TOUGH Targeting Optimal Use of GPS Humidity Measurements in Meteorology EVG1-CT-2002-00080 6 th Periodic Report 1 February 2003 31 January 2006 Section 6. Final report Co-ordinator: Dr. Henrik Vedel email: tough@dmi.dk, http://tough.dmi.dk
2
SECTION 6: FINAL REPORT FOR THE TOUGH PROJECT...4 6.1. Background...4 6.1.1. Summary of project goals...4 6.2. Scientific/technological and socio-economic objectives...5 6.2.1. Project scientific and technological objectives...5 6.2.2. Project socio-economic objectives...10 6.3. Applied methodology, scientific achievements and main deliverables...13 6.3.1. The NRT GPS observing system in TOUGH...13 6.3.2. Assimilation of GPS data, impact studies...27 6.3.3. Estimation and assimilation of GPS slant delays...37 6.3.4. Error modelling and GPS system research...39 6.4. Conclusions, including socio-economic relevance, strategic aspects and policy implications...41 6.4.1. Main achievements...41 6.4.2. Main deliverables...42 6.4.3. Socio-economic relevance and policy implications....42 6.4.4. Overall conclusions...42 6.5. Dissemination and exploitation of the results...43 6.6. Main literature produced...47 6.6.1. Peer Reviewed Articles...47 6.6.2. Non refereed literature...48 6.6.3. Planned future publications...51
SECTION 6: FINAL REPORT FOR THE TOUGH PROJECT. 6.1. Background 6.1.1. Summary of project goals. Knowledge of the atmospheric distribution of water vapour is of key importance in weather prediction and climate research. It is tightly coupled to processes like energy transfer, precipitation, and is an important greenhouse gas. However, currently there is lack of knowledge about the actual humidity field, both due to a shortage of observations and a sub optimal handling of humidity in the data assimilation systems, which are used to make estimates of the actual atmospheric field. Such fields are used to start numerical weather prediction models and for climate monitoring. Global Positioning System (GPS) signals are particularly sensitive to water vapour. The main purpose of this project is to develop and refine methods enabling the use of GPS data from existing European GPS stations in numerical weather prediction models, and to assess the impact of such data upon the skill of weather forecasts. The GENERAL OBJECTIVES for the project are to improve the use of GPS data for numerical weather prediction and climate monitoring. This shall be done by innovation of new techniques and methodologies enabling proper correction of error sources identified in previous work, as well as by initiating use of the more detailed information available in the form of the individual delays between each receiver and the GPS satellites visible to it, rather than the single average type delay used by current methods. In the project we will: - Carry out research to optimise the assimilation of ground-based GPS in numerical weather prediction models. This research will include a proper modelling of the GPS measurement errors and application of more advanced assimilation techniques. Each step/component in the optimisation of the assimilation techniques will be verified by impact studies. - Develop methods for use of GPS slant delays in numerical weather prediction. Use of slants will enhance the amount of information available from each ground station. - Running a research mode data collection, by co-ordinated pre-processing and distribution of ground-based GPS measurements from Europe through a few European processing centres in support of the proposed data assimilation research efforts. The data processing centres will provide pre-processed data from subsets of the total European network, and each subset of the data should have comparable error characteristics. These error characteristics will be documented through comparisons of data from stations included in several of the network subsets (network overlap). - Investigate the benefit of using ground-based GPS-data in numerical weather prediction using the improved assimilation software through extended parallel data assimilation and forecast experiments, with and without ground-based GPS measurements, covering all four seasons. After the project, the resulting methodologies can be utilised by European weather forecast agencies at large, and the results help pave the road for a future co-ordinated, operational
European GPS moisture observation system. The exploitation of this new source of Earth Observation data is expected to benefit in particular the prediction of precipitation. In the longer run it will benefit also climate monitoring. When the Galileo satellites are launched the amount of observations of this type will increase and some of the error sources can be more easily controlled. 6.2. Scientific/technological and socio-economic objectives 6.2.1. Project scientific and technological objectives. The main purpose of this project is to develop and refine methods enabling the use of Global Positioning System (GPS) data from existing European GPS stations in numerical weather prediction models, and to assess the impact of such data upon the quality of weather forecasts. After the project, the resulting methodologies can be utilised by European weather forecast agencies at large, and the results help pave the road for a future co-ordinated, operational European observation system. The exploitation of this new source of Earth Observation data is expected to benefit in particular the prediction of precipitation. Weather forecasting of today is strongly dependent on the application of numerical weather prediction (NWP) techniques. Starting from initial states representing the atmosphere at a certain time, numerical models are integrated forward in time to obtain the future state of the atmosphere. The initial atmospheric states, the quality of which are of crucial importance to the quality of the forecasts, are obtained from the time history of observations through a process that is generally referred to as atmospheric data assimilation. Thousands of observations are required for the determination of the state variables of the atmospheric models, the most important ones being vertical profiles of wind, temperature and moisture, in addition to the pressure at the surface of the earth, and radiances measured by satellites. Throughout the history of NWP, the observation and model initialisation of the moisture has been treated with less care than the other variables. The moisture initialisation has generally been carried out without coupling to the initialisation of temperature, surface pressure and wind. Only radiosonde observations of atmospheric moisture profiles have been available, and these observations are often not representative of the scales of motion described by the models and are also affected by observational errors. Remote sensing observations and modern data assimilation methods, based on e.g. variational techniques, have the potential of bringing the moisture field initialisation to a more advanced state. The measurement of the atmospheric delay of radio signals from navigation system satellites, such as the GPS, offer an opportunity for the NWP community to get access to high quality atmospheric moisture information from already established networks of GPS ground stations. The atmospheric delay of GPS radio signals is due to the sensitivity of atmospheric refraction to atmospheric pressure, temperature and moisture. The total delay of the radio signals between a GPS satellite and a GPS ground station is essentially dependent on the total atmospheric mass, i.e. the pressure at the surface, and the columnar atmospheric moisture content. Provided the surface pressure can be determined from another source of information, e.g., an NWP model, the delay of the GPS signals provides 5
