A Night-Rain Delineation Algorithm for Infrared Satellite Data Based on Microphysical Considerations

1218 VOLUME 42J O U R N A L O F A P P L I E D M E T E O R O L O G Y

q 2003 American Meteorological Society

A Night-Rain Delineation Algorithm for Infrared Satellite Data Based onMicrophysical Considerations

ITAMAR M. LENSKY AND DANIEL ROSENFELD

Institute of Earth Science, Hebrew University of Jerusalem, Jerusalem, Israel

(Manuscript received 7 February 2002, in final form 19 December 2002)

ABSTRACT

Information about the microstructure and precipitation potential of cloud revealed by the brightness temperaturedifference between a thermal IR channel (11 mm) and a mid-IR channel (3.7 mm) at nighttime is implementedinto a precipitation delineation algorithm. The algorithm is simple to implement and uses data available fromalmost all of the operational and research satellites. The delineation algorithm performs well also in cases ofwarm clouds over land for which passive microwave algorithms fail.

1. Introduction

Rain estimation from satellite data is often done intwo stages: 1) delineation of the rainy area and 2) es-timation of the rain rates of the delineated rainy area.This study will focus on the first stage—the delineation,at nighttime, using the emitted radiation. Lensky andRosenfeld (2002, manuscript submitted to J. Appl. Me-teor., hereinafter LAR) analyzed the emitted radiation,showing some insights into the precipitation formationprocesses at nighttime. Application of these insights intoa nighttime precipitation delineation algorithm will bedescribed here. Passive microwave (MW) rain algo-rithms have poor performance in detecting warm rainover land. The suggested delineation algorithm fills thisgap in the capabilities of passive MW rain detection.

The conditions for development of precipitation inclouds are 1) sufficient water (thick cloud) and 2) aneffective mechanism for converting the small clouddroplets suspended in the air into large precipitationparticles.

Precipitation forms when the particles become largeenough to have terminal fall velocity of a few metersper second so that they can get to the ground beforeevaporating. Effective mechanisms of converting cloudwater into precipitation are coalescence of cloud drop-lets (warm processes) and ice aggregation (cold pro-cesses). Thick clouds with large particles are likely tohave enough column-integrated water and effective co-alescence to develop precipitation. Albrecht (1989), Ro-senfeld and Gutman (1994), and Lensky and Rosenfeld

Corresponding author address: Itamar M. Lensky, Institute ofEarth Science, Hebrew University of Jerusalem, Jerusalem 91904,Israel.E-mail: [email protected]

(1997) used an effective radius of 14–15 mm as a thresh-old for precipitation for both cold and warm processesin the daytime. At nighttime, the brightness temperaturedifference (BTD) between a thermal IR channel (11 mm)and a mid-IR channel (3.7 mm) is used.

This study is unique in its attempt to delineate rainat nighttime using microphysical information that is re-trieved from the BTD between mid-IR and thermal IRchannels. Other approaches to rain-detection algorithms,using emitted IR radiation, can be found in the literature,such as Anagnostou et al. (1999) and Inoue and Aonashi(2000, hereinafter IA00).

Anagnostou et al. (1999) build upon Adler and Ne-gri’s (1988) convective–stratiform technique, whichuses average morphologic characteristics to identifyconvective and stratiform areas in clusters of 253-Kisotherms identified within the IR brightness tempera-ture (BT) array.

IA00 selected the following parameters as cloud in-formation:

1) radiance ratio of 0.6 and 1.6 mm,2) BTD between 11 and 12 mm (BTD45),3) BTD between 3.8 and 11 mm (BTD34), and4) BT in 11-mm visible wavelength (T4).

They noticed that BTD34 is more sensitive then BTD45

to optically thicker cirrus clouds and found that BTD34

, 8 K and T4 , 260 K are effective in delineating rainarea. We will show here how using microphysical con-siderations can improve the rain-area delineation.

