16
The Diurnal Cycle of Clouds and Precipitation along the Sierra Madre Occidental Observed during NAME-2004: Implications for Warm Season Precipitation Estimation in Complex Terrain STEPHEN W. NESBITT Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois DAVID J. GOCHIS Research Applications Program, National Center for Atmospheric Research,* Boulder, Colorado TIMOTHY J. LANG Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado (Manuscript received 15 June 2007, in final form 4 February 2008) ABSTRACT This study examines the spatial and temporal variability in the diurnal cycle of clouds and precipitation tied to topography within the North American Monsoon Experiment (NAME) tier-I domain during the 2004 NAME enhanced observing period (EOP, July–August), with a focus on the implications for high- resolution precipitation estimation within the core of the monsoon. Ground-based precipitation retrievals from the NAME Event Rain Gauge Network (NERN) and Colorado State University–National Center for Atmospheric Research (CSU–NCAR) version 2 radar composites over the southern NAME tier-I domain are compared with satellite rainfall estimates from the NOAA Climate Prediction Center Morphing tech- nique (CMORPH) and Precipitation Estimation from Remotely Sensed Information Using Artificial Neu- ral Networks (PERSIANN) operational and Tropical Rainfall Measuring Mission (TRMM) 3B42 research satellite estimates along the western slopes of the Sierra Madre Occidental (SMO). The rainfall estimates are examined alongside hourly images of high-resolution Geostationary Operational Environmental Sat- ellite (GOES) 11-m brightness temperatures. An abrupt shallow to deep convective transition is found over the SMO, with the development of shallow convective systems just before noon on average over the SMO high peaks, with deep convection not developing until after 1500 local time on the SMO western slopes. This transition is shown to be contem- poraneous with a relative underestimation (overestimation) of precipitation during the period of shallow (deep) convection from both IR and microwave precipitation algorithms due to changes in the depth and vigor of shallow clouds and mixed-phase cloud depths. This characteristic life cycle in cloud structure and microphysics has important implications for ice-scattering microwave and infrared precipitation estimates, and thus hydrological applications using high-resolution precipitation data, as well as the study of the dynamics of convective systems in complex terrain. 1. Introduction Improving the basic understanding and model repre- sentation of warm season precipitation processes within the North American monsoon system (NAMS) is a pri- mary goal of the North American Monsoon Experi- ment (NAME; Higgins et al. 2006). Depending on lo- cation within the NAMS, summer rainfall in the mon- soon region provides between 40% and 80% of the annual water resource (Douglas et al. 1993; Adams and Comrie 1997; Brito-Castillo et al. 2003; Gochis et al. 2006) and, therefore, there is a pressing societal need for improved estimates of rainfall for water resources decision making and hydrological prediction activities. Previous work has shown that a primary mode of vari- ability of precipitation within the core monsoon region * The National Center for Atmospheric Research is sponsored by the National Science Foundation. Corresponding author address: Prof. Stephen W. Nesbitt, Dept. of Atmospheric Sciences, University of Illinois at Urbana– Champaign, 105 S. Gregory St., Urbana, IL 61801-3070. E-mail: [email protected] 728 JOURNAL OF HYDROMETEOROLOGY VOLUME 9 DOI: 10.1175/2008JHM939.1 © 2008 American Meteorological Society

PDF (2978 KB)

Embed Size (px)

Citation preview

Page 1: PDF (2978 KB)

The Diurnal Cycle of Clouds and Precipitation along the Sierra Madre OccidentalObserved during NAME-2004: Implications for Warm Season Precipitation Estimation

in Complex Terrain

STEPHEN W. NESBITT

Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois

DAVID J. GOCHIS

Research Applications Program, National Center for Atmospheric Research,* Boulder, Colorado

TIMOTHY J. LANG

Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

(Manuscript received 15 June 2007, in final form 4 February 2008)

ABSTRACT

This study examines the spatial and temporal variability in the diurnal cycle of clouds and precipitationtied to topography within the North American Monsoon Experiment (NAME) tier-I domain during the2004 NAME enhanced observing period (EOP, July–August), with a focus on the implications for high-resolution precipitation estimation within the core of the monsoon. Ground-based precipitation retrievalsfrom the NAME Event Rain Gauge Network (NERN) and Colorado State University–National Center forAtmospheric Research (CSU–NCAR) version 2 radar composites over the southern NAME tier-I domainare compared with satellite rainfall estimates from the NOAA Climate Prediction Center Morphing tech-nique (CMORPH) and Precipitation Estimation from Remotely Sensed Information Using Artificial Neu-ral Networks (PERSIANN) operational and Tropical Rainfall Measuring Mission (TRMM) 3B42 researchsatellite estimates along the western slopes of the Sierra Madre Occidental (SMO). The rainfall estimatesare examined alongside hourly images of high-resolution Geostationary Operational Environmental Sat-ellite (GOES) 11-�m brightness temperatures.