a unique source of information related to the atmospheric moisture content. Normally the GPS data processing results in a single delay measure, reflecting the average properties of the atmosphere around the site. More advanced techniques, which determines the delay between the site and each GPS satellite on the sky are being introduced thereby enhancing the information about the local spatial variability of the atmosphere. The utilisation of data from GPS ground stations for numerical weather prediction, and also for climate monitoring and research, is the subject of the COST Action 716 (Exploitation of Ground-based GPS for Climate and Numerical Weather Prediction Analysis). Several of the members of COST 716 Action have furthermore contributed to EC-funded MAGIC (Meteorological Applications of GPS Integrated Water Vapour Measurements in the Western Mediterranean) Project. Considerable progress has been achieved both within COST 716 and within the MAGIC project. The quality of the data has steadily been improved and the extraction techniques work in near real time and are approaching operational status in Europe. COST 716 data assimilation tests for the June 2000 period using Central and Northern European model integration areas have indicated significant bias (systematic observation error) problems between the GPS total zenith delay measurements and model predictions. Preliminary results from MAGIC assimilation show a neutral impact in the overall statistics over 2 weeks of data, but indicate positive impact for rapidly evolving localised storm systems or in situations where the humidity field is not dominated by large-scale dynamics. Thus, GPS delays are potentially very useful to meteorology, but further research is needed before the GPS data can be used in an optimal way to the benefit of numerical weather prediction. It is based on these promising results that 7 meteorological institutes now join forces in this project in order to optimise the methods by which GPS data can be utilised in NWP models. In total 15 institutes will partake in the project, seven of which will process the GPS data into zenith delays do research on improving such processing. The GENERAL OBJECTIVES for the present project proposal are to improve the use of GPS data for numerical weather prediction and climate monitoring. This shall be done by innovation of new techniques and methodologies enabling proper correction of error sources identified in previous work, as well as by initiating use of the more detailed information available in the form of the individual delays between each receiver and the GPS satellites visible to it, rather than the single average type delay used by current methods. Considering the experiences and the achievements from the COST 716 Action and from the MAGIC Project, these general objectives may be stated more precisely through the following verifiable sub-objectives: - Carry out research to optimise the assimilation of ground-based GPS in numerical weather prediction models. This research will include, for example, a proper modelling of the GPS measurement errors and application of more advanced, 4-dimensional, assimilation techniques. Each step/component in the optimisation of the assimilation techniques will be verified by impact studies. - Develop methods for use of GPS slant delays in numerical weather prediction. - Running a research mode data collection, by co-ordinated pre-processing and distribution of ground-based GPS measurements from Europe through a few European 6
processing centres in support of the proposed data assimilation research efforts. This work will be closely linked with the COST 716 Action. The data processing centres will provide pre-processed data from subsets of the total European network, and each subset of the data should have comparable error characteristics. These error characteristics will be documented through comparisons of data from stations included in several of the network subsets (network overlap). - Investigate the benefit of using ground-based GPS-data in numerical weather prediction using the improved assimilation software through extended parallel data assimilation and forecast experiments, with and without ground-based GPS measurements, covering all four seasons. Special emphasis will be devoted to the verification of precipitation forecasts. - Promote the idea of an operational utilisation of ground-based GPS measurements to the numerical weather prediction community in Europe. State of the art in GNSS observations of water vapour The raw GNSS data consist of ranging measurements from visible navigation system satellites such as the Global Positioning System (GPS). If the positions of the satellites and receivers are precisely known, the ranging measurements can be used to detect delays due to the atmosphere. This is possible since the propagation speed of the radio signals is sensitive to the refractive index of the atmosphere, which is a function of pressure, temperature and humidity, and the ionospheric electron content. The ionospheric delay is dispersive and can be removed using observations on two frequencies. The remaining accumulated delay for a raypath is the integral of the refractivity along the trajectory of the ray through the atmosphere P e e d = Ndl N = k + k + k T T T 6 d 10 where 1 2 3 l 2 The refractivity N is described as a function of temperature T, the partial pressure of dry air P d, and the partial pressure of water vapour e and constants, k 1, k 2, and k 3, which have been determined experimentally (Smith et al 1953, Thayer 1974, Bevis et al 1994). Small scale horizontal variations may be neglected, to first order, so that observations at all satellite elevation angles can be mapped to a single zenith delay value which can then be transformed to integrated water vapour with auxiliary information on the surface pressure field (Bevis et al 1992). Since the concept was initially proposed, the quality of the data has steadily improved through several major efforts, for example the EC projects MAGIC (Haase et al 2001, Vedel et al 2001) and WAVEFRONT (Dodson et al 1999), and NEWBALTIC (Emardson et al 1998), and the U.S. ARM (Gou et al 2000), GPS/STORM (Rocken et al 1995), CORS (Fang et al 1998), and CLIMAP(Haas et al 2001), 7
ZTD GPS - RS Std dev (mm) 25 20 15 10 5 0 1999_01 Time dependence of the GPS - Radiosonde ZTD Difference 1999_03 1999_05 1999_07 1999_09 1999_11 2000_01 2000_03 2000_05 CAGL CART CASC GRAS GRAZ HFLK KOSG MEDI OBER SJDV VILL ZIMM ACOR BRST Figure 1 Time dependent behaviour of the standard deviation of the GPS-radiosonde ZTD difference over a 1.5 year time period in the Mediterranean area. MAGIC (Meteorological Applications of GPS Integrated Column Water Vapour Measurements in the Western Mediterranean) was a 3-year research project financed in part by the European Commission to develop the tools necessary for the meteorological users to integrate the GPS derived humidity products into their numerical weather prediction models, and test these models in severe storm situations. In the project, a prototype system for deriving and validating robust GPS integrated water vapour (IWV) and zenith tropospheric delay (ZTD) data sets was developed, both in post-processing and near-real-time mode. An extensive a database of 1.5 years of ZTD data is available for more than 50 sites in Spain, France, and Italy. The database has been validated through continuous comparisons with radiosondes. The comparison shows differences with a standard deviation on the order of 10 mm ZTD (see fig. 1) or the equivalent error in IWV of 1.6 kg/m 2. The continuous comparison with independent data sets demonstrated that there are long-term differences that require further investigation, especially for climate applications. Continuous comparisons with HIRLAM NWP fields show a standard deviation of 17 mm ZTD or 2.7 kg/m 2. A higher standard deviation for the HIRLAM fields than radiosondes indicates that there is significant information contained in the GPS observations that is unknown to the NWP model, and hence the potential to improve the model. State of the art in meteorological data assimilation The European weather services have invested scientific development efforts over the past 5-10 years into a new generation of data assimilation based on variational techniques. The 3-dimensional versions of these assimilation schemes (3D-Var) have recently been introduced operationally (Lorenc et al 1999, Gustafsson et al 2001). One of the advantages of these variational data assimilation schemes is the possibility to utilise observed 8
