They are insensitive to contamination by particulate matter, are not permanently damaged by liquids, and are resistant to most chemicals. A capacitive humidity sensor works like a plate capacitor. The lower electrode is deposited on a carrier substrate, often a ceramic material. A thin polymer hygroscopic layer acts as the dielectric, and on top of this is the upper plate, which acts as the second electrode but which also allows water vapour to pass through it into the polymer. The water vapour molecules enter or leave the hygroscopic polymer until the water vapour content is in equilibrium with the ambient air or gas. The dielectric strength of the polymer is proportional to the water vapour content. In turn, the dielectric strength affects the capacitance, which is measured and processed to give a relative humidity measurement.

The RS41-SG humidity sensor integrates humidity and temperature sensing elements. Preflight automatic reconditioning of the humidity sensor effectively removes chemical contaminants in order to improve humidity measurement accuracy. The integrated temperature sensor is used to compensate for the effects of solar radiation in real time. The sensor heating function enables an active de-icing method in freezing conditions during the flight (Table 3 from Vaisala, 2014; SparvEmbedded, 2016).

The RS41-SG has a number of variants and of particular importance here are the RS41-SG and RS41-SGP. Although both sonde types provide pressure, temperature, humidity, and wind measurements it is in the manner in which pressure is derived that the difference arises. The SGP variant has the same pressure sensor as the RS92 sonde but with revised electronics and calibration, while the SG has no pressure sensor at all. In the latter case, the values of atmospheric pressure are calculated from satellite ranging codes, combined with differential corrections from the MW41 ground station. Pressure calculation also uses temperature and humidity from the radiosonde and the hypsometric equation.

The S1H2 measures the pressure with a microelectromechanical (MEMS) piezoresistor pressure sensor. This technology etches a diaphragm into a silicone substrate. Micro-piezoresistors measure the deformation of the diaphragm due to changing pressure.

The difference in performance characteristics (Table 4) between the two sondes arise from the S1H2 making direct pressure measurements while those of the RS41-SG are derived indirectly. The WMO radiosonde inter-comparison experiment 2010 (Nash et al., 2011) showed that pressure measurements, derived from geopotential heights and radiosonde measurements of temperature, and relative humidity profiles were very reproducible and suitable for all radiosounding operations wherein GPS systems are set up correctly, which includes the Vaisala system. This shows that the Vaisala-derived pressure is a reliable reference to assess the Windsond pressure sensor, and the Windsond cost can be lowered by removing the pressure sensor in future versions of the Windsond system, depending on its GPS system accuracy. Using a Windsond without a pressure sensor, however, requires an accurate pressure measurement at the surface if the pressure above the surface is to be computed using GPS altitude information, which requires a complementary external pressure sensor which can reduce the versatility of the Windsond system.

The Vaisala system uses on-board GPS receiver pseudorange to measure latitude, longitude, and height and applies a differential correction using the Vaisala ground station's GPS receiver. Use of differential GPS techniques in principle improves the accuracy and resolution of measurements. However, wind speed and direction are determined independently from the GPS position using the GPS Doppler frequency shifts.

The Windsond GPS ground station is not a GPS receiver; therefore, latitude and longitude are determined using on-board GPS receiver pseudorange without differential correction. Similar to the RS41-SG, the S1H2 wind speed and direction are determined independently from latitude and longitude using the GPS signal without differential correction explaining the two systems' similar performance characteristics as seen on Table 5.

The Vaisala system determines height using the GPS pseudorange with differential correction while the Windsond uses sonde pressure. The Windsond altitude algorithm tested here does not include hypsometric correction and is corrected in later versions.

The performance comparison between the Windsond S1H2 and the Vaisala RS41-SG shows the potential of the Windsond system which is able to closely match the temperature, pressure and humidity of the Vaisala RS41-SG even after processing by the MW41. However, when a sudden temperature and humidity change happen the slower response time of the Windsond system leads to temporary bias in the profile. The main weakness of the Windsond S1H2 lies in its GPS sensor and antenna which leads to a systematical error in wind speed and components, which complicates the observation of phenomena such as the NLLJ. A more advanced signal processing can improve the GPS sensor performances. The robust performance of the pressure sensor associated with the altitude systematic error shows that corrections in the altitude retrieval algorithm implemented in the latest versions of the Windsond firmware can improve the altitude measurement. The consistent pressure measurements are leading to the use of pressure level as the vertical reference for comparing the Windsond S1H2 and the Vaisala RS41-SG during the reproducibility experiment.

The assessment of sonde reproducibility is essential to guarantee the reliability of the sounding data during the data analysis: alterations of sonde performance under different atmospheric conditions have to be taken into account for a complete understanding of the data. The reuse feature of the S1H2 requires an evaluation of the data alteration due to sonde reuse in addition to the reproducibility evaluation using new sondes under different atmospheric conditions.