We used for the development of the algorithm 236coincidence subsets (CS) of the Tropical Rainfall Mea-suring Mission (TRMM) satellite over Israel and Kwa-jalein, Republic of the Marshall Islands, from 1998 and516 CSs from 1999 for the validation (same locations).

SEPTEMBER 2003 1219L E N S K Y A N D R O S E N F E L D

FIG. 1. (a) The BTD between a 3.7- and 11-mm channel (T3.7211)produced by an RT model. (b) A schematic description of the inputfor the RT model. The cloud has top temperature of 260 K and varyingoptical depth t and effective radius re. The gray zone is where noprecipitation is expected because the cloud is not thick enough, theeffective radius is not large enough, or both.

The results are presented here, together with two casestudies that demonstrate the capabilities of the delin-eation technique as compared with the IA00 algorithmand a passive microwave rain-detection algorithm.

The next section will describe the data that were col-lected for this study. The concept of the night-rain al-gorithm will be demonstrated using a radiative transfermodel in section 3. The nighttime delineation algorithmis described in section 4. In section 5 results of the night-rain algorithm for the validation CSs are shown, togetherwith two case studies. The summary and conclusionswill be presented in section 6.

2. The data

The data for this study were taken from three differentinstruments on the TRMM satellite. Data from the vis-ible and infrared scanner (VIRS) were used for the de-lineation algorithm. The VIRS is a radiometer that hasfive channels in the visible and infrared wavelengths:0.65, 1.6, 3.7, 11, and 12 mm. The radiance from thefive channels is given in watts per meter squared. Duringnighttime, BT is calculated for channels 3, 4, and 5.The spatial resolution of the VIRS pixels is 2 km 3 2km at nadir.

Results of the algorithm using the VIRS data werecompared with the 2A12 surface rain algorithm of theTRMM Microwave Imager (TMI). A description of theTMI can be found in Kummerow et al. (1998). The 2A12algorithm provides rainfall rates based upon the ninechannels of the TMI. The TMI operational rain algo-rithm varies according to the geographical zone definedin the TRMM data: land, coast, and ocean. A descriptionof the algorithm can be found in Kummerow et al.(1996). The 2A12 rain algorithm was designed to workover ocean. Other algorithms such as Grecu and An-agnostou (2001), which were developed to estimateoverland rainfall areas and rates, have superior perfor-mance to the 2A12 rain algorithm over land. Both thedelineation algorithm and the 2A12 delineation werevalidated against direct measurements from the TRMMprecipitation radar (PR).

The training dataset is from two CSs: over Israel(ISBN) and over Kwajalein (KWAJ) from 1998 (all to-gether 236 overpasses). The validation dataset is fromthe same two CSs but from 1999 (altogether 516 over-passes). All of the TRMM data are version 5.

3. Radiative transfer model simulations

Figure 1a, (after Fig. 2 of LAR) shows the BTD be-tween a 3.7-mm mid-IR channel and an 11-mm thermalchannel (T3.7211) produced by a radiative transfer (RT)model (Ricchiazzi et al. 1998). The input for the RTmodel is described schematically in Fig. 1b. The resultsare from 31 runs for cloud with a top temperature of260 K, varying optical depth t and effective radius re,and no solar radiation. The gray zone is where no pre-

cipitation is expected because the cloud is not thickenough, the effective radius is not large enough, or both.The RT model showed large negative T3.7211 values foroptically thick clouds with small effective radius (T3.7211

; 248C) and large positive T3.7211 values for semitrans-parent clouds (T3.7211 ; 68C). Rain clouds, that is, thickclouds with large effective radius, produced BTD in theinterval (21 , T3.7211 , 4) for the clouds specified inthe input.