An abrupt shallow to deep convective transition is found over the SMO, with the development of shallowconvective systems just before noon on average over the SMO high peaks, with deep convection notdeveloping until after 1500 local time on the SMO western slopes. This transition is shown to be contem-poraneous with a relative underestimation (overestimation) of precipitation during the period of shallow(deep) convection from both IR and microwave precipitation algorithms due to changes in the depth andvigor of shallow clouds and mixed-phase cloud depths. This characteristic life cycle in cloud structure andmicrophysics has important implications for ice-scattering microwave and infrared precipitation estimates,and thus hydrological applications using high-resolution precipitation data, as well as the study of thedynamics of convective systems in complex terrain.

1. Introduction

Improving the basic understanding and model repre-sentation of warm season precipitation processes within

the North American monsoon system (NAMS) is a pri-mary goal of the North American Monsoon Experi-ment (NAME; Higgins et al. 2006). Depending on lo-cation within the NAMS, summer rainfall in the mon-soon region provides between 40% and 80% of theannual water resource (Douglas et al. 1993; Adams andComrie 1997; Brito-Castillo et al. 2003; Gochis et al.2006) and, therefore, there is a pressing societal needfor improved estimates of rainfall for water resourcesdecision making and hydrological prediction activities.Previous work has shown that a primary mode of vari-ability of precipitation within the core monsoon region

* The National Center for Atmospheric Research is sponsoredby the National Science Foundation.

Corresponding author address: Prof. Stephen W. Nesbitt, Dept.of Atmospheric Sciences, University of Illinois at Urbana–Champaign, 105 S. Gregory St., Urbana, IL 61801-3070.E-mail: [email protected]

728 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

DOI: 10.1175/2008JHM939.1

© 2008 American Meteorological Society

JHM939

Page 2: PDF (2978 KB)

exists on diurnal time scales (Negri et al. 1994; Gochiset al. 2003, 2004, 2007; Janowiak et al. 2005; Lang et al.2007). However, the underlying mechanisms respon-sible for the diurnal variability in precipitation are oftenpoorly represented in regional- and global-scale modelssimulating the NAMS (Gutzler et al. 2005). Sinceweather systems that modulate the intraseasonal vari-ability of the monsoon (e.g., “gulf surges”; Bordoni andStevens 2006; Bordoni et al. 2004; Stensrud et al. 1997)as well as moisture (Berbery 2001) and potential vor-ticity transports (Saleeby and Cotton 2004) influenceconvective processes across many scales, understandingthe various modes and processes responsible for diur-nally driven convective precipitation are critical to im-proving models and, thus, improving predictions of theNAMS.

The Sierra Madre Occidental (SMO) dominates theregional geography of the NAMS, extending from sealevel to over 3000 m in elevation in several locationsnear the west coast of Mexico, which contains deepcanyon systems, steep slopes, and high plateaus (Fig. 1)that greatly impact low- to midlevel atmospheric flowfeatures in the region (Johnson et al. 2007). The SMOproduces diurnally modulated upslope (downslope)

flows during the day (night), thus exerting control onmoisture transport and convergence along the SMOmountain block (Berbery 2001). Given these factors,the initiation of convection is favored earlier in the di-urnal cycle over high mountain peaks relative to theslopes or surrounding plains (Gochis et al. 2004), whiletotal precipitation is maximized along the SMO westernmidslopes (see Fig. 1). The vertical structure and mi-crophysical characteristics of midday developing con-vection over high terrain are of great interest since littleis known about the details of the precipitation fre-quency and intensity characteristics in the terrain of theSMO.

