4.15 USING METEOSAT SECOND GENERATION HIGH RESOLUTION VISIBLE DATA FOR THE IMPOVEMENT OF THE RAPID DEVELOPPING THUNDERSTORM PRODUCT

Transcription

1 4.15 USNG METEOSAT SECOND GENEATON HGH ESOLUTON VSBLE DATA FO THE MPOVEMENT OF THE APD DEVELOPPNG THUNDESTOM PODUCT Oleksiy Kryvobok * Ukrainian HydroMeteorologial nstitute Kyiv, Ukraine Stephane Senesi Meteo-Frane, DPEV/P Toulouse, Frane Christophe Morel Meteo-Frane, DPEV/P Toulouse, Frane 1. NTODUCTON The paper presents the results of the use of High esolution Visible (HVS) hannel data of MSG for the improvement of the disrimination of onvetive systems from all the traked louds systems in the apid Developing Thunderstorm (DT) produt. The priniple is to estimate of young onvetive loud parameters due to the inreased spatial resolution of the HVS hannel (1 km at subsatellite point against 3 km for the other hannels). This should allow the HVS hannel to desribe smaller louds than the 10.8 hannel and improve the earliness of the first detetion of onvetive systems during their growing phase in order to be able to warn and provide information to a foreaster when severe weather may our. VS =N + (1-N) (1) where N the part of pixel overed by loud, - the radiane sensed from the loudy part of pixel, - the radiane sensed from the sky part of pixel. After alibration of satellite data we rewrite Eq. (1) in terms of refletane () as VS =N + (1-N) (2) where VS - the measured refletane, - the top of atmosphere (TOA) refletane of loud, - the TOA refletane of sky surfae. The loud fration N is evaluated from Eq. (2) as 2. CLOUD FACTON ESTMATON USNG VSBLE DATA N vis = (3) 2.1 Basi definitions of loud fration estimation using visible data The radiane, measured by a satellite radiometer above a partly loudy overed pixel at visible wavelengths ( VS ), is given by Corresponding author address: Oleksiy Kryvobok Ukrainian Hydrometeorologial nstitute, 37 Nauki str., 03028, Kyiv, Ukraine; o_kryvobok@hotmail.om From Eq. (3) the loud fration is determined by,. n our further reasoning is TOA refletane of fully loudy pixel and refletane of sky pixel. 2.2 Clear sky pixel TOA refletane is TOA We used the sky nearby pixel method for the estimation of sky TOA refletane. The idea of the method is to find the sky pixel on

