International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products Satellite Precipitation Estimation over the Tibetan Plateau and Perspectives for new Satellite Missions Federico Porcù1*, Uarda Gjoka1,2, Stefano Dietrich3, Paolo Sanò3, Daniele Casella3, Alberto Mugnai3 1. Department of Physics and Earth Sciences – University of Ferrara, via Saragt 1, I-44122 Ferrara, Italy ([email protected]) 2. Department of Information Technology – Aleksander Moisiu University, Durres, Albania 3. ISAC-CNR, via Fosso del Cavaliere 100, 00185 Rome, Italy ABSTRACT Precipitation estimation is a challenge for atmospheric remote sensing: a number of satellite sensors, with different sensitivity to precipitation, are commonly used to feed estimation techniques. Precipitation signature in the radiation measured from an orbiting sensor varies across the wavelength: is generally low in the visible-infrared and higher in the microwave. On the other hand, due to diffraction reasons, microwave sensors are only operated on low orbit satellites, resulting in high revisiting time and large footprint at the ground. To overcome these limitations, multisensor approaches are pursued, combining microwave and VIS-IR measures in order to mutually mitigate disadvantages and enhance capabilities. In the frame of CEOP-AEGIS an effort was undertaken to perform precipitation estimation on the Tibetan Plateau, where the knowledge of precipitation systems is very low and the ground-based observation system is poor. A summary of the results achieved in the Project is presented, with emphasis on satellite precipitation estimation, showing advantages and drawbacks of the considered techniques. A new Artificial Neural Network multisensor technique has been implemented on the Plateau, by using infrared METEOSAT-7 channels, ground radar rainrate measurements and microwave satellite estimates. Comparison with ground data and global scale precipitation products are considered and the role of orography and diurnal cycle on the precipitation intensity and spatial distribution is evaluated. The results are also considered with a look in the next future, when other sensors, dedicated to precipitation measurement, such as the Dual-frequency Precipitation Radar, on board the GPM Core Observatory, will be available. KEYWORD: PRECIPITATION, REMOTE SENSING, RADAR. INTRODUCTION The quantitative estimation of spatial distribution of precipitation in the Tibetan Plateau (TP) is a key aspects for the understanding of water cycle processes and the estimation of water resources. Over TP observation relies on a sparse raingauge network and few ground based weather radar with coverage strongly limited by beam blocking due to orography. The satellite point of view is thus a valuable option for precipitation monitoring over TP, and several studies have been carried out to highlight precipitation characteristics [1]. The need of satellite observations was originally addressed by Ueno [2], who implemented different visible-infrared (VIS-IR) and passive microwave (PMW) International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products algorithms to estimate precipitation for the monsoon season, adapting different algorithm to the TP climatology, highlighting the need estimation algorithms improvement and adaption. To assess the relationship between satellite data an precipitation at the ground in experimental way, two important issues must be addressed: the understanding of the precipitation structure over TP to emphasize the algorithm concept, and the measurement of accurate aerial precipitation amounts to calibrate and validate the algorithms. Both these issues have been addressed in 1998, during the GAME/Tibet project, when the study of precipitation by a Doppler Radar and a dense raingauge network has been carried out, taking also advantage of the observation of the Tropical Rainfall Measuring Mission (TRMM) and Chinese GEO satellite FY/2 [3]. A new algorithm for precipitation over land by deriving the optical thickness from the Brightness Temperature (BT) of the TRMM Microwave Imager (TMI) was proposed [4]. Based on a microwave radiative transfer equation, two indices, Index of Soil Wetness and Polarization Index, are used. Daily rain amounts have also been validated over the Dasa Basin, about 300 km2 wide, with 11 raingauges. Three major shortcomings for quantitative precipitation estimation were outlined: 1) the assumption of no emission from atmosphere and rain at 85 GHz was too simplistic; 2) the temporal sampling of TMI is inadequate to resolve the rapidly varying precipitation