In climate research, the role of water vapour can hardly be overestimated. Water vapour is the most important natural greenhouse gas and is responsible for the largest known feedback mechanism for amplifying climate change. Water vapour also strongly influences atmospheric dynamics and the hydrologic cycle through surface evaporation, latent heat transport and diabatic heating, and is, in particular, a source of clouds and precipitation.

Atmospheric water vapour is highly variable, both in space and in time. Therefore, measuring it remains a demanding and challenging task. The Zenith Total Delay (ZTD) estimates of GNSS, provided at high temporal resolution and under all weather conditions, can be converted to Integrated Water Vapour (IWV) if additional meteorological variables are available. In the past years, several long-term (20+ years) reprocessed GNSS tropospheric delay and water vapor time series datasets have been produced and have become available for climate studies. These include worldwide datasets (e.g. the International GNSS Service ‘IGS troposphere repro 1’), pan-European datasets (e.g. the EUREF Permanent Network ‘EPN troposphere repro 2’), but also regional datasets. However, although homogeneously reprocessed, these time series might still suffer from remaining inhomogeneities due to e.g. instrumental changes, environmental changes, etc., impacting the interpretation of the trends and long-term variability of the total column of water vapour.

In this contribution, we will describe the community activity on homogenization, in which we undertook an assessment of the performance of different break point identification algorithms on different synthetic datasets with inserted offsets. Furthermore, we will provide some examples of the use of GNSS retrieved IWV datasets for validation of (regional) climate models (e.g. EPN troposphere repro 2). Finally, we will show the results of different studies in which global GNSS IWV dataset are used, in complement to numerical weather prediction model reanalyses and satellite IWV retrievals, to understand and interpret the spatial and temporal variability of IWV.

The presentation will hence provide a review of the progress made in and the status of using GNSS tropospheric datasets for climate research, highlighting the challenges, and outlining the major remaining steps ahead.

Water vapor is the most variable parameter of the troposphere and accounts for most of the errors in the tropospheric modeling. The majority of water vapor is distributed below the height of 3 km and the assessment of the quality of tropospheric products in the lower troposphere still remains a challenge. Usually, the Global Navigation Satellite Systems (GNSS) estimates are compared with reference data. For the assessment of the vertical profiles, the radiosonde measurements are often used. However, such measurements have very low spatial and temporal resolution. Moreover, in Switzerland, the GNSS estimates are often underestimated in the lower troposphere compared to the radiosonde. To overcome these obstacles, we have designed a campaign to evaluate height and time dependent errors using a unique Unmanned Aerial Vehicle (UAV) called meteo-drone.

The meteo-drones were originally designed for automatic measurements of meteorological parameters in vertical profile (up to 3 km) with 4 Hz sampling rate. One profile measured with both ascending and descending drone takes approximately 30 minutes and there are multiple flights possible during one night. The measurements have been conducted in March 2019, in Marbach, Switzerland. We have collected pressure, temperature and humidity profiles from nine flights. From the measured meteorological parameters, we have calculated the refractivity and zenith tropospheric delays (ZTDs). Additionally, we have installed a low-cost dual-frequency GNSS receiver and antenna on the drone, from which we plan to obtain the tropospheric estimates at different heights. Moreover, the ZTDs are inferred from the neighboring permanent geodetic stations using the in-house developed least-squares collocation software COMEDIE (Collocation of Meteorological Data for Interpretation and Estimation of Tropospheric Pathdelays). From comparisons of the ZTDs interpolated from over 30 geodetic stations with the ZTDs calculated from meteorological parameters on the drone, we can establish the height and time dependent uncertainties of the tropospheric parameters.

Nowadays, many worldwide stations provide multi-GNSS observations in real-time. Such data can be analysed for obtaining different precise products to support various real-time applications. While global products such as satellite orbits and clocks together with code and phase satellite hardware biases are essential for Precise Point Positioning (PPP), local augmentation services providing atmospheric parameters are necessary for a very fast ambiguity resolution.
Geodetic Observatory Pecny (GOP) has been providing tropospheric parameters in near real-time using RINEX files since 2001, and has provided such products in the framework of the EUMETNET GNSS Water Vapor Programme (E-GVAP) service since 2004. In 2011, GOP initiated the development of the PPP real-time troposphere monitoring in support of low-latency, high temporal resolution, anisotrophy monitoring, all strongly benefiting of the new multi-GNSS observations. Initial fully operational service were completed and demonstrated within the GNSS4SWEC project and Real-time Demonstration campaign (2015-2018). We originally used a traditional processing strategy using code and carrier-phase observations forming the ionosphere-free linear combination. Recently, we have also implemented a strategy utilizing undifferenced and uncombined observations which does not eliminate the first order of ionospheric delays and thus need to be modelled by introducing new parameters. Advantage of this strategy is that both tropospheric and ionospheric parameters can be estimated efficiently with high accuracy in a single processing run. Moreover, multi-frequency and multi-constellation observations can be very flexibly and optimally used in the processing.
We routinely process real-time data streams from almost 200 stations in Europe. Estimated atmospheric parameters will be evaluated by comparing with the reference products and cross-validated when obtained at the nearby collocated stations. We will also show how precise tropospheric and ionospheric delays can shorten an initial convergence in PPP and improve kinematic positioning.