2 the image, whih has the same or almost the same TOA refletane as the loudy ontaminated pixel, in the ase if the latter would be loud free. t is well-known that TOA refletane in visible hannels is determined by surfae refletane and atmospheri ontribution. We use the visible refletane map (Gutman at el, 1995), the land over map and monthly atmospheri integrated water vapour ontent limatology data (Oort, 1981) to find the most representative loud free nearby pixel. The visible refletane map and the land over map are used to find the loud free pixel whih has the losest surfae refletane as a loudy pixel on HVS image, beause the refletane map for this hannel is not available. We proeed from the supposition that loudy and the loud free pixels have the same surfae refletane, if they belong to the same ategory on land over map and have the minimal differene in visible refletanes. TOA refletanes of two pixels will be the same, if the integrated water vapour ontent differene will be minimum in over these pixels. This approah is appliable for low (VS0.6, VS08) and for high resolution (HVS) data over a small domain around the pixel of interest where satellite and sun angles are almost onstant. For VS06 data we use only the visible refletane map, beause the map and data are obtained in the same spetral region. For VS08 data we use only integrated water vapour ontent information, beause spetral refletane in the hannel is very low over sea (where this data are used). Comparison with methods based on adiative Transfer Model (TM) alulation shows (Kryvobok, 2005) that the nearby sky pixel refletane, whih is observed by satellite itself, is more plausible that simulated by TM, beause this value orresponds to the real atmospheri state instead of simulated one, whih is based mostly on limatologial data. 2.3 Estimation of fully loudy pixel refletane The estimation of fully loudy pixel refletane is based on simulation of loudy refletane on TOA by TM. Theoretial modeling and real observations (Cahalan et. al, 1994; Kobayashi, 1993; iihimaki, 2001) show that for optially thik loud (>100) TOA refletane to a great extent, does not inrease with further inrease of optial and geometrial thiknesses of loud. Convetive louds are really optially thik, even if their geometrial size is not so large (O Hirok et. al, 1998). After the review of last studies (osenfeld et. al, 1998; Hess et. al, 1998) we have found the observed typial values of onvetive louds properties, whih we used for the TM alulation of fully loudy refletane. We onsider in our omputation the onvetive loud, whih is equivalent to the growing umulonimbus about 700m height. 2.4 Cloudy pixels diagnosti on visible hannels We use threshold tehnique for loudy pixel diagnosti. Threshold tehnique is based on a pixel by pixel analysis of radiane and omparison of this value with a threshold. The pixel is lassified as loud ontaminated if a value of measured radiane is more than a threshold. One of the main advantage of the threshold method is that it is relatively easy to adapt thresholds to varying meteorologial onditions, viewing geometry using external data (NWP data, TM alulation, limatologial atlas). Aording to the Wieliki and Parker (1992) the refletane threshold in visible hannels is the most sensitive for low louds identifiation. As they noted a refletane threshold an be used effiiently for loud altitudes below 3 km. The tests are applied to the visible or near-infrared TOA refletanes aim to detet louds having a refletane higher than the underlying surfaes. The visible refletane threshold is better to use over land, where loud free refletane is not so high and less dependent on vegetation, near infrared refletane threshold is better to use over the sea, beause the solar refletion over sea is very low. These thresholds are alulated using TM depending on different atmospheri states, viewing geometry and different kind of surfaes. n order to get more realisti threshold values, whih an be pre-omputed we need to add TOA refletane a pre-defined margin (to aount for noise and variation of surfae refletane). 3. METHODS OF CLOUD FACTON ESTMATON FOM SEV DATA 3.1 Estimation of pixel loud fration as mean value of N high-resolution pixels The loud fration estimation of a low resolution pixel is based on the estimation of the loud fration of eah high resolution pixel in the orresponding low resolution pixel. For eah high

3 resolution pixel we estimate the loud fration N H as N H =( threshold0.75 )/( threshold 0.75 ) (4) where the refletane of HVS pixel, threshold0.75 the threshold refletane, 0.75 the refletane of the fully loud overed HVS pixel. We used the abbreviation High resolution Cloud Fration Field (HCFF) for the loud fration field estimated from HVS data. The loud fration (N) of low resolution pixel is 9 N=( N H, i )/9, (5) i= 1 where 9 is the number of HVS pixels inside one low resolution. t is the most suitable method for the loud fration estimation of small onvetive louds. The auray of the method depends on the threshold refletane and We used the abbreviation Converted Cloud Fration Field (CCFF) for loud fration field estimated using this method. 3.2 Estimation of loud fration using low resolution VS data Data of VS0.6 hannel is preferable for loud fration estimation over land, beause loud free refletane is not so high and less dependent on vegetation N=( threshold0.6 )/( 0.6 threshold0.6 ) (6) Data of VS0.8 hannel is used over sea, beause the solar refletion over oean is very low N=( threshold0.8 )/( 0.8 threshold0.8 ) (7) where threshold0.6, threshold0.8, 0.6, 0.8 the refletane thresholds and fully loudy pixels refletanes for orresponding hannels (VS0.6,VS0.8). The method gives a reasonable result for louds with a size omparable of a pixel size, but it depends on a threshold refletane and.we used the abbreviation Low resolution Cloud Fration Field (LCFF) for the loud fration field obtained from low resolution data. 4. DESCPTON OF CLOUD CONVECTVE SYSTEMS We have studied two developing onvetive systems on 4 July These systems were observed over North part of Afria (Figure). The first system (indiated as on Figure) is an example of the very rapid developing thunderstorm haraterized by ontinuous growth of one loud from 11:15 to 12:00 UTC. Only few very small louds are observed at 11:00 UTC and in the next min they merged and beame as one very large mature onvetive loud. The dynami of the development of this system is shown on Figure. On the DT produt the system was deteted first at 11:30 UTC with orresponding harateristis: phase of development growing, min temperature 46 C. At the next 15 min it was deteted also as growing with a min temperature 52 C and at 12:00 UTC the system was deteted as dereasing with a min temperature 50 C. The seond onvetive system (indiated as on Figure) was loated lose first one. This system is haraterized by isolated small onvetive louds, whih begun to merge in a mature phase of their development. They began to grow, the most rapidly, at 12:00 UTC. At 12:15 UTC the loud system is onsisted from some smaller onvetive louds, whih begin to merge. This onvetive system is muh smaller than the first one. First detetion on the DT produt was at 12:15. Charateristis of the seond system on the DT produt were the following: phase of development growing, min temperature 37 C.