patterns; 3) the relationship between precipitation layer optical thickness and actual rainrate depends upon the unknown hydrometeor type. Similar approach was pursued in [5] using BT to estimate precipitation rate over TP. They used 4 raingauges data to assess and calibrate the rainrate at the ground with two parameters computed as linear combinations of BT: Scattering Index and Polarization Corrected Temperature. An intercomparison study shows that [5], tuned exclusively for TP, is advantageous over a global techniques. More recently, Yin et al. [6] used different MW and MW-IR based algorithm to infer monthly precipitation over TP. A total of 50 parameters describing orography (height, slope, aspect, orientation, among others) computed over a 1x1 km grid are used to “correct” the outputs of the different techniques, through a principal component analysis. This analysis is strongly dependent on the cumulation time used for precipitation evaluation and can be only qualitatively referred to shorter time intervals, as requested for hydrological purposes. SATELLTE PRECIPITATION ESTIMATION TECHNIQUES Clouds and precipitation structures interact with electromagnetic radiation selectively: different hydrometeors scatter and/or absorb radiation over different frequency ranges, in relation to their phase, shape, size, density and habit. At VIS-IR wavelength clouds are mostly opaque bodies: the radiation interacting with precipitation layers usually do not reach the satellite sensor. VIS-IR radiation carries information on the cloud top structure, and an estimate of the rainrate below can only be indirectly inferred, by means of statistical pattern recognition techniques. In the MW spectrum, instead, cloud droplets and crystals are transparent to low frequencies (around 10 GHz), so that the emission of raindrops can be measured from satellite. Al higher frequencies (e.g. 87 GHz), scattering from cloud top ice particles becomes relevant: multispectral techniques are able to infer the vertical distribution of hydrometeors and compute precipitation at the ground, based on physical properties of the particle-radiation interactions. Since PMW sensors, due to diffraction problems, have useful sized footprint only if on board Low Earth Orbit (LEO) satellites, more advanced techniques make use of both VIS-IR (ensuring continuous International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products coverage) and PMW (providing direct information of cloud/precipitation structure) data, to provide blended precipitation products [1,7]. A two step strategy has been designed to retrieve precipitation over the TP: 1) two microwave-based, physically-founded algorithms provide high quality precipitation fields; 2) different techniques are used to merge the MW high quality maps and high resolution VIS-IR GEO observations. Physically based, microwave algorithms The climatology of the precipitation over TP is clearly divided in a monsoon season, occurring in warm months (JJAS), where diurnal convection takes place and night time stratified precipitation occur [2,3,8], and a dry season for the rest of the year characterized by sparse snowfalls. The data from the 94-GHz nadir-looking Cloud Profiling Radar (CPR) on board the LEO CloudSat spacecraft [9] are used to retrieve snowfall rates over TP during dry months. The algorithm developed and applied to TP hydrological basin is derived by Kulie and Bennartz [10]. The data considered is the vertical reflectivity profiles (2B-GEOPROF) and makes use also of ancillary data from the ECMWF model. The algorithm selects CPR vertical profiles with a vertically continuous cloud path (i.e. reflectivity exceeding the -15 dBz threshold) for at least 1 km. A lower limit of 1.3 km above the ground is set to avoid clutter and a temperature threshold of 273 K is used to screen out wet snow cases. Finally, a reflectivity snow rate-reflectivity (S-R) relationship of the type Z=aSb, is applied to the lowest cloudy bin of the considered profile. The a and b parameters in the Z-S relation should be experimentally computed, by on site observation of ice particle shape, not available in this project, then we used values obtained by scattering simulations on bullet rosettes, as a typical, averaged, hydrometeor shape. For the monsoon season we considered an algorithm based on PWM, which is expected to work reasonably well over land: the Cloud Radiation Database (CRD) [11]. This algorithm is a rainfall retrieval scheme that works on conical scanner data: for this project data from SSMIS are used. The algorithm is based on a cloud