Satellite rainfall estimates : new perspectives for meteorology and climate from the EURAINSAT project

Satellite meteorology is facing a crucial period of its history since recent missions have revealed instrumental for quantitative rainfall measurements from space and newly conceived missions are at hand. International partnership is rapidly developing and research projects keep the community focused on rapidly developing research and operational issues. A perspective is given through the structure of EURAINSAT, a project of the 5th Framework Programme of the European Commission. Its key objective is the development of algorithms for rapidly-updated satellite rainfall estimations at the geostationary scale. The project is fostering international research on satellite rainfall estimations building a bridge between Europe and the U.S. for present and future missions. Mailing address: Dr. Vincenzo Levizzani, Istituto di Scienze dell’Atmosfera e del Clima (ISAC), CNR, Via Gobetti 101, 40129 Bologna, Italy; e-mail: v.levizzani@ isac. cnr.it


Introduction
The latest generation sensors on board the Geostationary Operational Environmental Satellites (GOES) (Menzel and Purdom, 1994) and the upcoming METEOSAT Second Generation (MSG) Spinning Enhanced Visible and Infrared Imager (SEVIRI) (Schmetz et al., 1998) significantly enhance the ability of sensing cloud microstructure and precipitation forming processes (Levizzani et al., 2000(Levizzani et al., , 2001) ) from a geostationary platform.A potential exists for improved instantaneous rainfall measurements from space by combining infrared (IR) and visible (VIS) with passive microwave (MW) observations with a more global perspective.
IR and VIS satellite rainfall estimates have long since been available and suffered from the diffi culty in associating cloud top features to precipitation at ground level.IR methods were used for climate purposes or combined with radar measurements for nowcasting (recent examples are Vicente et al., 1998;Porcù et al., 1999;Amorati et al., 2000) and multispectral approaches start to become operational (e.g., Ba and Gruber, 2001).
Physically-based passive MW methods were developed mainly using data from the Special Sensor Microwave/Imager (SSM/I) and are based on several different physical principles (see for example Wilheit et al., 1994;Smith et al., 1998).Limitations of MW algorithms include the relatively large footprint and the low earth orbits not suitable for most of the operational strategies.
The combined use of MW and IR data for rainfall estimations was already recognized some time ago (Vicente and Anderson, 1993).Adler et al. (1993) used SSM/I data for monthly average rainfall estimations over wide areas and global products such as those of the Global Precipitation Climatology Project (GPCP) were conceived (for recent advances see Huffman et al., 2001).However, the need for hourly and instantaneous combined estimations was clearly recognized (e.g., Vicente, 1994;Vicente and Anderson, 1994;Levizzani et al., 1996).Several methods exist (Turk et al., 1999;Sorooshian et al., 2000;Todd et al., 2001) that make use of IR and MW at various degrees of complexity and targeting different rainfall regimes.Some of these are running operationally, granting that their validation requires additional work in the years to come.In particular, global rapidupdate estimates with near real-time adjustment of the thermal IR co-localized with MW-based rainrates are operationally very promising (Turk et al., 1999).
The algorithms for the Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) require a special mention given the novelty and potential of the active instruments for future missions: an example is the TRMM algorithm 2A-25 (Iguchi et al., 2000).Combined MW and rain radar algorithms are relatively new and were developed for the TRMM Microwave Imager (TMI) and the PR (e.g., TRMM algorithm 2B-31, Haddad et al., 1997).
Finally, rainfall and humidity assimilation, and microphysical parameterizations for Numerical Weather Prediction (NWP) models, above all Limited Area Models (LAMs), and General Circulation Models (GCMs) open up the road to very effective operational meteorological applications that incorporate the verifi cation of model output (Turk et al., 1997).