To delineate the precipitation, the cloud-top temper-ature and BTD TRMM data from all of the overpassesover Israel (continental clouds) and Kwajalein (maritimeclouds) with rain events in 1998 were accumulated. Thedata of the maritime and continental clouds were dividedin to groups according to the cloud-top temperature.Each group contained all of the pixels within a cloud-top temperature interval of 18C. We are now lookingfor two BTD thresholds. The lower BTD threshold willdelineate rain pixels from pixels with small BTD dueto small particles near cloud top, and the large BTDthreshold will delineate rain pixels from pixels withlarge BTD due to semitransparent clouds. LAR showedthat typical curves of BTD versus cloud-top temperaturefor maritime clouds are shifted to larger BTD valuesbecause even young maritime clouds have large particlesize. Therefore, whereas in continental clouds we wishto exclude both pixels with ‘‘too-small BTD’’ and pixelswith ‘‘too-large BTD,’’ in microphysically maritimeclouds the too-small-BTD case is rare.

1220 VOLUME 42J O U R N A L O F A P P L I E D M E T E O R O L O G Y

FIG. 2. (a) The BTD intervals associated with the CSI in (b) thehighest CSI for each 18C cloud-top temperature interval for cloudsover Israel. (c) As in (a) but for clouds over Kwajalein. (d) As in (b)but for clouds over Kwajalein.

FIG. 3. The BTD intervals used to delineate precipitation for con-tinental clouds (ISBN) and maritime clouds (KWAJ), and that sug-gested by IA00 (IA).

FIG. 4. Same as Fig. 1a but taking the observation perspective.The boxes marked a and b have ambiguous BTD.

The critical success index (CSI) test was used to findthe couples of BTD values that would best delineate theprecipitation according to the PR, that is, the maximumnumber of precipitating pixels and minimum number ofnonprecipitating pixels within the checked BTD intervaland the minimum number of precipitating pixels out ofthe checked BTD interval. The CSI is defined as (Dos-well et al. 1990)

CSI 5 hit/(hit 1 miss 1 false alarm),

where ‘‘hit’’ is the number of pixels in the checked BTDinterval with PR rain, ‘‘miss’’ is the number of pixelsout of the checked BTD interval with PR rain, and‘‘false alarm’’ is the number of pixels in the checkedBTD interval with no PR rain.

Figure 2b shows the highest CSI for each 18C cloud-top temperature interval for clouds over Israel for thetraining dataset. The BTD intervals associated with thisCSI are shown in Fig. 2a. Figures 2d and 2c are thesame, but for clouds over Kwajalein. Figure 3 showsBTD intervals for continental clouds (ISBN) and formaritime clouds (KWAJ) and those suggested by IA00.A comparison of the different approaches will be dis-cussed in section 5.

Table 1 summarizes the counts (hit, miss, false alarm,and no rain), CSI, probability of detection (POD), andfalse-alarm rate (FAR) for the VIRS algorithm describedin this paper (LR), the TMI algorithm (2A12), and IA00.The data were divided by cloud-top temperature (colderand warmer than 2308C), and surface (land, coast, andsea). The data are the validation dataset from 1999, overKwajalein, Israel, and the eastern Mediterranean Sea.

4. The nighttime rain-delineation algorithm

The shaded zone of Figs. 2b and 2d (continental andmaritime clouds, respectively) is the best BTD range asa function of cloud-top temperature, according to theCSI test. This test was done on the basis of individualpixels. Some of the false-alarm cases can be treatedwhen examining neighbor pixels.

Figure 1 was drawn with the perspective of the RTmodel in mind, that is, the input in the lower part, withthe raining and nonraining zones as expected from theproperties of the cloud in the input, and the RT modelresults in the upper part. Figure 4 shows the upper partof Fig. 1, but this time with the observations perspective,that is, the measured BTD. The precipitating zone is thesame, where the upper and lower shadings respectivelystand for nonprecipitating semitransparent clouds withBTD larger than the higher BTD threshold and cloudswith small particle size, with BTD smaller than the low-er BTD threshold. In the range of BTD between the twothresholds there are three locations of ‘‘valid BTD’’

SEPTEMBER 2003 1221L E N S K Y A N D R O S E N F E L D

TABLE 1. CSI, POD, FAR, and counts (hit, miss, false alarm, and no rain) of pixels for the algorithm described in this paper (LR), theTMI algorithm (2A12), and the IA00 algorithm (IA). The data were divided by cloud-top temperature (colder and warmer than 2308C) andsurface (land, coast, and sea). The data are from the validation dataset from 1999 over Israel and the eastern Mediterranean Sea (ISBN) andover Kwajalein (KWAJ).