4 11: 00 11:15 11:30 11:45 12:00 12:15 Figure. HVS images of the rapid developing thunderstorm (11:00 12:15 UTC) on 4 July 2004 The red and yellow retangles show the development of onvetive systems and. 5. ANALYSS OF CLOUD FACTON FELD a) Convetive system. The loud fration field of onvetive system shows that at 11:15 UTC there are some loudy objets on the images. They are haraterized by small onvetive louds of few pixels size. We obtained inhomogeneous loud fration fields over the region. The values of HCFF are almost 100%. t means that some of onvetive louds were as large as few pixels and well developed. On low resolution images the values of loud fration fields are max 79% (CCFF) and 56% (LCFF). On the next images the loud fration fields are haraterized by inreased values 100% and by inreased number of pixels, whih loud fration value are 100%. Clouds merged and loud fration field beame more homogeneous with dereased values on the edge of the louds, where partly loudy pixels are loated. This was observed on HCFF and CCFF from 11:30 to 12:00 UTC. The values of LCFF are not higher than 90 95% for this period of time. We have also analyzed the evolution of HCFF over the small domain (6 6 HVS pixels). The analysis shows that at 11:15 UTC when very intensive loud formation (appearane of small louds) was the HCFF was inhomogeneous. The values of loud fration are varying from 2 to 92%. There were an intensive development and a merging of small louds on the next images. The loud fration beame homogeneous with values about 100%. a) Convetive system. The onvetion development of the onvetive loud system was not so obvious. The loud fration fields on low resolution images are inhomogeneous with values from 14 to 29% at 11:30 UTC and even 78% at 12:00 UTC. On the ontrary on HCFF at 11:30 UTC these values are