quantities with complicated, e.g. non-linear, relations to the forecast model variables. Thus it is, for example, possible to directly assimilate the atmospheric delay data as measured at the ground-based GPS stations. Early trials to assimilate simulated ground-based GPS measurements with simplified variational data assimilation schemes were carried out by the Mesoscale Meteorology group at the National Centre for Atmospheric Research (NCAR), Boulder, USA (Kou et al 1996, de Pondeca et al 2000). The main limitation of these early NCAR trials with variational data assimilation of GPS data was the lack of a background error, thus the forecast errors were not described properly and therefore the assimilation became sub-optimal. The more mature variational data assimilation schemes developed by European weather services for operational purposes included proper background error constraints. The meteorological services involved in the COST 716 Action and the MAGIC Project developed and tested 3D variational methods for the assimilation of ground-based GPS data. Assimilation tests were carried out for a 2 weeks period in June 2000. The overall large scale statistical impact on forecasts of temperature, wind, and humidity fields was neutral for the GPS ZTD data set, which was not unexpected given the number of GPS ZTD observations compared with conventional observations. However, rainfall forecasts for specific case studies were improved, especially in localised regions of high precipitation (see fig 2, next page). This was a very encouraging result, that was undetectable in the overall statistics, but has the potential to have a significant socio-economic impact, since these intense short duration high precipitation events are a principal cause of weather related damage in the Mediterranean region. On the other hand, COST 716 data assimilation tests for the same June 2000 period and for Central and Northern European model integration areas have indicated significant bias (systematic observation error) problems associated with the GPS Total Zenith Delay measurements. These bias problems were temporarily avoided by introduction of Bias Reduction Algorithms, based on a comparison between GPS measurements and forecast model data. The origin of the problem is yet not clear, however. Simulation studies and results from trials to model the spatial correlation of GPS observation errors support the possibility of slowly varying and horizontally correlated observation errors associated with the GPS measurements. 9
Figure 2 (left panel) observed 12 hour accumulated precipitation for an event the 9 June 2000 which produced high rainfall in the Pyrenees and north-eastern Spain, (centre panel) forecast precipitation without GPS data, (right panel) forecast precipitation with GPS data. European geodesists and meteorologists have joined forces in the COST 716 Action on Exploitation of ground-based GPS for climate and numerical weather prediction application, with participation from 17 European countries. A benchmark data collection, near-real time processing, data distribution and data assimilation test was successfully carried out for a two-week period in June 2000. A near-real time data collection, processing and distribution exercise is continuously ongoing from April 2001 until February 2002. A working group (WG4) on the design of an operational ground-based European GPS network for meteorological purposes has started its activities. Innovation goals by the project - Optimisation of the 3 dimensional assimilation of ground-based GPS data by a proper modelling of observation error biases and spatial/temporal correlation - Development of 4-dimensional assimilation to utilise the temporal resolution of the GPS data. - Processing, validation and assimilation of GPS slant delays. - Development of methods for assimilation of GPS slant delays in 3 dimensional data assimilation - Investigation of the optimal use of the GPS data in meteorology by extended parallel data assimilation and forecast experiments distributed over all seasons, by objective and subjective verification. 6.2.2. Project socio-economic objectives The project contributes specifically to Objective 1 and Objective 2 of Key action 7.2 Development of generic Earth observation technologies. Objective 1: Introduce scientific results into new or existing applications, and Objective 2: Improve the exploitation of Earth observation. Measurements from an existing network of ground-based GPS stations, developed mainly for geodetic purposes and available with very minor additional costs, are utilised for numerical weather prediction purposes (Objective 2). This is a new application of existing geodetic measurements, and it will allow numerical weather prediction centres to specify the initial moisture field with an accuracy and detail that has not been possible in the past. Improved and more detailed forecasts are expected, in particular of precipitation. Furthermore, the utilisation of ground-based GPS measurements in numerical weather prediction has been made possible through the introduction of advanced data assimilation schemes like 3- and 4-dimensional variational data assimilation (3D-Var and 4D-Var), that have been developed through significant scientific efforts over the past 5-10 years (Objective 1). The operational numerical weather prediction models that provide European citizens with 1-2 day forecasts cover approximately a quarter of the globe, even for the limited area 10