radiation database constructed as follows [12]. A cloud vertical profile data set is assembled by means of cloud resolving model outputs (the Non-hydrostatic Modeling System of the University of Wisconsin is used to this end), then a radiative transfer algorithm is applied to simulate the radiances upwelling from the modeled cloud profiles. For the algorithm set-up over TP and used in this work, the global database was completed by three model runs over cases studies of convective precipitation over TP. When a set of satellite radiances ad different frequency is measured from the satellite sensor, the database is searched for the cloud profile whose simulated radiances better match the observed ones, and the precipitation rate related to the selected profile is assigned at the ground. This selection of profiles is carried out using the Bayesian distance among the measured BT vectors, and the corresponding simulated vectors of the previously selected set of profiles of the CDR database. This algorithm is currently applied in different regions (Europe and U.S.) with encouraging results. PMW-VIS-IR blended techniques Several blending techniques are proposed in the literature, based on different principles, algorithms and data [13]. In this work the blended product are obtained at spatial and temporal resolution of METEOSAT-7 IR data: 5x5 km (at nadir) and 30 minutes, respectively. METEOSAT-7 is on a GEO orbit at 57 degrees of latitude. We started attempting three different approaches: Passive Microwave combined Global Convective Diagnostic (PM-GCD) [14], Calibrated Negri Adler Wetzel (C-NAW) [15], and Artificial Neural International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products Network (TANN) [16]. The first approach, based on the dependence of convective rainrate from the temperature difference between the infrared window (λ = 11.0 μm) and the water vapor (WV, λ = 6.5 μm) channels. The correlation between rainrate and temperature difference is usually good if applied over Europe and at global scale [14], but over TP was not satisfactory, probably due to variable water vapor tropospheric content and to the presence of nighttime stratiform rain, not sensed at IR wavelength. For this reason, the PM-GCD was abandoned, after first sensitivity studies. The C-NAW [15] first uses IR GEO observations to delineate high precipitation areas by means of the coldest 10% fraction of the cloud-top temperature and then MW observations from LEO satellites to estimate mean rainfall values within such areas. This technique mostly applies to convective clouds, but is also successfully used for stratified of mixed cloud types, if the precipitation coefficient are carefully calibrated. Over the TP the calibration was performed by using the CPR snowfall technique, for the cold months, and the output of the CRD algorithm for the monsoon seasons. The use of ANN to provide a blended product is open to a number of options, as seen in the literature [16]. During training phase, the ANN learn to reproduce, given a set of input (i.e. satellite data), a known output (i.e. precipitation at the ground), setting up a transfer function that weights input to obtain corresponding output. We used as input data the BT from METEOSAT-7 channels in the IR and WV, and their local variability features (i.e. local average and local standard deviation over a 3x3 pixel neighborhood) then used two different approaches for the reference precipitation data for ANN training. Details on the ANN structure and training-testing-validation procedures can be found in [16]. For the first release of the product (hereafter referred as TP ANN calibrated with radar, TANN-R) the training set used weather radar as reference ground data. Five weather radar (two over Qinghai and three around Lhasa), operated by China Meteorological Administration (CMA), provided calibrated precipitation data for selected case studies. The ANN training was performed over these case studies for the monsoon season, and over the CPR snowfall data for the rest of the year. For the second release of the product (TANN-S) the training is performed on a supervised dataset where reference rainfall values are obtained by CDR estimates, for the monsoon season. For the dry season, the same calibration as TANN-R is used. The advantages of this approach is that the training set spans all over the season, and not only on selected case studies: a wider variety of precipitation systems is then included in the dataset. A possible drawback is that the CRD estimates is expected to perform poorer than the radar