Hit

No. %

Miss

No. %

False alarm

No. %

No rain

No. % CSI POD FAR

ISBNWarm2 312 500

Land49.9%

Coast27%

Sea23.1%

LR2A12IALR2A12IALR2A12IA

2784182

27464984

5243125351138731475

0.240.020.240.800.080.500.660.720.28

4064666641028700

13 16010 559

79697607

10 005

0.350.580.361.392.111.691.491.421.87

13 7282546

18 379807373916302417952854264

1.190.221.591.291.191.010.780.990.80

1 133 6671 144 8491 129 016

601 948602 630603 719518 893517 787518 808

98.2299.1897.8196.5296.6296.8097.0796.8797.05

0.140.020.110.230.020.160.220.230.09

0.410.030.400.360.040.230.310.340.13

0.830.930.870.620.930.670.540.580.74

Cold1 050 000

Land56.6%

Coast26%

Sea17.4%

LR2A12IALR2A12IALR2A12IA

17 9971035

22 42414 390

253317 473

96826593

15 404

3.030.173.775.280.936.415.283.598.40

913326 095

47068972

20 82958898848

11 9373126

1.544.390.793.297.642.164.826.511.70

45 9471788

103 94519 196

489742 741

98951895

28 356

7.730.30

17.57.041.80

15.685.391.03

15.46

520 985565 144462 987229 946244 245206 401155 009163 009136 548

87.7095.1477.9484.3889.6375.7584.5088.8774.44

0.250.040.170.340.090.260.340.320.33

0.660.040.830.620.110.750.520.360.83

0.720.630.820.570.660.710.510.220.65

KWAJWarm3 955 000

Coast14.7%

Sea85.3%

LR2A12IALR2A12IA

3164246139

26 00935 131

2449

0.560.040.020.771.040.07

11 19814 11614 22372 75563 63396 315

1.972.482.502.151.882.84

5316439457

61 84060 25010 169

0.940.080.081.831.780.30

548 621553 498553 480

3 226 0973 227 6873 277 768

96.5397.4097.4095.2595.3096.79

0.160.020.010.160.220.02

0.220.020.010.260.360.02

0.630.640.770.700.630.81

Cold2 852 500

Coast16.7%

Sea83.3%

LR2A12IALR2A12IA

40 07628 16734 314

219 055246 173173 767

8.445.937.229.22

10.367.31

23 26135 17029 023

130 630103 512175 918

4.907.406.115.494.357.40

44 9917605

31 639266 266173 239171 679

9.471.606.66

11.207.297.22

366 734404 120380 086

1 761 4871 854 5141 856 074

77.2085.0780.0174.0978.0078.07

0.370.400.360.360.470.33

0.630.440.540.630.700.50

0.530.210.480.550.410.50

outside of the box of the precipitating pixels (as weexpect from the input to the RT model). The rightmostand the leftmost locations are associated with surfaceradiation upwelling through very thin clouds or directlyfrom the ground. These locations are marked with theletter a. These pixels are easy to treat, because theirbrightness temperature is usually warmer by 108–158Cthan the warmest clouds. The third location of validBTD outside of the box of the precipitating pixels isoften associated with pixels residing on the edges ofclouds with small particles, and therefore with BTDsmaller than the lower BTD threshold. If these cloudsare viewed on the background of pixels with positiveBTD, such as surface or semitransparent clouds withvery large BTD, then somewhere between the core of

the cloud with small particles (and small BTD) and theadjacent pixels with positive BTD, there will surely besome pixels with valid BTD located on an ‘‘envelope’’of pixels on the cloud edge. This case is marked withthe letter b in Fig. 4. To detect these cloud edges, pixelswith valid BTD are checked for adjacent pixels withboth higher and lower BTD than the BTD thresholds.