regional models such as HIRLAM. These models require, particularly for rainfall prediction, a dense, evenly distributed and accurate observation of the water vapour field. The global NWP models that provide boundary conditions for the operational models require improved observations as well. This means that resolving problems with deficiencies in the observing system for European users requires European scale efforts. Co-operative initiatives have existed for some time for radiosonde and other observation networks, as well as for atmospheric remote sensing from space. The ground-based GPS data offer the advantage of an existing station network and existing data processing capabilities. It is necessary when considering using GPS as an atmospheric remote sensing source that the data collection and data processing efforts are shared between European member states so that the observation system is dense enough to have significant impact. In particular for detailed rainfall forecasts during the warm season with the next generation mesoscale NWP models, a station density of approximately 30 km is required. Furthermore, a carefully co-ordinated European GPS data processing effort is necessary in order to guarantee a homogenous and well documented data quality. Furthermore, a carefully coordinated European GPS data processing effort is necessary in order to guarantee a homogenous and well documented data quality. Therefore, the contribution that GPS data can make to resolving the problem also requires European scale efforts. European added value for the consortium The objectives proposed in TOUGH benefit by carrying out the research as a collaborative effort at the European level. Bringing together a strong international group of people working on similar problems will increase the efficiency of the development. Problems will need to be tackled in terms of NRT delivery, orbit determination, stability of reference frames, and understanding of noise sources that require international co-operation. The joint efforts of the geodetic and the meteorological communities in the present proposal will give added value in the form of improved data quality for both of these communities. By using the GPS information, the meteorologists will be able to improve the quality of the modelling output products, and these improved products can help the geodetic community to improve their data processing (feedback loop) and products for geodetic applications. Improving the quality of life and health and safety Daily forecasts and warnings of severe weather situations have become a vital asset for many areas of life and activities for public and industrial businesses with an everincreasing economical importance. The use of accurate and timely weather information, both actual and forecast, is essential for the operation of air transport. As air traffic grows, new aeronautical systems are being developed in order to optimise aircraft operation and thus to ensure a high level of safety and in particular in the meteorological area. The humidity observations are expected to make the significant improvements in weather prediction for European citizens. The most improvement should come in precipitation forecasts of European models. Therefore the public should benefit most importantly from more reliable warnings that are especially useful for the prediction of floods, increasing the safety of European citizens. 11
TOUGH makes an important contribution to the development of skills in Europe: There will be many of the activities in the next five years in research disciplines related to GPS systems and the use of the future European GALILEO system. Public uses of precise satellite time and positioning information together with a growing commercial market made it clear that modern highly developed countries need access to such skills. TOUGH project will contribute to develop such skills. One of the major error sources in precise global time and position information is the contribution from atmospheric phenomena. TOUGH research results will increase the ability of European users to use GPS information. Modern numerical weather prediction systems and climate research activities rely more and more on satellite Earth observations. TOUGH will develop skills in Europe in the most forefront data assimilation theories and applications related to GPS information. TOUGH reinforces the development of state of the art and innovative GPS techniques at the higher education level that will be directly and indirectly transferred to increase the skills of the European workforce. Preserving and/or enhancing the environment Water vapour structure plays a very important role in environment forecast and monitoring. TOUGH will improve knowledge of the water vapour structure by providing complementary observation data and by improving data assimilation systems. A better understanding of the water vapour structure has also important implications for climate change researches. TOUGH will start to investigate the long term error characteristics of GPS measurements, which will be a stable independent data source for climate monitoring. Economic benefits There are short-term economic benefits due to increasing the cost-effectiveness of the European observing systems. For example, the objectives of the EUCOS programme are to increase the cost-efficiency of the European observing system over the continents while staying at the same overall cost. This is proposed by replacing radiosondes with AMDAR airplane soundings, however the AMDAR soundings do not contain humidity information. The ground-based GPS ZTD data will provide supplementary humidity information that allows this cost-redistribution with less negative effect on the forecasts. Thus the costeffectiveness of the GPS observations are very beneficial to end users at the European level as well as at the level of the national meteorological agencies. There are long term economic benefits from the improvement in NWP, in particular for forecasting of severe weather, as the ground-based GPS data are expected to improve the humidity analysis and lead to better forecasts for humidity and precipitation. The improved weather analyses and forecasts deliver benefits to the European public, to aviation, and to the fishing and shipping industries, both in terms of safety (meteorological hazards and dangers detection and avoidance) and cost effectiveness (e.g., improved flight profile and conduct). 12
Strategic impact The investment in the research for developing methodologies to exploit the GPS observations will place Europe in a leading position on the international scene in this domain. Current research in the U.S., for example, does not have the equivalent density of potential observations, nor the breadth of experience with different assimilation approaches. These are key to successful exploitation of the data. Improvement of weather prediction quality will add to the competitiveness of meteorological agencies that supply services to the European public, to aviation and to marine transport industries. Participation in the project will help the research institutes involved to remain at the forefront of their respective fields. 6.3. Applied methodology, scientific achievements and main deliverables 6.3.1. The NRT GPS observing system in TOUGH. A primary goal in TOUGH was the establishment of a European wide NRT GPS ZTD producing observing system as part of the project. Both to ensure NRT GPS ZTD data from an area large enough to be of importance to regional European NWP models, and to enable a tight collaboration between NRT GPS ZTD producers (geodetic centres) and users (NWP centres). 8 of the TOUGH partners (ACRI-ST, NMA, Chalmers, LPT, GOP, ASI, IEEC and METO) did NRT GPS ZTD processing from raw GPS data. It should be noticed that the latter institute started NRT GPS ZTD processing not as part of the TOUGH project, but as a preparation for use of NRT GPS ZTD data in the UK. In addition a number of other institutes processed NRT GPS ZTD data from other sites, such that at then end of TOUGH in total 12 centres were producing GPS ZTDs. In the last year of TOUGH GFZ (GeoForschungsZentrum Potsdam) became an associated member of TOUGH, which was an important step as a significant of all NRT GPS ZTD data processed in Europe are from Germany. The fine relations between geodesists and meteorologists established in COST Action 716 were maintained through the TOUGH period, resulting in a NRT GPS ZTD observing network which is geographically more extended than what could have been accomplished by the TOUGH partners themselves. From TOUGH we are extremely thankful to all centres having produced NRT GPS ZTDs to the TOUGH data-server, besides GFZ these include BKG, ROB, and SGN. Further we are extremely thankful to all owners of GPS sites that have provided data for the NRT GPS ZTD estimation. The GPS ZTD processing centres applied different strategies for their NRT estimation of GPS ZTD and used different softwares, with different setups for the processing. Initially in the project there was a focus on timeliness, the requirement from the User requirements document (D15) that the GPS ZTD estimates should be available for NWP within 1h 45min of their valid time. Later more focus was paid to the quality of the NRT ZTDs. 13