Once convection initiates in the high terrain of theSMO, it has been shown to occasionally grow upscaleinto mesoscale convective systems (MCSs) that propa-gate toward the Gulf of California within the time-mean easterly flow (Farfan and Zehnder 1994; Jano-wiak et al. 2005; Lang et al. 2007). However, Lang et al.(2007) showed that there is significant intraseasonalvariability in the degree to which convection organizesalong the SMO and this variability influences precipi-tation production by convective systems of varyingtypes within the diurnal cycle (Nesbitt and Zipser2003). These important life cycle processes by whichconvection is initiated and grows upscale along theSMO must be better understood in order to be properlyrepresented in numerical prediction models. The mor-phology of cloud and microphysical structures mustalso be well understood in order to develop confidentestimates of precipitation from remote sensing obser-vation platforms. These needs underline the impor-tance of a multiscale approach to studying monsoonconvective systems on diurnal to intraseasonal timescales.

Since the operational precipitation observing net-work in northwest Mexico is not extensive or spatiallydense, satellite observations are potentially a desirabletool to monitor the occurrence and intensity structuresin monsoon precipitation. However, biases in satelliteprecipitation estimates are only beginning to be ex-plored in this region (e.g., Janowiak et al. 2005; Gebre-michael et al. 2007; Hong et al. 2007), and are likely tobe quite sensitive to the vertical structure of the con-vection (Nesbitt et al. 2004; Fiorino and Smith 2006;Liu et al. 2007). Infrared-based techniques rely on therelationship between cloud-top temperature and sur-face rainfall (Xie and Arkin 1997; Sorooshian et al.2000), while overland passive microwave techniquescurrently rely on empirical relationships between the85–89-GHz ice-scattering optical depth and the surfaceprecipitation (Ferraro 1997). Both of these techniquessuffer when precipitation is shallow because the cloud-

FIG. 1. (a) Map showing the region of study, including terrainelevation (shaded, m), the location of ground-based scanning ra-dars (�), and 210-km range rings from each radar, as well as thelocations of NERN gauges (�). Mean 1998–2004 3B42 July–August precipitation is contoured with the thick lines. (b) Mapshowing the locations of the 0.25° grids used in this study, shadedby elevation band.

AUGUST 2008 N E S B I T T E T A L . 729

Page 3: PDF (2978 KB)

top temperature is often too warm and there is oftenlittle or no cloud or precipitation ice to scatter radiationat these microwave frequencies.

Preliminary analyses show that convective clouds inthe NAME region may have vertically varying struc-tures as functions of topography (Hong et al. 2007;Rowe et al. 2008). This variability is hypothesized tocause biases in satellite-based rainfall estimates in theregion. Therefore, characterization of uncertainties insatellite rainfall estimates within NAME may rely inlarge part on an understanding of the vertical profile ofprecipitation and cloud structures as functions of topog-raphy. Such an understanding is particularly importantin the application of remotely sensed precipitation es-timates for hydrological prediction in a region wherestrong gradients in precipitation character drive largedifferences in the surface hydrological response (Vivoniet al. 2007). In this study, we will examine the diurnallyevolving, vertical structure of clouds and other signa-tures of convection in an attempt to characterize theprocesses that control the diurnal cycle of clouds andprecipitation behavior along the western slopes of theSMO.

This paper is organized in the following manner. Insection 2, the data sources and data reduction are out-lined. In section 3, we will discuss the differences inobserved precipitation rates and frequencies in the re-gion of interest. Sections 4 and 5 examine the varyingconvective vertical structures and microphysics as afunction of elevation in the SMO, followed by discus-sion and conclusions in sections 6 and 7, respectively.

2. Data

a. Ground-based precipitation measurements

Surface gauge measurements at 15-min time resolu-tion were obtained from the NAME Event Rain GaugeNetwork (NERN) from the National Center for Atmo-spheric Research/Earth Observing Laboratory(NCAR/EOL) during the period of the NAME EOP (1July–1 August 2004). Maintenance, calibration, anddata processing details of the NERN network are pro-vided by Gochis et al. (2003, 2004, 2007). During the2004 NAME EOP, the NERN consisted of 86 single,tipping-bucket-type gauges distributed across six topo-graphic transects shown in Fig. 1. Three scanning pre-cipitation radars were operated during the period 7July–21 August 2004. These included two Servicio Me-teorológicio Nacional (SMN) C-band operationalDoppler radars located at Guasave and near Cabo SanLucas, Mexico, as well as the NCAR S-band dual-polarization Doppler radar (S-Pol) near La Cruz deElota, Mexico (see Fig. 1 for the locations and nominal