5. Results and case studies

Figures 5a–c show one case study of clouds over theMediterranean Sea from 2054 UTC 29 January 1999,and Figs. 5d–f show one case study of clouds over Kwa-jalein from 1523 UTC 5 May 1999. Figures 5a and 5dshow a red–green–blue (RGB) composition of T3.7212,

1222 VOLUME 42J O U R N A L O F A P P L I E D M E T E O R O L O G Y

FIG. 5. Two case studies: (a)–(c) clouds over the eastern Mediterranean Sea from 2054 UTC 29 Jan 1999 and (d)–(f ) clouds over Kwajalein(98N, 1668E) from 1523 UTC 5 May 1999. The RGB composition in (a) and (d) is T3.7212 (T35inv) in red, T3.7211 (T34inv) in green (both,color inversed), and T11 (T4) in blue. The rain delineation algorithms in (b) and (e) are LR in khaki, the removed cloud edges (edge)highlighted in gray, the IA00 delineation algorithm in olive green, and where both algorithms (both) predict precipitation colored in lightgreen. In (c) and (f ) the validation PR-measured precipitation is red (PR), the TMI-measured precipitation is blue (TMI), and where bothPR and TMI measured precipitation (both) is in purple.

(inversed) in red, T3.7211 (inversed) in green, and T11 inblue. Figures 5b and 5e show the rain-delineation al-gorithm (LR) in khaki, the removed ‘‘cloud edges’’(edge) in gray, the IA00 delineation algorithm in olivegreen, and where both algorithms (both) predict precip-itation in light green. Figures 5c and 5f show the PR-measured precipitation in red (PR), the TMI-measuredprecipitation in blue (TMI), and where both PR and TMImeasured precipitation (both) in purple.

The black clouds in Figs. 5a and 5d are semitrans-parent cirrus clouds with large BTD (T3.7211 . 10 K;T11212 . 10 K) and low T11 (;250 K). The bright yellowclouds are clouds with small particles—negative T3.7211

(,22 K). The clouds with darker yellow–brown are theclouds with the T3.7211 in the interval within which weexpect precipitation.

We chose to show in the ISBN case study those cloudsthat are in the temperature range of the IA00 algorithm,

SEPTEMBER 2003 1223L E N S K Y A N D R O S E N F E L D

FIG. 5. (Continued)

that is, where most of the cloud tops are colder than260 K. Thin cirrus clouds with large BTD that appearblack in Fig. 5a are successfully screened in Fig. 5b byboth algorithms. In Fig. 5b, the shallower clouds overthe Mediterranean Sea and Cyprus that are colored inkhaki are warmer than the IA00 260-K temperaturethreshold. In Fig. 5c, we can see that these are precip-itating clouds, according to the PR. Note the yellowcloud in the upper right-hand corner of Fig. 5a withtypical T3.7211 of 228C. In Fig. 5b, we can see that thecloud passed the IA00 algorithm criteria and was col-ored in olive green. It did not pass LR algorithm criteriabecause of the negative BTD that indicates that the par-ticles near the cloud top are too small to produce pre-cipitation, except for the pixels on the cloud edge thatwere removed (gray). Figure 5c shows that, indeed, the

PR did not detect precipitation in this cloud. Otherclouds that are colored in olive green are clouds thatpassed the IA00 criteria but did not pass the LR criteria,that is, with BTD smaller than 8 K but larger than theLR ISBN threshold for continental clouds, as shown inFig. 3. The PR in Fig. 5b shows that these clouds donot precipitate.