TOUGH deliverables D51 to D64 contain work related to estimation and validation of the NRT GPS ZTDs. In addition the yearly reports from the processing centres provide details on the processing methods adopted at each individual processing centre. As an example of the approaches adopted figure 6.3.1.1 shows the GPS stations processed by ACRI-ST, with a division into sub-networks to speed up the data processing. Figure 6.3.1.2 shows the processing strategy adopted by ACRI-ST. Figure 6.3.1.3 shows the processing strategy adopted by ASI. The latter centre makes both a NRT and a post processed solution. The post processed is more precise, being based on better estimates of the GPS satellite orbits, etc. At ASI this produces both station coordinates, which in the form of a monthly mean coordinate is used in the NRT solution, and it produces post processed ZTDs which can be used for validation of the NRT ZTDs. During the course of the projects the processing strategies and softwares improved. A comprehensive account of the processing strategies and softwares used early in the project, including those of non- TOUGH partners, can be found in the COST Action 716 Final report (Elgered et al 2005). The most recent processing strategies adopted by the TOUGH partners are described in the TOUGH reports available at the TOUGH homepage. Fig 6.3.1.1. Stations processed by ACRI-ST. They are processed as two sub networks (red and yellow) to speed up the solution. Grey dots represent IGS stations used in both subnetworks. (ACRI-ST, D57) 14
Fig 6.3.1.2. Analysis strategy adopted by ACRI-ST. (ACRI-ST, D57) Figure 6.3.1.3. Analysis strategy adopted by ASI, which makes both NRT (upper part, every hour) and a post processed (lower part, once a day) solutions (ASI, D57). The GPS ZTDs from all the processing centres were assembled at a central ftp-server at METO, see figure 6.3.1.4. Initially the GPS ZTD data had to be downloaded by ftp from the server by the NWP institutes, but during the course of the project most of the data became available also via the GTS network, which is a separate network used by the national met-services for interchange and distribution of meteorological data. 15
. Fig 6.3.1.4. Data flow for the NRT GPS ZTD data. Thorn is the ftp-server operated by METO for TOUGH. (METO, D64) During the course of the project the number of GPS sites from which GPS ZTD data were available increased tremendously. Figure 6.3.1.5 shows the evolution in available GPS ZTD data from May 2001 to January 2006. Notice that both the number of GPS sites processed by the individual data processing centres and the number of processing centres increased. Fig 6.3.1.5. Number of GPS sites for which NRT GPS ZTD were available as function of time. (METO, D64) For operational NWP a main issue is timeliness. Each new NWP forecasts is started a certain time after the nominal time of the data assimilation, less than two hours for regional (national) forecasts. In TOUGH emphasis was put on the ability to meet such goals. Figure 6.3.1.6 shows the time for 75 percent of the GPS ZTD reports from different processing centres to reach the ftp-server thorn. It can be seen that all processing centres for an 16
extensive period were able to meet the timeliness requirement. After having established that more attention was paid to increasing the quality of the GPS ZTD data. Figure 6.3.1.6. Monthly summary of time delay for 75% of observations to arrive in METO database for each processing centre; the target. The horizontal black line is the target. From May 2003, all ACs have been delivering their NRT data within 3 hours and from November 2003, are mostly within 1h 45min. By May 2005, this was generally reduced to 1h50m. The performances noted here for January 2005 are not representative (METO, D64). Corresponding to the increase in number of GPS sites (figure 6.3.1.5) there was a gradual change in the geographical density of the GPS ZTD observing network. Figure 6.3.1.7 shows the distribution of the GPS sites toward the end of the project, January 2006. Figure 6.3.2.3 gives an idea about the distribution of GPS sites for which NRT ZTDs were available at various times at which NWP impact experiments were performed. 17
Fig 6.3.1.7. Distribution of GPS sites with NRT GPS ZTD for January 2006. (METO, D64) A separate issue of great importance for use of NRT GPS ZTDs in NWP is the quality of the GPS ZTD data. To monitor that, radiosonde and NWP model data were extracted and used for calculation of independent ZTD estimates for comparison. Further NRT GPS ZTD were compared to post processed GPS ZTD, for which more precise GPS satellite orbits and clock errors are available. Figure 6.3.1.8 shows a comparison between the GPS ZTD sequence for a month for station WSRA, as a time series and as a scatter plot. The NWP data are from the METO local model 18
Figure 6.3.1.8 28-day time series and scatter plot of observed ZTD (station WSRA processed by GFZ) vs. ZTD derived from the METO mesoscale model. (METO, D64) KNMI has NRT comparisons not only against NWP (HIRLAM) but also vs. radiosondes. Using nearby synoptic observations of pressure and temperature, ZTD values are converted to IWV, and these are also validated. An example of such a plot is shown in figure 6.3.1.9. The example is for a GPS station which is processed by 4 different processing centres. One of the recommendations from TOUGH for an operational GPS ZTD observing network is exactly that all processing centres shall process a set of common stations for inter comparison purposes. During the cause of the TOUGH project the quality of the NRT GPS ZTDs have been gradually increasing. As an example ACRI-ST determined correlation coefficients computed between their NRT GPS ZTDs and NWP ZTDs for 6 months in 2003 and for the entire years 2004 and 2005 and found that the coefficients of correlation are all greater than 0.9 in 2003 except for 5 sites where the values are between 0.82 and 0.89. In 2004, coefficient of correlation are better than 0.91 for all but 1 site that shows a correlation coefficient of 0.89. And finally, coefficient of correlation are better than 0,93 in 2005. The mean value of the correlation coefficients is 0.93 in 2003, 0.94 in 2004 and 0,97 in 2005, demonstrating an improvement of the correlation coefficients during with time. (ACRI-ST, D57). 19
Figure 6.3.1.9. Time series comparing GPS ZTD and IWV against model (HIRLAM) and radiosonde equivalents, for site TERS. From KNMI website (S. de Haan). Quality problems On occasion, there are various anomalous data points, which can be classed as: Constant offsets Variable bias Transients (occasional large spikes ) Consistent low level noise, jumps or other inaccuracies Consistent severe noise Mixed quality of data for the same station; some processing strategies appear to give more problems than others. Where ongoing problems have been identified, the AC has been informed so that the issue can be investigated and (if possible) corrected. Studies have been done of the variation in offsets between GPS derived ZTD/IWV and corresponding radiosonde and NWP measures as function of both time and time of the day. Figure 6.3.1.10 show the bias, standard deviation and RMS for the stations POTS and BOGO based on NRT GPS ZTD from GOP and HIRLAM ZTD from DMI (GOP, 2006 yearly report) In these cases both bias and standard deviation varies with the time of the year. This could indicate necessity of using time varying bias corrections and observation error estimates in NWP data assimilation. 20