coverage areas of the radar data). Reflectivity datafrom these three radars were quality controlled andcomposited every 15 min (when available), and reflec-tivity and rain-rate estimates were interpolated to a0.05° latitude–longitude grid by Colorado State Univer-sity (CSU) and NCAR (Lang et al. 2007). This studyuses version 2 of the CSU composites (Rowe et al.2008), which feature improved corrections for beamblockage, hail contamination (using a 250 mm h�1 cut-off), and the use of a polarimetrically tuned Z–R rela-tionship (Z � 133 R1.5) obtained following the meth-odology of Bringi et al. (2004). In the radar data, wehave only analyzed locations in a particular hour withmore than 90 accumulated scans over the entire experi-ment period. This limitation helps remove points thatwere only scanned intermittently (i.e., those points oc-curring only at far ranges due to changes in the maxi-mum range set for particular radars during the experi-ment).

b. Satellite data

Gridded 3-hourly 0.25° horizontal resolution satelliterainfall estimates were obtained from three sources.The first is the National Oceanic and Atmospheric Ad-ministration (NOAA) Climate Prediction Center Mor-phing technique (CMORPH), which advects precipita-tion estimates from the Special Sensor Microwave Im-ager (SSM/I), the Advanced Microwave SoundingUnit-B (AMSU-B), and the Tropical Rainfall Measur-ing Mission (TRMM) passive microwave sensors intime using higher temporal resolution geostationaryinfrared (IR) data (Joyce et al. 2004). The secondsource is the Precipitation Estimation from RemotelySensed Information Using Artificial Neural Networks(PERSIANN), which uses neural network classifica-tion/approximation procedures to compute an estimateof rainfall rate using geostationary IR brightness tem-peratures. The algorithm uses TRMM 2A12 passive mi-crowave rainfall estimates as input into the neural net-work (Hsu et al. 1997; Sorooshian et al. 2000; Hong etal. 2005). Note that this study uses the operational ver-sion of PERSIANN, while a related study by Hong etal. (2007) uses a parallel version of PERSIANN with acloud classification scheme (PERSIANN-CCS) that isdescribed as using the measured IR brightness tem-perature characteristics to select a more appropriatemicrowave rain-rate estimate (Hong et al. 2004). Thethird source is TRMM 3B42 version 6 (hereafter 3B42),which calibrates IR precipitation estimates withTRMM 2B31 combined radar and microwave estimatesof precipitation, as well as other passive microwave sat-ellites; monthly accumulations are adjusted to gauge

730 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Page 4: PDF (2978 KB)

accumulations over land (Huffman et al. 2001; Huffmanet al. 2007).

To examine the vertical structure of convection as afunction of the diurnal cycle, 4-km-gridded, half-hourly,globally stitched IR brightness temperatures from geo-stationary satellites (Janowiak et al. 2001) were ob-tained from the NOAA/Climate Prediction Center viathe National Aeronautics and Space Administration(NASA) Goddard Space Flight Center Data and Ac-quisition Center for the period 1 July–31 August 2004.These data were linearly interpolated to the 5-km gridof the CSU–NCAR radar composites described above.

3. Comparison of precipitation estimates

a. Comparison methodology

The first objective of this study is to compare theestimated diurnal cycles of precipitation frequency andintensity from the various precipitation products in theregion. Each dataset was placed onto a common 0.25°� 0.25° grid at 3-hourly time resolution for the report-ing period 0000 UTC 1 July–2100 UTC 31 August 2004.Note that the period of accumulation corresponding toeach reporting time varies among the satellite esti-mates. For a reporting time of 1200 UTC, for example,3B42 reports estimated precipitation for the period0930–1130 UTC, while CMORPH and PERSIANN re-port precipitation from 1200 to 1500 UTC. Thesedatasets were kept in their native time resolution toavoid temporal interpolation errors, while radar andgauge data were interpolated to the 3B42 three-hourtime bins. Radar (0.05°) and gauge (point) data wereupscaled via linear interpolation (in boxes with at leastone radar or gauge observation) to the same 0.25° grid.Once the precipitation estimates were combined, eleva-tion data from the U.S. Geological Survey’s (USGS)Global 30 Arc Second Elevation Data Set (GTOPO30;at �1 km horizontal resolution) were linearly interpo-lated to the same 0.25° grid, and overland points werebinned into three elevation bands: below 500 m (coastalplain), 500–2250 m (foothills), and above 2250 m (highpeaks). Rain rates and precipitation frequencies werecalculated based on accumulations or the number of3-hourly periods with measurable precipitation dividedby the total number of observations, respectively (in-cluding rain and no-rain periods).