In the Kwajalein case study we can see again that thethin cirrus clouds that appear black in Fig. 5d are suc-cessfully screened in Fig. 5e by both algorithms. Thereis a case of multilayer clouds in the lower left-hand partof the figure, a semitransparent cirrus cloud over low-level precipitating cumulus clouds. This is a case inwhich both algorithms fail to delineate the precipitationunder the high-level cirrus clouds. The IA00 algorithmdetects the cumulonimbus clouds but fails to detect the

1224 VOLUME 42J O U R N A L O F A P P L I E D M E T E O R O L O G Y

FIG. 6. Results of the VIRS and TMI delineation vs the PR delineation for all of the continental-cloudsrain events in the validation dataset of 1999 over Israel. Each point represents one window of 50 3 50 VIRSpixels. The fraction of rainy pixels detected by the PR in the window is on the abscissa. The fraction of rainypixels detected by the VIRS and TMI in the window is on the ordinate; (a) warm rain, (b) cold rain overland, (c) warm rain, (d) cold rain over coast, (e) warm rain, and (f ) cold rain over over the MediterraneanSea. Error statistics can be found in Table 2.

SEPTEMBER 2003 1225L E N S K Y A N D R O S E N F E L D

FIG. 7. Same as Fig. 6, but for maritime clouds over Kwajalein: (a) warm rain, (b) cold rain over coast,(c) warm rain, and (d) cold rain over sea.

smaller cumulus precipitating clouds, as can be seen inFig. 5e.

Figures 6 and 7 show results of the VIRS and TMIdelineations vs the PR delineation for all of the rainevents in 1998 in Israel (Fig. 6) and Kwajalein (Fig. 7).The rain in Israel is from warm and cold mostly con-vective winter clouds, with typical top heights of 3–7km above sea level. The rain in Kwajalein is from equa-torial maritime clouds. Each point in the scattergramrepresents one ‘‘window’’ of 50 3 50 VIRS pixels. ThePR and TMI data were transformed to the VIRS co-ordinates for comparison and validation. The fractionof rainy pixels detected by the PR is on the abscissa,and the fraction of rainy pixels detected by the VIRSand TMI is on the ordinate. Figure 6a shows the warmrain over land. Note that the TMI (2A12) fails in thiscategory. For cold rain over land, the TMI has poorresults in Fig. 6b. The performance of the VIRS is aboutthe same for warm and cold clouds. Over coastal zones,the TMI performs better than over land (Figs. 6c,d), but

still the VIRS performance is slightly better. Over sea(Figs. 6e,f ) the TMI has better results, and it performsbest for maritime clouds over Kwajalein (Fig. 7). Table2 shows error statistics: correlation, bias, standard error,and variance for the data plotted in Figs. 6 and 7 forthe algorithm described in this paper, the IA00 algo-rithm, and the TMI algorithm (2A12).

6. Summary and conclusions

A rain-delineation algorithm for nighttime based onmicrophysical considerations was suggested. The al-gorithm uses BTD between a mid-IR channel and athermal IR channel to detect potentially precipitatingclouds, that is, optically thick clouds with large particlesnear cloud top. The algorithm is simple to implement,and it uses data available on most operational and re-search meteorological satellites. The results of the al-gorithm were compared with the operational TMI rainalgorithm and the delineation suggested by IA00. The

1226 VOLUME 42J O U R N A L O F A P P L I E D M E T E O R O L O G Y

TABLE 2. Error statistics: correlation (CC), bias, standard error (Stderr), and variance (Var) for the data plotted in Figs. 5, 6, and 7 forthree algorithms (Alg): the algorithm described in this paper (LR),the IA00 algorithm (IA), and the TMI algorithm (2A12). The datawere divided by cloud-top temperature as in Table 1 and by surface(land, coast, and sea), indicating the number of windows in eachcategory.