Fgure 6.3.1.10. Example of seasonal variation of bias and offset between NRT GPS ZTD and the corresponding model estimate. From GOP That this is not a particular behaviour for just two stations is seen in similar comparisons for more stations. Figure 6.3.1.11 (bias) and 6.3.1.12 (standard deviation) show a comparison of ACRI-ST derived GPS IWVs against HIRLAM deduced IWVs. (ACRI-ST, D57). Figure 6.3.1.11. Bias of GPS versus HIRLAM IWV, July 2003 to Jan. 2006. (ACRI-ST, D57) 21
Figure 6.3.1.12. Standard deviation of GPS versus HIRLAM IWV offsets. (ACRI-ST, D57) For this larger sample systematic seasonal changes in the biases are less visible than in figure 6.3.1.10, whereas the seasonal change in the standard deviations of the offsets is very pronounced. The warmer summer atmosphere contains more moisture, and water vapour has a greater tendency to vary spatially caused by much stronger convective systems. The NWP models have difficulties simulating that, and the symmetry assumptions adopted in GPS processing software will more often be in error. Both effects leading to an increase in the standard deviations of the IWV (and ZTD) offsets. The NWP data assimilation and forecast cycles follow the same patterns day after day. Typically a data assimilation is done every six or three hours after which a new forecast sequence is made. This has a visible effect on the GPS ZTD versus NWP ZTD statistics. Figure 6.3.1.13. Bias of GPS ZTD HIRLAM ZTD versus time of day. (GOP, D57). 22
Figure 6.3.1.14. Standard dev. of GPS versus HIRLAM ZTD against time of day. (GOP, D57). Figure 6.3.1.13 show the bias of the GPS-HIRLAM ZTD as function of the time of the day. Clear jumps are visible from 05 to 06, from 11 to 12, from 17 to 18, and from 23 to 00. These corresponds exactly to the times at which data analyses are performed (00, 06, 12, 18 UTC). The HIRLAM data are extracted at DMI from a model without GPS data in the data assimilation. The humidity information to the model comes from radiosondes, data from which are primarily available for the 00 and 12 UTC analyses. It is interesting to note the tendency of the biases to increase at certain sites at those time, while the standard deviations seem to decrease at the same times at most sites. In NWP data assimilation systems certain assumptions are made about the distribution of the distribution of the errors of the offsets between observations and model estimates thereof. If the assumptions (Gaussian distribution with zero mean) are not more or less met the data assimilation may not lead to proper results, even the observations are precise. It is therefore important to consider the offset distribution. Figure 6.3.1.14 is an example of that. Here three NRT processing methods (different versions of the same processing software), varying vertically, are used for three GPS sites, varying horizontally, are compared against NWP HIRLAM ZTD estimates. It is clear that the upper, oldest version of the processing software produces results more far from the assumptions of the data assimilation system. 23
Figure 6.3.1.14. Distribution of GPS versus HIRLAM ZTD offsets for three GPS sites (BOR1, GOPE, POTS, left to right) for three different versions of the same processing software. (from partner GOP). The same method can be used to characterise the NRT GPS ZTD results obtained at different processing centres against more precise GPS ZTD estimates made using post processed data for a set of common stations. In figure 6.3.1.15 this is done for 9 GPS stations and 9 processing centres. This can be used to identify ways in which to refine the NRT data processing. 24
25 Fig 6.3.1.15. Distribution of NRT ZTD versus post processed ZTD. Processing centre at top, station name at left. (From GOPE presentation at final meeting.) B O R 1 G O P E H E R S P O T S W T Z R O N S A M A R 6 C A G L M A T E ACRI ASI BKG GFZ GOP IEEC LPT NKG NKGS
At the final meeting the NRT processing of GPS data was reviewed based on all the results obtained in TOUGH. Secondly, recommendations were made as how to set up processing of ground based GPS data in a future operational continuation: Achievements The number of unique stations increased from 150 in Feb. 2003 to 550 in Jan. From 5 Analysis Centres to 12 at the end of TOUGH From 225 k observations/month to 1000 k at the end of TOUGH All centres met timeliness requirements for at least 4 contiguous months at some time during the project. Data delivery to end users were robust during the project. Today 99% of the data are on GTS in BUFR format. The analysis centres have demonstrated consistent results (among themselves and with NWP models), despite the different softwares and strategies used Procedures for handling issues of data quality and homogeneity processing/delivery robustness are recommended Recommendations for the operational continuation. For commercial and political reasons there will be several (regional) analysis centres handling European GPS data in the foreseeable future, given that we want to meet requirements of spatial resolution (station separations ~50 km or less) and for redundancy the consistency of products is a prime consideration. All analysis centres shall use the same orbits and clocks (the best possible and available) and the same models (IERS conventions, effects such as earth tides) There is a need for centralized monitoring agree on a common set of European stations included by all analysis centres, at least 10 key stations Coordination between operators is important when selecting new station locations. A main concern is to understand the error characteristics. orbit and clock errors the geodetic datum mapping functions ocean loading (period of 12 hours) antenna phase centre variations with AZ, EL and time (aging) multipath very site (and time?) dependent! Analysis Centres monitoring activities shall include: Data quality, including multipath characteristics coupled to station environment 26
Timeliness and amount of data from each station Centralized monitoring activities shall coordinate: ZTD characteristics (mean, variability, random walk?) and station coordinates (consistent time series) In particular the key sites are monitored in order to assess the level of agreement in ZTD and coordinates Assimilation Centres monitoring activities shall include: Station performance (observation-model first guess) All monitoring results should be made publicly available normally via a central web site, urgent matters are handled separately Within EUMETNET the EUMETNET GPS Water Vapour Programme (E-GVAP) has been started (April 2005). It is currently joined by 11 countries. Its purpose is to transform the current ground-based NRT GPS delay observing system into a system for operational meteorological use. This will be done in liaison with the geodetic community. It will include a data monitoring and feedback facility. The above recommendations will be considered by E-GVAP. Further information about E-GVAP can be found at http://egvap.dmi.dk. Also in a number of non E-GVAP countries processing in NRT of ground-based GPS delays is being (or has been) improved, and collaboration established with between the geodetic and meteorological sides. The data from those processing centres will also be monitored by the E-GVAP monitoring facility provided the NRT GPS data are made available to E-GVAP. A main issue to be solved is assuring access to GPS data for NRT processing into delays for countries in which NRT GPS delays are currently sparse. For commercial and political reasons this will often have to be solved on the national level. In conclusion, we know how to make a NRT GPS delay observing system of the quality needed for operational meteorology, and the work has begun. Meteorological institutes not yet involved in this work are recommended to start work in this field. 6.3.2. Assimilation of GPS data, impact studies A second main goal in TOUGH was to determine whether NRT GPS ZTD data are in fact beneficial to NWP. Prior to TOUGH there were indications of that, but no clear conclusions. The impact studies have been carried out by DMI, INM, LAQ, and METO. The results are reported in the TOUGH deliverables D42 to D49 27