The analysis of Global Navigation Satellite Systems (GNSS) data includes identifying, estimating and mitigating error sources that affect the signal path while passing through the atmosphere. Of major importance among those effects is the existence of water vapour in the troposphere and its varying spatio-temporal distribution, which translates into so-called zenith wet delay (ZWD) and tropospheric gradients (GRD), respectively. In the standard GNSS data analysis, multiple day time series of ZWD and GRD are usually derived by independent daily estimations that are then combined in subsequent steps. That standard analysis approach is chosen mainly due to the fact that high observation frequency of the original GNSS data and thus large data sets make the continuous processing of multiple days computationally expensive. Such an approach is however sub-optimal as it results in discrepancies present on the daily boundaries. Furthermore, that standard approach is also sub-optimal with regards to the robust parameter estimation in a probabilistic sense. To address those deficiencies, a lightweight Kalman filter (KF) has been developed in the framework of the c5++ space-geodetic analysis software, which is able to provide very long time series of tropospheric parameters that are estimated continuously by employing the Precise Point Positioning (PPP) technique. To enhance the performance of that KF module, one phase ambiguity parameter per satellite-station pair is used and it gets reset when a satellite comes into view or when a cycle slip occurs. To enhance the stability and accuracy of this new analysis approach, the square root variant of the Kalman filter and related smoother are employed. In the interests of mapping the atmosphere in the most reliable and robust way, combination on the observation level among multiple co-located receivers is supported as well. Under the assumption that those receivers share the same atmosphere, common tropospheric parameters are derived while all other error sources are handled in a commonly known fashion. The current implementation of this new analysis approach utilizes GPS data, but the addition of Galileo is planned for the near future as well as the support for GLONASS and BeiDou constellations. The developed KF is tested using the data from two co-located GNSS receivers at Onsala Space Observatory (OSO). First, single-receiver monthly solutions are computed and compared to those obtained from other analysis software packages. Next, a combined solution from the two OSO receivers is performed and the resulting tropospheric parameters are studied in the context of station position repeatabilities. In addition, the derived time series of ZWD and GRD are compared to the independently derived results from a co-located water vapour radiometer (WVR) and from the analysis of geodetic Very Long Baseline Interferometry (VLBI) obseravtions with a co-located VLBI station. Thirdly, an analysis of how the computational cost scales with the varying time span of the estimation process is presented. Finally, the new analysis approach is stress-tested in a multi-month continuous analysis with the aim of revealing the restrictions and potential risks of the presented analysis setup.

In order to improve weather forecast, numerical weather models currently assimilate quantities that describe the water vapor content of the troposphere, such as zenithal tropospheric wet delays or IWVC (Integrated Water Vapor Content). Unfortunately, both quantities are hard to model since they vary both spatially and temporarily.
Meteorological agencies are highly interested in assimilating tropospheric information from GNSS measurements since they are available at high temporal rate and in all weather conditions. For this reason, the troposphere is also monitored by networks of high-grade (geodetic) GNSS receivers. The shortcoming of this technique is the limited amount of these base stations.
GNSS smartphone receivers, however, could provide additional data that would – in the best scenario – significantly improve the coverage and density of the GNSS-based tropospheric delay estimation, while decreasing the infrastructure and maintenance cost of high-grade GNSS receiver networks. The most recent chipsets are multi-constellation receivers (Samsung S8 and Huawei P10), and the first multi-frequency smartphone is on the market (Xiaomi Mi8) equipped with Broadcom BCM47755 chip.
The objective of this paper is to examine the feasibility to use smartphone measurements to retrieve zenithal tropospheric wet delays for weather monitoring. In this context, we extend the existing crowdsourcing tropospheric delay estimation algorithms to fuse observables from a smartphone receiver network.
The novelty of the proposed method lies in that (A) we introduce suitable constraints to keep the algorithm numerically stable, (B) we search for the “optimal” density on how to span the grid, and (C) we show that increasing the grid density over this optimal value does not improve the results.
In addition, the specific property of a smartphone receiver network is that receivers are added to it or removed from it at random times, leading to novel challenges for tomographic techniques. We propose to assess the tropospheric delay estimation algorithm performance when the amount and distribution of the receivers vary with respect to time.
Results are obtained by modelling the smartphone GNSS and barometer measurements, using the measurement campaign real data and realistic models done and developed by us in a separate work.
The paper presents the results of these activities and is organized as follows:
- Firstly, we review the tomographic techniques previously used to estimate the distribution of tropospheric water vapor, and then compare these against the accuracy and precision of other weather measurement techniques.
- Secondly, the crowdsourcing tropospheric delay estimation algorithm used to assess the system performances is presented. This algorithm is a state-of-the-art Bayesian tomographic model.
- Thirdly, the test-bed developed to emulate this system is presented.
- Finally, the system performances are assessed. A sensibility analysis is proposed to analyze the impact of major parameters, such as density, quality (noise level) and characteristics (GNSS constellations and signals) of smartphones. In detail, we examine the low and high limits of the receiver density, where density is the number of receivers per one tomographic grid cell.