5 65% (32 42 for CCFF). On the next images we see ontinuous inrease of loud fration values, whih are 100% (100% for CCFF) at 12:00 UTC. The dynami of the loud fration field of small loud over the domain (9 9 HVS pixels) shows that between 11:15 and 11:30 UTC we an not see any onstant enter of development of onvetive loud. The values of HCFF are varying from 6 to 64%. Very likely those louds dissipated or moved outside the domain after formation. An intensive formation of the louds begun at 11:45 UTC, half of pixels beame loudy with loud fration values from 2 to 93%. Then, when louds developed and merged (12:00 UTC), HCFF beame homogeneous, mostly, with high values (80 100%). n respet to that the main task of the study is to find the usefulness of HVS data for the DT produt to improve an earliness of first detetion of onvetive systems during their growing phase, we analyzed more detailed the loud fration field earlier than was the first detetion on the DT produt. As already noted, the first detetion of the onvetive system was at 11:30 UTC and the onvetive system - at 12:15 UTC. Analysis of the loud fration field at 11:15 UTC for the onvetive system (15 min before the first detetion on the DT produt) shows that the highest loud fration values (HCFF) are reahed 92% (five pixels, less than one low resolution). For CCFF and LCFF the maximum values are 78 and 56 orrespondingly, all other values are muh less. Analysis of the loud fration field at 11:45 and 12:00 UTC for the onvetive system shows that the loud fration values are reahed about 90%, 30 minutes before (5 pixels) and, 15 minutes before (almost for all pixels), the first detetion on the DT produt. Correspondingly, CCFF values are reahed 70% (one pixel), 80 90% (3 pixels) and LCFF values are 41% (one pixel), 78% (one pixel). Thus high value ( 90%) of onvetive loud fration an be observed on HCFF even before 30 minutes the first detetion on the DT produt, only few pixels have this range of values. We have obtained high value of loud fration ( 80 90%) on CCFF only 15 minutes before the first detetion on the DT produt and, mostly, for one pixel in the onvetive systems. The analysis of loud fration fields shows that the method based on analysis of visible satellite data gives plausible results. n the ase of small broken louds we obtained the realisti loud fration field, inhomogeneous with mostly low values of N. Over a large mature loud we obtained a high value of the loud fration over the thikest part of loud (enter) and lower value on the edge of the loud where really partly loudy pixels (part of pixel overed by loud) or pixels, whih are not thik enough are loated. Over a large loud we obtained a homogeneous loud fration field, beause real onvetive loud is thik and over a large area without any gaps inside of it. Cloud over evolution shows a real growth of louds - gradual inrease of N and inrease of loudy pixels. At the same time we need to note of an underestimation of loud fration field using low resolution (VS0.6) data. On our point of view it is, beause simulated loudy refletane values are higher that really should be in this hannel. We expet that more detailed TM and more detailed mirophysial and geometrial information for onvetive louds will derease the unertainty in estimation of the TOA loudy refletane. Another problem of the used method for loud fration estimation is a dependeny on the atmospheri state. ntegrated water vapour ontent and aerosol onentration strong influene on the sky TOA refletane. Current version of the method uses limatologial water vapour ontent and aerosol onentration, whih gives some unertainty in the TOA refletane estimation and subsequently on loud fration. Use of NWP data instead of limatologial will derease of unertainty in the TOA refletane estimation and improve evaluation of the loud fration. The weakness of the method is, also, that it needs the refletane map of surfae in HVS hannel, whih is not available now. The spetral interval of HVS hannel over part of visible and nearinfrared, where surfae types have ompletely different refletane harateristis (vegetation). The sheme, whih we used there, based on joint analysis of land over map and visible refletane map in 0.6 µ (AVH hannel 1), that does not reah a high auray in the sky TOA refletane and subsequently in the loud fration. 6. CONCLUSON The study shows that more realisti loud fration fields were obtained using HVS data than low resolution data. Computed loud fration fields show plausible values when ompared to a visual interpretation of the HVS image; nevertheless, some unrealisti values raise the question of designing a reliable sheme for dynamially adapting the optial thikness of louds used in the refletane simulation to the atual loud field. The next question would be to get a referene

6 observation data set for objetive validation of frational loud over for the more diffiult ase (regarding radiative transfer modeling) of broken, non-plane parallel, umulus louds. So, the possible use of derived frational loud over should be only in a qualitative way at that stage. 7. EFEENCES Gutman G., Tarpley D., gnatov A., and Olson S., The enhaned NOAA global land dataset from the Advaned Very High esolution adiometer, Bulletin of the Amerian Soiety,76 (7) Oort A.H., 1983, Global atmospheri irulation statistis, NOAA professional Paper No. 14. Kryvobok, O., 2005, Monitoring harateristis of young onvetive louds for DT produt using MSG data, Visiting Sientist eport, SAF/NWC/OP/MFT/SC/P/02. Laura iihimaki, 2001, A disussion of loud albedo and how it ould be influened by (anthropogeni) CCN onentrations. Private ommuniation O Hirok, W. and Gautier C., A Three- Dimentional adiative Transfer Model to nvestigate the Solar adiation with a Cloudy Atmosphere. Part :Spatial Effets, J.Atm.Si., 55, osenfeld, D. and. M. Lensky Satellitebased insights into preipitation formation proesses in ontinental and maritime onvetive louds. Bull. Amer. Meteor. So., 79, Hess, M., P. Koepke and. Shult Optial properties of aerosols and louds. Bull. Amer. Meteor. So.,79, Calahan,., idgway, W., Wisombe, W., ndependent Pixel and Monte-Carlo Estimates of Stratoumulus Albedo, J.Atm.Si., 51, Kobayashi, T., Effets Due to Cloud Geometry on Biases in the Albedo Derived from adiane Measurements, J.Climate, 6,

7