To complete both assessments, sondes were launched and retrieved until they got lost. To ensure, according to the authors, the best compromise between ensuring a satisfying recovery rate and a full LLC coverage, the cut-off was set at an altitude of 650 m a.g.l. At the preset cut-off altitude, two heating coils were activated and the string connecting the sonde to the balloon was burnt through. During the sonde descent, after the sonde loses contact with the ground station at approximatively 100 m a.g.l., the system automatically predicts and displays the expected landing point on a map view.

The ground station was carried to the predicted location and upon getting closer, approximately within 50 m, contact between the sonde and the ground station was established, the sonde immediately started to emit loud beeps (about 15 s time interval) and flashes of light. Signal strength increased when approaching the sonde and the vice versa. Once retrieved the sonde was switched off.

When reusing the sonde, the cup and lid were checked for any physical damage. The lid of the cup was then opened to confirm that there is no physical damage to any part (i.e. the heating coils or the printed circuit board). A 4 m polyester string (sewing thread) was wound around a cardboard (4cm×2cm×0.3 cm) cut-out with the ends left free: one to attach to the balloon the other to tie to the heating coil.

The sonde renewal strategy was based on sonde damage or loss. If a sonde was lost or any physical damage was not amendable for the next routine flight a new sonde was introduced. This strategy has been chosen to fully evaluate the degradation of the sonde, in terms of both retrieval and data quality but reduced the number of reproducibility flights with new sondes. The number of times each sonde had flown, as well as sonde recovery success, is detailed in Fig. 8. The results will be analysed and associated with the different reasons for a sonde loss.

Flights where an S1H2 has been launched simultaneously with another RS41-SG have been selected for the reproducibility and data alteration from the sonde reuse study. During the simultaneous flights, the RS41-SG and S1H2 were attached to different balloons and consequently did not climb at the exact same ascent rate. The comparison of each pair requires the data to be aligned at the same vertical level, and the systematic underestimation of the altitude by the S1H2 associated with the robust performances of the S1H2 pressure sensor led to the use of 1 hPa pressure ranges. For each pair, temperature, relative humidity, total, zonal, and meridional winds have been boxed in the pressure ranges. The pairs have then been sorted by the number of times the S1H2 was used, the median value for each range, and the number of uses that have been computed before a similar statistical comparison is performed on the median values.

Figure 8 details the sonde flight number, the flight success and the sonde recovery for each flight. More than 70 % of the sonde launches have been recovered, with sonde 468 being used eight times. The recovery rate could have been improved with more experience using the system and if the receptor had not been damaged due to the difficulties of carrying a laptop with an antenna in the tropical rainforest and different hazards such as tropical animals. The radio receiver RR2 with a Bluetooth connection seems promising for soundings in a difficult or harsh environment for overcoming these difficulties. Only five flights have been identified as unsuccessful, showing the overall robustness of the S1H2 radio antenna through the experiment.

The scatter plot in Fig. 9 compares respectively temperature, relative humidity, altitude, wind speed, meridional wind, and zonal wind recorded by the S1H2 and the RS41, boxed in a 1 hPa range and sorted according to the number of soundings of the S1H2, as indicated by the different markers, with colours indicating the corresponding altitude according to the RS41-SG with the maximum altitude set to 1000 m a.g.l. The presence of data over 650 m a.g.l. is explained by failures of the cut-off system leading to the loss of the sonde, but supplementary data are available for the comparison. For every parameter, the different markers are superposed randomly indicating the absence of performance degradation over time with the use of the S1H2 system. However, sonde S1H2 464 used for the sixth time systematically underestimated relative humidity and overestimated meridional wind, but sonde 468 when used for the eighth time did not show a particular anomaly, suggesting a contamination of the 464 sonde relative humidity sensor. Temperature and relative humidity of sonde 468 during its eighth flight at 800 m a.g.l. (yellow) showed the presence of a cloud top where the lag in the S1H2 answer was identified as it was in the performance flight.

Figure 10 shows the linear regression coefficient and the correlation between the boxed S1H2 and the RS41-SG data for each use. For temperature and altitude, the markers are superposed while for the other parameters markers are more spread but no clear trend can be identified. Sonde 464, which was used for the sixth time, with a low correlation and linear regression coefficient for relative humidity and large meridional speed linear regression coefficient, respectively, confirms the contamination damage on the sonde identified in Fig. 9. The relative humidity and low correlation of sonde 468 when used for the eighth run can be explained by the cloud top found in Fig. 9. The low or negative linear regression coefficient values for speed confirm the lack of accuracy in the performance flight and underline a need for improvement in the wind speed calculation from the GPS data.

The reproducibility experiment showed the robustness of the recovery system as well as the sensors. No clear performance degradation has been identified through the flights and the sondes have been recovered up to seven times. Similar performance weaknesses have been identified, such as the GPS sensor correction and the sensitivity to abrupt temperature and humidity changes.

However, the maximum altitude was limited to 650 m a.g.l. to ensure a satisfactory recovery rate which limits the use of the sonde recovery feature, and a sonde at its sixth use showed signs of contamination. A check of the sonde sensor values with ground instrumentation is consequently necessary before reusing the sonde to increase confidence in the measurement.