estimate. Global reference products Two well assessed global product are also considered as reference in this work: the TRMM Multi-satellite Precipitation Analysis (TMPA) 3B42 product [17], and the Climate Prediction Center MORPHed precipitation (C-MORPH) [18]. Both products, freely available, have a ground resolution of 0.25x0.25 degrees and the rainrate is averaged over three hours. The TMPA is a calibration-based sequential scheme for combining precipitation estimates from multiple sensors, as well as gauge analyses [17]. The 3B-42 estimates are produced in four stages; (1) the microwave precipitation estimates are calibrated and combined, (2) infrared precipitation estimates are created using the calibrated microwave precipitation, (3) the microwave and IR estimates are combined, and (4) rescaling to monthly data is applied, with the use of available raingauge data. The C-MORPH uses a different approach: half hourly IR data from GEO satellites are used as a mean to transport the microwave-derived precipitation features during the intervals between International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products subsequent PMW observations. Propagation vector-matrices are produced by computing spatial lag correlations on successive IR images of GEO satellite, which is used to propagate the microwave derived precipitation estimates. The shape and the intensity of the precipitation features are modified (morphed) during the time between microwave sensor scans by performing a time-weighted linear interpolation. The method is flexible to permit precipitation morphing from any satellite PMW sensor (AMSU, TMI, AMSR-E, SSM/I). Both products are validated with raingauges observation over different regions of the word, showing a general underestimation of moderate rainrates, with better performances of 3B42, which, however makes direct use of raingauges data. No validation is available over TP yet. RESULTS AND DISCUSSIONS Three years of satellite data are processed, and 30 minutes of instantaneous rainrate/snowrate over the METEOSAT-7 spatial grid are retrieved. As mentioned, the lack of a complete ground reference dataset prevent a systematic validation campaign, especially for the dry season. Even for the monsoon season the ground radar have a limited spatial coverage that prevents a statistically significant validation. Therefore, the analysis presented here is mainly an intercomparison among the different estimates. As an example, in Figure 1 three hourly rainfall maps for the 29/08/2009 at 18:00 UTC are shown. Radar network shows a precipitation structure with low precipitation in the 3 hours, and one small peak with cumulated rain around 10 mm (a). CMORPH (b) completely misses the precipitation structures, detecting only low rain accumulation but slightly misplaced with respect to radar observations. CNAW (c) detects several precipitating areas in the region, with rain amount comparable with radar observations, but also a misplacement is noted. The 3B42 (d) estimates is rather good in detecting rain areas, but the amount of rain is underestimated. TANN-R (e) and TANN-S (f) show similar shapes in the precipitation patterns, and the wet areas is the largest among the estimates. The higher resolution shows the presence of a structure of precipitation field not seen at lower resolution. TANN-S estimates higher rain amount than any other technique. A further approach in the intercomparison among precipitation product is presented in Figure 2a, where the TP basin averaged daily rainrate is plotted for year 2010. In Figure 2b the black line is the rain amount computed as the mean value of the five techniques used (C-NAW, TANN-R, TANN-S, 3B42 and CMORPH), while the red line is the corresponding standard deviation. All the techniques are sensitive to the onset and end of the monsoon season, when also show higher agreement (low standard deviation). For the first two months the precipitation is very low: well below 0.5 mm for both the global products, but 3B42 shows few peaks above 1 mm, while blended techniques have values between 1 and 2 mm on the average. A very large peak, probably due to snow at the ground is estimated by TANN-R, and other by other techniques but at markedly lower values, except CMORPH. Around day 80 CMORPH has a 10-day wide peak not sensed by other techniques, while between day 100 and 150 the techniques agree with rather constant averaged value around 1 mm, with some isolated peaks observed by 3B42. Around day 150 there is the onset of monsoon season TANN-R and TANN-S and the precipitation amount increases for all