Figure 6.3.2.1. 24 h precipitation. Observed, left. With standard amount of GPS data, middle. With additional Spanish GPS data, right. From partner INM. Figure 6.3.2.2. Verification scores for 24 h precipitation scores. Most important scores are equitable threat score (upper middle) or true skill score (upper right). RE = control experiment, RH9 = experiment with 2mRH, GP9=experiment with GPS data, RG1 = experiment with both 2mRH and GPS data. From partner INM. 28
Fig 6.3.2.3. GPS sites providing data for a sequence of the DMI impact experiments. From partner DMI. Fig 6.2.3.4. Percentage of assimilations with convergence problems when assimilating GPS data at DMI. In the other runs there were no convergence problems. 29
Fig 6.3.2.5. Example of verification of NWP forecasts against meteorological observations. Period is Jan. 2005. GGG=with GPS, 0.45deg. resolution., GGN=same without GPS. DDG=with GPS, 0.15deg. resolution, DDN=same without GPS. Notice improvements in bias and RMS of heights and degradation of 850hPa temperature. From partner DMI. 30
Fig 6.3.2.6. Example of subjective verification. Left is 12h observed precipitation, middle forecast precipitation in control run, right forecast precipitation when GPS data are included. From partner DMI. At the final TOUGH meeting an overall comparison of the impact studies was done. The following tables provide in condensed format an overview of the results. 31
Table 6.3.2.1. Summary of setup of assimilations studies done by partners METO, LAQ, INM, DMI. 32
Table 6.3.2.1 continued. Summary of setup of assimilations studies done by partners METO, LAQ, INM, DMI. 33
Table 6.3.2.2. Summary of results of long term impact studies and case studies done by partners METO, LAQ, INM, and DMI. 34
From the assimilation studies are final conclusions are: In the TOUGH project 4 partners have carried out extensive impact studies, covering all seasons. The studies include a number of more detailed case studies of strong precipitation events. The NWP models utilised cover a wide range of model types and data assimilation system types. Notice that in the studies the NRT GPS data have been used in addition to all the data types normally used in an operational NWP setup, not in isolation. Under these circumstances it is well known that a large impact from a single extra observing system is not likely. It has been found that ground-based NRT GPS data in general has a neutral to positive impact on NWP weather forecasts. We are confident that the GPS ZTD data when treated properly, are beneficial to the NWP models. It has further been demonstrated that production and assimilation of the GPS ZTD data in NRT on a European scale is possible. Based on that we recommend that European NWP centres prepare themselves for the use of ground-based NRT GPS data in their operations. However, it has also been found that the NRT GPS ZTD observing systems, as well as the NWP pre-processing and data assimilation systems, are not yet at a stage where they can be considered mature for large scale operational use. On the observing system side there are intermittent problems with the consistency of the NRT GPS products between different producers. In part this is because during part of the TOUGH project there was focus on timeliness issues. Now they can be demonstrably met focus on procedures for handling issues of data quality and homogeneity in processing and delivery are recommended. On the NWP side there are difficulties in dealing efficiently with inhomogeneous data quality during the pre-processing and assimilation of the GPS data. Screening and data assimilation functions less well when some of the statistical assumptions behind do not hold. Standard assumptions are that all observation data of a given type, (e.g. NRT GPS ZTDs) can be described with the same statistical error characteristics (e.g., observation error of a given size, Gaussian error distribution centred on true state. It is straightforward to operate with observation errors individual to each site, but not to relax the assumptions about the observation error distribution.) A separate, but related, issue is that the NWP data assimilation systems used here, as their control variable for humidity all use variables for which the statistical assumptions (Gaussian error distribution) about the errors of the control variables used in the data assimilation are not valid. No clear conclusion has been found regarding the necessity of bias correction of the GPS ZTD data, and the timescale over which to estimate the biases. It may vary from region to region and NWP model setup to setup. It is clear that part of the bias is due to the NWP models, and that it varies regionally and with time. 35
It has been found (again) that verification of precipitation is difficult. Different types of verification (e.g. objective versus subjective, and raw rain gauge measurements versus gridded precipitation observations) can result in different conclusions. Much attention was paid to this in TOUGH. Most institutes prefer to assimilate GPS ZTD, which is well suited to their assimilation systems. However, at LAQ the best results were obtained via assimilation of GPS-PW, with GPS-PW being derived currently via NWP model pressure information. Measurements of pressure, temperature at the GPS sites will enable an NWP model independent conversion from GPS ZTD to GPS PW, which might be particular use full in regions with complex terrain. Similarly use of pressure measurements can improve on the assumptions made in the processing of GPS data to ZTD. Based on the findings in TOUGH all NWP partners contributed to the following recommendations for use of NRT GPS observations in operational NWP: NWP centres should begin prepare themselves for use of NRT GPS in their operational, regional NWP forecasts. This includes: Start downloading NRT GPS delays to their databases and monitoring of it. Upgrade their data assimilation system so that it can assimilate GPS ZTD data or IWV data. Start producing the statistics of the observed NRT GPS - NWP first guess ZTD (or IWV) offsets on a station by station and producer by producer basis. This will: o Help determine whether bias correction is required for model and region of interest o Help deciding how to set up the pre-processing o Help decide which NRT GPS data to select from sites processed by more than one processing centre o Help decide whether the distribution of the offsets for some sites is so non Gaussian that the data should not be assimilated. o Help decide on the various limits to use in the variational quality control during data assimilation. Make impact studies with/without NRT GPS delays to determine optimal setup. Move toward use of data assimilation systems that can utilise the high time resolution of NRT GPS data (such as 4DVar). The NWP centres are encouraged to change their data assimilation control variable for humidity to a property with proper statistical errors (Gaussian distribution). For the NWP models utilised here that was not the case. This is not a problem specific to use of GPS data, but for GPS data it is of greater importance than for many other observation, as the ZTD (or IWV) is an integral measure, not an in situ measure, requiring significant skill of the data assimilation system to distribute properly the humidity deduced from a ZTD. Consider instalment of pressure and temperature sensors at GPS sites (or vice versa), in particular in regions with complex terrain. As mentioned several European meteorological institutes are preparing themselves for NRT GPS as an observing system. It is less clear whether a corresponding amount of 36