EURAINSAT: the project
EURAINSAT is a project partially funded by the European Commission with the aim of developing new satellite rainfall estimation methods at the geostationary scale for an operational use in short and very short range weather monitoring and forecasting.It will exploit the new channels in the VIS, NIR and IR of the MSG SEVIRI (see fi g. 1) that will be launched in mid 2002.The project started in January 2001 and will last until December 2003.
The SEVIRI channels in the VIS and IR portion of the spectrum will gain better insights into the microphysical and dynamic structure of precipitating clouds allowing for a more precise identifi cation of precipitation levels.The method(s) will work as follows: 1) microphysical characterization of precipitating clouds with VIS/IR sensors; 2) creation of microphysical and radiative databases on cloud systems using cloud model outputs and aircraft penetrations; 3) tuning of MW algorithms on the different cloud systems (convective, stratiform, ...); 4) combination of data from the different algorithms and application to a rapid update cycle that makes use of the different sensors at the geostationary scale.
The consortium has two objectives in mind: 1) contribute to improving the knowledge of clouds and precipitation formation processes using meteorological satellite sensors, and 2) make available new precipitation products for weather analysis and forecasting.SEVIRI will in fact provide better multispectral measurements for the identifi cation of the physical processes of cloud formation and evolution.The 15 min image repetition time is also more compatible with the time responses of cloud systems.
The following key geographic areas and major meteorological events are considered: -Flood-producing episodes (e.g., Northwestern and Southern Italy).
-Several cases involving the presence/ absence of ice, polluted air masses and maritime conditions.
-Sustained light rain and «insignifi cant» rain cases (very diffi cult to detect from satellite) in U.K. and Northern Europe.
-Tropical and sub-tropical cases over Africa, where the Niger catchment was selected given the relatively regular ground raingage network.
The project has gathered together a substantial part of the satellite rainfall community.The team actively participates into the development of scenarios and concept for the future Global Precipitation Measurement (GPM) Mission and the International Precipitation Working Group of the Coordination Group for Meteorological Satellites (CGMS).More information on EURAINSAT and its findings can be gained at the web site http://www.isac.cnr.it/~eurainsat.

Microphysical characterization of cloud processes
A method was developed by Rosenfeld and Lensky (1998) to infer precipitation-forming processes in clouds based on multispectral satellite data.The method was originally based on the Advanced Very High Resolution Radiometer (AVHRR) imagery on polar orbiting satellites (Lensky and Rosenfeld, 1997).The forthcoming MSG SEVIRI is expected to enhance the capabilities of extracting cloud physical properties (Watts et al., 1998) more relevant for cloud genesis and evolution and not anymore limited by the insuffi cient number of passages.The effective radius (r e ) of the particles and cloud optical thickness are extracted and used for radiative transfer calculations that defi ne the cloud type and improve its characterization.
Precipitation forming processes are inferred also using data from the AVHRR, the TRMM VIS and IR Sensor (VIRS) and the MODerateresolution Imaging Spectroradiometer (MODIS) on board NASA's Terra spacecraft.In particular, microphysical and radiative parameters from satellite sensors are instrumental for defi ning the characteristics of precipitating clouds.An example of global cloud effective radius values derived from MODIS and ready to be used for EURAINSAT is given in fig. 2.
«Microphysically-maritime» clouds grow in relatively clean air with small Cloud Condensation Nuclei (CCN) and low droplet concentrations, which produce very effi cient coalescence and warm rain processes.«Continental» clouds normally grow, on the contrary, in polluted air masses having large CCN and high droplet concentrations, i.e. the coalescence is relatively ineffi cient.The better knowledge of cloud microphysical structure and precipitation forming processes will facilitate the development of a new generation of improved passive MW rainfall algorithms.
One more promising line of action is the potential use of lightning detection for discriminating between convective and stratiform regimes while estimating precipitation.Data from ground-based lightning detection networks and satellite sensors like the TRMM Lightning Imaging Sensor (LIS) are applied to IR rainfall estimations (Grecu et al., 2000) and show considerable potential for rapid update applications.
Above all, lightning detection represents a fast-response fundamental parameter for discriminating active convection and quantitative relationships have been found between lightning discharges and other measurables of rainfall (Petersen and Rutledge, 1998).Dietrich et al. (2001) have recently shown that the use of concurrent data from TRMM PR and LIS give unique information about the link between electrification and convection for the discrimination of convective and stratiform regimes.The authors hint at applications for rainfall retrieval using multispectral and MW techniques.The work is based upon the fi ndings of Solomon and Baker (1998) who examined thunderstorm development and lightning fl ash rates in tropical maritime, subtropical continental, and midlatitude continental storms.The dependence of lightning occurrence and fl ash rate on cloud condensation nucleus concentration, primary and secondary glaciation mechanism, liquid water flux, and updraft velocity is of fundamental importance.

IR methods
IR-based rainfall estimation methods were the fi rst to be applied to a wide variety of scales and phenomena.With the advent of MW sensors they were more and more confi ned to large scale and climate applications.At the instantaneous time scale, methods based on thermal IR data are being integrated by ancillary information, such as data from radar, other VIS, IR and NIR channels, lightning detection, model output and other meteorological parameters.
For example, Porcù et al. (1999) have conceived a simple method to ensure a more physical relationship between cloud structure and precipitation rates as derived from the IR thresholding Negri-Adler-Wetzel (NAW) technique (Negri et al., 1984) by using SSM/I observations.A low-rainrate event with orographic forcing and small scale precipitation in Piedmont (NW Italy) on 13-16 October 2000 was considered over a 25 000 km 2 area to document the impact of the calibration on low precipitation NAW areas; nine SSM/I overpasses were available over the target basin (Porcù et al, 2000).The impact on the basin-averaged rainfall is rather low, resulting in a slight decrease of satellite overestimation for higher rainrates.The effect of calibration is more evident over the directly calibrated locations, given the very low occurrence of high precipitation areas during the event.The calibration helps for the higher rainrate peaks, while there is a marked underestimation of lower rainrates that is not affected by the calibration.The scattering of the points also increases showing an overall overestimation even for the highest rainrates.to the peak rainrates and does not reduce the underestimation for light rain, since it has no effects on the rain/no rain threshold.