the techniques: average values are now between 3 and 5 mm, except for CMORPH, that keeps below the other curves. Sharp peaks, during 1 o 2 days, are seen during the monsoon season: sometimes they are observed by more than one technique, but in general not with the same values. Even TANN-R and TANN-S are far to be similar, with the latter generally overestimating. International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products Figure 1. Three hourly cumulated rain as seen from different techniques: (a) QPE from Tibet ground radar network; (b) CMORPH; (c) 3B42; (d) CNAW; (e) TANN-R; (f) TANN-S. Figure 2. TP-averaged daily rain amount for the year 2010 as retrieved from the five techniques (a), and mean value (black line, left scale) and standard deviation (red line, right scale) of the daily amount (b). Around day 300, after the monsoon season has finished, the amount of rain decreases with the agreement of all the techniques, except CMORPH, that in this case tends to overestimate precipitation amount. TANN and CNAW are highly correlated. International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products CONCLUSIONS AND FURTHER PERSPECTIVES The aim of this work is the implementation and applications over three years of data of an array of satellite precipitation techniques. Two physically based techniques have been implemented for snow rate, based on CPR data (for cold months) and rainrate derived from SSMI/S data (for monsoon season). These techniques, given the high revisiting time, are not suitable for direct use as precipitation monitoring tool, but are used as calibrator of multisensor techniques based on GEO IR data. Two approaches are tested to merge MW retrieval to IR data: entity based and fully statistical. The first approach (C-NAW) requires the definition of precipitating area as separated entities, and the assignment of rainrates as determined by coincident MW estimates. The full statistical approach has been pursued by implementing an ANN technique, trained on ground radar data (TANN-R) or PMW retrieval (TANN-S) for the wet season, and on Cloudsat-CPR snowfall retrieval during dry months. Finally, two global precipitation products have been considered for reference and intercomparison: the CMORPH and the TMPA product 3B42. All the techniques have been implemented for the 3 years and the results compared at different spatial and temporal scales. The analysis of daily rain amount has shown that in general CMORPH and TMPA are able to estimate rain amount larger than the ones estimated by other techniques during the monsoon season. In cold months global techniques underestimate precipitation amount and areas, resulting in a dry bias with respect to IR calibrated techniques. Case studies compared with ground radar data on convective episodes shown that global products tend to underestimate precipitation areas, while IR calibrated techniques provides reliable rainrate patterns. Unfortunately, the number of radar case studies was not large enough to allow significant validation studies, and also non data were available for cold months. Annual precipitation cumulated maps show marked differences among the techniques: IR calibrated techniques generally overestimate precipitation amount by a factor of 2 with respect of global products. Reasons for discrepancies are probably in the role the IR data have on the techniques: when the rain area delineation is based on IR, overestimation is likely to occur, while when the MW estimate is used, underestimation of low precipitation areas is expected. Further studies are needed to resolve these ambiguities, and the forthcoming full exploitation of the Global Precipitation Measurement mission will surely help on this task, and probably give a clearer picture of precipitation properties over TP. In particular, the GPM Core Observatory (GPM-CO), to be launched next year on a 65 degree sun-synchronous orbit, will host the Dual-frequency Precipitation Radar (DPR), a Ka-Ku band radar, and a high resolution, multichannel PMW rain radiometer called the GPM Microwave Imager. The GPM-CO will serve as the calibration reference system and the fundamental microphysics sensor package, especially on region with scarce ground observations, supporting an integrated satellite measuring system made up of six to ten constellation satellites, equipped with PMW sensors. ACKNOWLEDGEMENT The work described here was supported by the European Commission (Call FP7-ENV-2007-1 Grant nr. 212921) as part of the CEOP-AEGIS project (www.ceop-aegis.org) coordinated by the Université de Strasbourg. The