effort will be put on improving to the operational level the ability of the NWP preprocessing and data assimilation systems to utilise in an optimal way the NRT GPS ZTDs emerging from this new type of observations. In both TOUGH and COST716 is was clear that progress in this area came slower and at a higher cost than on the observing system side. Much of the progress was made as a result of external funding - which is significantly less post-tough. It is necessary that the NWP institutes themselves include also this type of work in their plans. Contacts between institutes to discuss results and further progress will be beneficial. The EUMETSAT GRAS SAF, as part of its activities under the Continuous Development and Operations Phase (CDOP -- to start in March 2007) is proposing to develop and maintain a formal deliverable software package to assist NWP centres to assimilate ZTD data in their operational models. This is to encourage more NWP centres to exploit this new data type without them having to develop their own data-specific software from scratch. Now-casting Now-casting was not studied in TOUGH. Never-the-less we find it important to include in the recommendations. The time resolution of NRT GPS data is high. Currently 2-4 observations/hour with an update every hour, with a potential for faster updates in the future. From the observations, maps showing the evolution of the distribution of IWV can be made and used as a guidance for forecasters. This is a field which deserves more focus. Maps of IWV are now being produced in E-GVAP, visible via http://egvap.dmi.dk/ 6.3.3. Estimation and assimilation of GPS slant delays A third goal in TOUGH was to prepare for use of GPS slant total delays (STDs) in NWP. This includes methods for both estimation of observed GPS STDs, for determination of their NWP counterparts and for assimilation into NWP models. At TUD software has been made and used to estimate GPS STDs (D33, D34, D35). The network for which this was done is shown in figure 6.3.1. A particular aspect of this is that in determination of GPS STD the asymmetric effects of the surroundings and the GPS setup itself become more important than for GPS ZTD estimation, and must be corrected for. Figure 6.3.2 is an example of this. The multipath map shows systematic deviations from symmetry in observed residuals over a month long period. 37
Figure 6.3.3.1. GPS sites used for GPS STD estimation. (TUD, D33) Figure 6.3.3.2. Example of multipath map (for station APEL). (TUD, D33) KNMI (D37) and FMI (D38 and 39)have both made STD observation operators, which given a NWP field determines the NWP estimate of STD at the location of a specific GPS site. Further they have determined error estimates of the GPS STDs. The observation operator for GPS STD have been implemented into the HIRLAM model, including a model for the observation errors. Using this KNMI (D40) have made a sensitivity study of the likely impact of GPS STDs based on synthetic GPS ZTDs and DMI (D41) have made an impact study based on real data from the dataset provided by TUD. The two studies reveal first of all that a properly working setup for assimilation of GPS STDs is now available in the NWP model HIRLAM. An example of the impact on the data analysis is shown in figure 6.3.3. The placement and size of the increments indicates the software is working properly. 38
Figure 6.3.3.3. Difference in relative humidity at 850 hpa in analysis made with and without GPS STD data. HIRLAM model at 0.15 degrees. (DMI, D41). The impact of GPS STDs on forecasts was found to be small in both experiments. DMI found a positive impact in subjective verification of precipitation, like in their studies with GPS ZTDs, but in the much smaller region, corresponding to the small region of the experiment. Notice, however, that the resolution of the experiments was not terribly high. As the GPS STDs provide a measure of the atmospheric spatial variability on local scales, higher resolution models will be needed to gain an significant advantage from use of slants relative to GPS ZTDs. The results found in TOUGH are very promising. GPS slant total delays GPS slant total delays include information about the local variability of the water vapour field. More so when the Galileo satellite system is launched. In TOUGH methods were developed for estimation and assimilation of NRT GPS slant total delays. The results are promising. This is an area of research which deserves further support on the national and European level. 6.3.4. Error modelling and GPS system research. The fourth main goal in TOUGH was to improve the understanding of GPS system and processing errors, and subsequently try to correct for them. This work was closely related to the work on NRT GPS ZTD estimation and the NWP data assimilation experiments. Spatial error correlations. The spatial correlations of the errors of GPS ZTDs have been estimated and modelled by several groups (Chalmers and SMHI D19 and FMI D18) The estimation is complicated, as the offsets of the GPS ZTD is always calculated against a property which does also have correlated errors. Figure 6.3.5.1 gives an example. Notice that the GPS ZTD contains spatially correlated errors on both small and large scales. 39
Fig 6.3.4.1. Horizontal correlations of the ZTD innovations for a one year period. Dashes = model fit to data. Dash dot = NWP model correlations. Red line = model fit to GPS ZTD correlations. (Chalmers, D19). A model enabling correction for the spatial error correlations was implemented in the HIRLAM model by SMHI and tested in a short impact experiment. It was found that the implemented correction software works. Whether the impact on the data analysis is positive is difficult to say it will require an idealised experiment to test it, as the GPS observations in ordinary assimilation experiments represent a minute fraction of all observations and have a correspondingly small weight Temporal error correlations. The temporal error correlation of GPS ZTDs was estimated and modelled by Chalmers (D22). Figure 6.3.4.2 gives an example. Fig 6.3.4.2. Temporal correlation of the differences between GPS and WVR wet delays at Onsala for a one year period. The e-folding time is about 1 day. (Chalmers, D22.) In a month long impact 4DVar data assimilation impact study DMI (D23) used a simple method to correct for the temporal error correlation, the existence of which means the GPS ZTD observations should be given less weight in the assimilation than otherwise. Compared to not correcting for the temporal error correlation (D24) the impact was positive for precipitation and neutral otherwise. 40