MW passive and active methods
Many methods have been proposed for measuring rainfall from MW satellite sensors.Simple methods using polarization-corrected brightness temperatures (PCT) (e.g., Kidd, 1998) have been proposed together with more physical approaches that rely upon microphysical characterization by: -stratifying clouds into different microphysical types and examining how much of the variability in the bias of MW rainfall estimation is explained by the microphysical characterization; -developing a library of passive MW signatures from different cloud types, and -using a microphysical cloud classifi cation for improving cloud radiative transfer modeling based on statistical multivariate generators of cloud genera.Figure 3 shows an example of MW rainfall retrieval over the Mediterranean and Western Europe using the algorithm by Turk et al. (1999) and data from SSM/I, TMI and the Advanced Microwave Sounding Unit-B (AMSU-B).More details can be found in Berg et al. (1998).
The scheme of Mugnai et al. (1993) and Smith et al. (1992) is a good example of such methods, especially in the very complex environment of severe storm microphysics.Cloud modeling and MW radiative transfer has been recently applied to stratiform rainfall by Bauer et al. (2000).Panegrossi et al. (1998) have shown the importance of testing the physical initialization and the consistency between model and measurement manifolds.Research trends concentrate on improving the interpretation of active and passive MW measurements through better modeling of cloud processes such as the melting layer (Bauer, 2001a;Olson et al., 2001a,b).The TRMM satellite has set the path to new algorithms that still mostly work over oceans (e.g., Bauer, 2001b), but new developments are at hand over land (Grecu and Anagnostou, 2001).Polarization and texture information from passive radiometers together with PR data complete the scenario of new methods (Olson et al., 2001c).Ferreira et al. (2001) have recently contributed to the improvement of the TRMM 2A-25 algorithm (Iguchi et al., 2000) proposing the use of G drop size distributions and R-k relations instead of R-Z (where R is rainrate, k the attenuation coeffi cient, and Z the refl ectivity).
The importance of separating convective and stratiform precipitation and stratiform and transition regions in the a priori cloud model database of MW algorithms is demonstrated by Kummerow et al. (2001).The authors document the latest improvements to the Goddard PROFiling algorithm (GPROF) as applied to the TRMM data.The new algorithm also uses the emission and scattering indexes instead of individual brightness temperatures.These improvements, together with the elimination of some classifi cation ambiguities over land, are general and apply to other algorithms as well.
Fundamental ancillary data are finally provided by active and passive MW radiometry for Cloud Liquid Water (LWC) profi ling: measurements from ground based MW radiometers combine with Z profi les from cloud radars and cloud model statistics to lower the errors in LWC measurements by as much as 10-20 % (Löhnert et al., 2001).

Combined multispectral and MW methods
Cloud microphysical information, when combined with MW measurements, can lead to improvements in satellite-based rainfall measurements, especially from clouds in the extra tropics and over land (e.g., Bauer et al., 1998).EURAINSAT concentrates on exploiting SEVIRI data in the VIS, NIR and Water Vapor (WV) for cloud characterization and screening within a rapid cycle of rainfall estimation based on SSM/I, TMI and geostationary IR data.Data from MODIS simulate MSG SEVIRI data during the pre-launch phase.Moreover, the project shares in the cloud-related work from the MODIS team (King et al., 1997).
There are two main research lines: -Develop new MSG-MW rainfall algorithms incorporating the observed cloud microstructure and precipitation forming processes.State of the art cloud and mesoscale models (Khain et al., 2000;Tripoli et al., 2001) and radiative transfer models will be instrumental to detailed cloud and rainfall type discrimination.The need to use such frontier cloud model was recently shown by Khain et al. (2001) who demonstrated the weakness of existing cloud parameterizations while trying to give reason of highly supercooled water in convective clouds (1.8 g m -3 at -37.5 C) as found by Rosenfeld and Woodley (2000).
-Introduce such methods into rapid update rainfall cycles for near real time rainfall estimations over oceans and land with the widest possible area coverage.Mid-latitude Europe, the Mediterranean basin, North Africa, the Middle East and equatorial and tropical African regions are the main targets for operational and climatological applications.Applications to the Mediterranean have been reported by Meneguzzo et al. (1998).