TMPA 3B42 data used in this study were acquired as part of the NASA's Earth-Sun System Division and archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC) Distributed Active Archive Center (DAAC). International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products CMORPH products have been downloaded from the NOAA-CPC ftp site: ftp.cpc.ncep.noaa.gov/ precip/global_CMORPH/. The METEOSAT-7 MVIRI data were downloaded from the EUMETSAT Earth Observation Portal: http://www.eumetsat.int/Home/Main/DataProducts/ProductNavigator/ index.htm?l=en). The availability of radar and raingauge data was granted by prof. Liping Liu of the CMA, and the help of Wei Zhuang in data managing is acknowledged. REFERENCE [1] Gjoka U., Analysis of precipitation characteristics on the Tibetan Plateau using remote sensing, ground–based instruments and cloud models, PhD Thesis, University of Ferrara, 2012, 141 pp. [2] Ueno, K., Characteristics of Plateau-scale precipitation in Tibet estimated by satellite data during 1993 monsoon season. J. Meteor. Soc. Japan, 76 (1998) 533-548. [3] Uyeda, U., Yamada H., Horikomi Y., Shirooka R., Shimizu S., Liping L., Ueno K., Fujii H., Koike T., Characteristics of convective clouds observed by a Doppler radar at Naqu on Tibetan Plateau during the GAME-Tibet IOP. J. Met. Soc. Japan. 79 (2001) 463-474. [4] Fuji, H., Koike T., Development of a TRMM/TMI Algorithm for Precipitation in the Tibetan Plateau by Considering Effects of Land Surface Emissivity. J. Meteor. Soc. Japan 79 (2001) 475-483. [5] Yao, Z., Li W., Zhu Y., Zhao B., Chen Y., Remote Sensing of Precipitation on the Tibetan Plateau Using the TRMM Microwave Imager. J. Appl. Meteor. 40 (2001) 1381-1392. [6] Yin, Z-Y, Zhang X., Liu Z., Colella M., Chen X., An Assessment of the Biases of Satellite Rainfall Estimates over the Tibetan Plateau and Correction Methods Based on Topographic Analysis. J. Hydrometeor, 9 (2008) 301–326. [7] Levizzani, V., P. Bauer and J. F. Turk, (eds.), Measuring precipitation from space: EURAINSAT & the future, Advances in global change research, Vol. 28, Springer, 2007. [8] Shimizu, S., Ueno K., Fujii H., Yamada H., Shirooka R., Liu L., Mesoscale characteristics and structures of stratiform precipitation on the Tibetan Plateau. J. Meteor. Soc. Japan, 79 (2001) 435-461. [9] Stephens, G. L., and Coauthors, The CloudSat mission and the A-Train. Bull. Amer. Meteor. Soc. 83 (2002) 1771–1790. [10] Kulie, M. S., Bennartz R., Utilizing spaceborne radars to retrieve dry snowfall. J. Appl. Meteor. Climatol. 48 (2009) 2564–2580. [11] Sano P., Casella D., Mugnai A:, Schiavon G., Smith E. A., Tripoli G. J., Transitioning from CRD to CDRD in Bayesian retrieval of rainfall from satellite passive microwave measurements: Part 1. Algorithm description and testing," IEEE Trans. Geosci. Remote Sens., in press, 2013. [12] Mugnai, A., Casella D., Cattani E., Dietrich S., Laviola S., Levizzani V., Panegrossi G., Petracca M., Sanò P., Di Paola F., Biron D., De Leonibus L., Melfi D., Rosci P, Vocino A., Zauli F., Puca S, Rinollo A., Milani L., Porcù F., Gattari F., Precipitation Products from the Hydrology SAF, Nat. Hazards Earth Sys. Sci., in press, 2013. [13] Kidd, C., Levizzani V., Status of satellite precipitation retrievals, Hydrol. Earth Syst. Sci. 15 (2011) 1109-1116. [14] Casella, D., Dietrich S., Di Paola F., Formenton M., Mugnai A., Porcù F., Sanò P., PM-GCD – a combined IR–MW satellite technique for frequent retrieval of heavy precipitation, Nat. Hazards International Workshop on Terrestrial Water Cycle Observation and Modeling from Space: Innovation and Reliability of Data Products Earth Sys. Sci. 11 (2012) 231-240. [15] Kotroni, V, Lagouvardos K., Defer E, Dietrich S, Porcù F, Medaglia C.M, Demirtas M, The Antalya 5 December 2002 storm: observations and model analysis. J. Appl. Meteor. 45 (2006) 576-590. [16] Capacci, D., Porcù F., Evaluation of a satellite multispectral VIS/IR daytime statistical rain-rate classifier and comparison with passive microwave rainfall estimates, J. Appl. Meteor. Clim. 48 (2009) 284-300. [17] Huffman, G.J., Adler R.F., Bolvin D.T., Gu G., Nelkin E.J., Bowman K.P., Hong Y., Stocker E.F., Wolff D.B., The TRMM Multi-satellite Precipitation Analysis: Quasi-Global, Multi-Year, Combined-Sensor Precipitation Estimates at Fine Scale. J. Hydrometeor. 8 (2007) 38-55. [18] Joyce, R. J., Janowiak J. E., Arkin P. A., Xie P., CMORPH: A method that produces global precipitation estimated from passive microwave and infrared data at high spatial and temporal resolution. J. Appl. Meteorol. 5 (2004) 487-503.
© Copyright 2026 Paperzz