Alg CC Bias Std err Var

ISBNLand

Warm339

Cold189

LRIA2A12LRIA2A12

0.630.440.270.760.590.28

3.264.350.052.765.640.02

0.0820.1340.0060.0760.1510.005

0.0010.00200.0230.0590

CoastWarm463

Cold218

LRIA2A12LRIA2A12

0.790.40.390.780.570.43

0.970.770.721.442.640.31

0.030.050.0340.050.1050.025

0.0030.0030.0010.0260.0680.003

SeaWarm54

Cold13

LRIA2A12LRIA2A12

0.920.580.620.710.890.99

0.621.040.380.421.810.74

0.0450.1580.0760.0630.2290.05

0.0010.0020.0010.0050.1480.048

KWAJCoast

Warm770

Cold600

LRIA2A12LRIA2A12

0.530.060.740.770.80.94

0.640.060.651.321.020.97

0.0440.0150.0330.0340.030.016

0.00300.0020.0580.0510.045

SeaWarm812

Cold541

LRIA2A12LRIA2A12

0.580.160.850.760.770.91

1.030.171.021.440.971.24

0.0620.0320.0330.0420.0350.023

0.0080.0010.0050.0630.0450.049

TMI performs very well for maritime clouds over ocean,but for warm clouds over land the performance is poor.The proposed algorithm, on the other hand, handlesthese cases well, identifying the clouds with cloud-topmicrostructure favorable for precipitation.

Acknowledgments. This research has been obtainedfrom the cooperation between the Hebrew University ofJerusalem and NASDA in ADEOS-II research activity.

REFERENCES

Adler, R. F., and A. J. Negri, 1988: A satellite infrared technique toestimate tropical convective and stratiform rainfall. J. Appl. Me-teor., 27, 30–51.

Albrecht, B. A., 1989: Aerosols, cloud microphysics and fractionalcloudiness. Science, 245, 1227–1230.

Anagnostou, E. N., A. J. Negri, and R. F. Adler, 1999: A satelliteinfrared technique for diurnal rainfall variability studies. J. Geo-phys. Res., 104, 31 477–31 488.

Doswell, C. A., III, R. Davies-Jones, and D. L. Keller, 1990: Onsummary measures of skill in rare event forecasting based oncontingency tables. Wea. Forecasting, 5, 576–585.

Grecu, M., and E. N. Anagnostou, 2001: Overland precipitation es-timation from TRMM passive microwave observations. J. Appl.Meteor., 40, 1367–1380.

Inoue, T., and K. Aonashi, 2000: A comparison of cloud and rainfallinformation from instantaneous visible and infrared scanner andprecipitation radar observations over a frontal zone in East Asiaduring June 1998. J. Appl. Meteor., 39, 2292–2301.

Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplifiedscheme for obtaining precipitation and vertical hydrometeor pro-files from passive microwave sensors. IEEE Trans. Geosci. Re-mote Sens., 34, 1213–1232.

——, W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Trop-ical Rainfall Measuring Mission (TRMM) sensor package. J.Atmos. Oceanic Technol., 15, 808–816.

Lensky, M. I., and D. Rosenfeld, 1997: Estimation of precipitationarea and rain intensity based on the microphysical propertiesretrieved from NOAA AVHRR data. J. Appl. Meteor., 36, 234–242.

——, and ——, 2002: Satellite-based insights into precipitation for-mation processes in continental and maritime convective cloudsat nighttime. J. Appl. Meteor., 42, 1227–1233.

Ricchiazzi, P., S. Yang, C. Gautier, and D. Sowle, 1998: SBDART:A research and teaching software tool for plane parallel radiativetransfer in the earth’s atmosphere. Bull. Amer. Meteor. Soc., 79,2101–2114.

Rosenfeld, D., and G. Gutman, 1994: Retrieving microphysical prop-erties near the tops of potential rain clouds by multispectralanalysis of AVHRR data. J. Atmos. Res., 34, 259–283.