Applications
Applications embrace, among others, water availability, global change studies, nowcasting, hydrogeological disaster management, agriculture, famine reduction, and monitoring of remote areas.

Rainfall assimilation for NWP
Data assimilation procedures that improve cloud and humidity characterization in current analysis schemes for LAMs are at hand.Most important are the sensitivity to orography and modeling of moist processes.The model BOLAM (Buzzi et al., 1998) is used in the project to conduct rainfall assimilation experiments that quantify the impact of satellite data onto the forecasting chain.The nudging technique of Falcovich et al. (2000) is adopted.Motivations for using nudging techniques are that nudging to over-saturation is more gradual, and the reference profi le is useful in tropical areas.The model is now applied to autumn rainfall episodes over the Mediterranean.
The model RAMS (Pielke et al., 1992) is also used and runs operationally in Tuscany with two grids, that are two-way nested and with the highest horizontal spatial resolution of 4 km around Tuscany and the Arno river basin.The complex cloud microphysics scheme is fully activated (Walko et al., 1995), while the Kuotype convection parameterization (Molinari, 1985) is activated only over the 20 km outer grid and explicit (resolved) convection is allowed over the inner grid.A higher spatial horizontal resolution of 2 km will be reached to ensure fully consistent explicit representation of the convection.Quantitative Precipitation Forecasts (QPFs) are produced hourly over the inner grid.Diabatic initialization and rainfall assimilation will be conducted in the operational chain before the end of the project.

Rainfall and climate change
The importance of multispectral cloud characterization methods has recently been demonstrated by observing and documenting the inhibiting effects of forest fi re (Rosenfeld, 1999), urban pollution (Rosenfeld, 2000) and desert dust aerosols (Rosenfeld et al., 2001) on precipitation formation processes.It is very likely that rainfall processes have been substantially underestimated against, for example, greenhouse gases in evaluating possible causes of climate changes.A more quantitative assessment is, however, necessary.
In particular, Ramanathan et al. (2001) argue that manmade aerosols produce brighter clouds with reduced precipitation efficiency and rainfall suppression.This can lead to a weaker hydrological cycle, which connects directly to water availability and quality, a major environmental issue of the 21st century.
Satellite rainfall data are used also to assess the impact of particular weather systems on the geographical, seasonal and interannual distribution of total rainfall.Rodgers et al. (2000) have concentrated on the impact of tropical cyclones on the North Pacifi c climatological rainfall.The same authors have repeated the study for the North Atlantic (Rodgers et al., 2001).Such studies are crucial for the quantifi cation of climatic effects and their relationships with indicators such as the El Niño Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO).Databases of global products such as those of the GPCP (Huffman et al., 2001) will be very useful for supporting the defi nition of new local indexes that better describe regional variations, observed over smaller basins like the Mediterranean.

Attenuation of satellite communications due to rainfall
Over the last few years, there has been an increasing demand of large-bandwidth information services coupled with high availability and low-fade margin communication systems (Watson and Hu, 1994).This scenario has prompted exploring channel frequencies at Ka band and above, and developing sophisticated countermeasure techniques to mitigate outage periods (Jones and Watson, 1993).The implementation of most advanced adaptive countermeasure techniques is related to the possibility of monitoring in quasi real-time the beacon attenuation in a given region and period.Spatial and frequency diversity methods, power link control, data rate and error correction on the downlink/uplink can be effectively adopted only if the propagation conditions are known in real time.
Exploitation of remote sensors and their products represents a natural way to optimize the performances of a satellite communication system with low power margins, specifi cally while applying fade mitigation techniques.MW signatures of precipitation, as given by a spaceborne multi-frequency radiometer, have been shown to be the base for estimating the path attenuation in K-band satellite communications (Marzano et al., 2000).A general approach should attempt to estimate rainfall intensity and attenuation by polar-orbiting MW radiometers and temporally track the rain system by means of geostationary IR radiometers.A statistical approach can be used to derive a prediction model of path attenuation from MW brightness temperature and surface rainrate (Crone et al., 1996).

Fig. 3 .
Fig. 3. 3-h average rainrate using a combined SSM/I, TMI and AMSU-B algorithm (29 September, 2001).Note the storm system stretching from the Tyrrhenian Sea to Tunisia.The black triangle indicates lack of data coverage after the mosaic of the satellite overpasses.Image courtesy of J.F. Turk (http://www.nrlmry.navy.mil/sat_products.html).