Well-known uncertainties are present when compar-ing precipitation measurements from different plat-forms (e.g., from gauges, radars, and satellite estimates)due to spatial and temporal sampling differences, andretrieval errors (Tustison et al. 2001, 2003; Steiner andSmith 2004). In the results herein, the quantifications ofthese uncertainties are left to future work as 1) the

observation network was not designed to fully quantifythe subpixel variability in satellite and radar measure-ments, including variations on very small scales in to-pography (Gembremichael et al. 2007), and 2) we areunaware of any standard method that exists to quantifysuch uncertainty. Both of these facts motivate futurehigh-resolution gauge and disdrometer measurementsto examine these uncertainties. Recent work by Roweet al. (2008) indicates that the statistical distribution ofrain rates is similar between the NERN gauge measure-ments and the NAME radar composites; this arguesthat the spatial character of the sampling uncertaintyfrom these measurements to a satellite pixel scale maybe similar. Given these uncertainties, and the fact thatwe have a much more extensive observation networkthan operational precipitation analyses in NW Mexico,it is felt that the length and number of observationscollected in NAME are able to capture the bulk differ-ences in the precipitation characteristics we aim to de-pict.

b. Precipitation estimate comparisons as a functionof elevation

Figures 2 and 3 show the total time series and meandiurnal cycles, respectively, of the estimated rainfall ac-cumulations from the five precipitation datasets. Fromthe rainfall time series presented in Fig. 2, it is apparentthat significant precipitation occurs nearly every daysomewhere within the analysis domain, and there is astrong diurnal component to the variability in all of theprecipitation time series. There is significant day-to-dayvariability in the amplitude of the diurnal cycle of therain rates, particularly at low elevations. All estimatestend to show modest increasing values of maximumprecipitation with decreasing elevation, as shown pre-viously by Gochis et al. (2007).

Comparisons of mean diurnal precipitation rates areshown in Fig. 3a. (All times have been adjusted tomountain standard time, e.g., UTC � 7 h.) These diur-nal averages were then subtracted from the time seriesshown in Fig. 2 for each day, and the ratio of the vari-ance of the resultant time series (with the mean diurnalcycle removed) to the original time series is calculated,and then subtracted from 1. This ratio represents thefraction of variance explained by the diurnal time serieswhose values are provided in Table 1. Along with aqualitative inspection of Fig. 2, the values of the frac-tional variance show that the diurnal cycle is indeedmore regular at the higher elevations of the SMO (inboth phase and magnitude) compared with the coastalplain region. Precipitation at lower elevations is moresubject to a higher degree of intraseasonal variability aswell as extreme events compared with higher eleva-

AUGUST 2008 N E S B I T T E T A L . 731

Page 5: PDF (2978 KB)

tions. These results are consistent with those of Gochiset al. (2004) and Lang et al. (2007), the latter studyshowing that the late night/early morning maximum inprecipitation along the coastal plain is primarily due to

organized convective systems (e.g., MCSs) that propa-gate westward off of the SMO, or propagate northwest-ward from their region of origin near Puerto Vallarta.

Figure 3 presents a similar result, displaying the

FIG. 3. Time series of (a) mean diurnal precipitation rate [mm day�1 (3 h)�1] and (b) mean diurnal precipitationfrequency for the period 0000 UTC 1 Jun–21 UTC 31 Aug 2004 for the five precipitation estimates for 0.25° gridpoints (top) above 2250 m, (middle) between 500 and 2250 m, and (bottom) below 500 m.

FIG. 2. Time series of mean precipitation rate [mm day�1 (3 h)�1] for the period 0000 UTC 1 Jun–2100 UTC 31 Aug 2004 for thefive precipitation estimates for 0.25° grid points (top) above 2250 m, (middle) between 500 and 2250 m, and (bottom) below 500 m.

732 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Fig 2 live 4/C

Page 6: PDF (2978 KB)

phase-lagging of the estimated diurnal precipitationrate (panel a) and the frequency maximum (panel b) aselevation decreases (moving from top to bottom pan-els). In comparing the precipitation datasets, there aredifferences in the estimated amplitude and timing ofthe diurnal cycle in this region. At high elevations (Fig.3a, top panel), the CMORPH and radar data have aprecipitation rate and frequency peak in the early af-ternoon, with corresponding higher magnitudes in pre-cipitation rate. However, peak rain rates in the NERN,PERSIANN, and 3B42 products occur in the late after-noon. Over the middle elevation bin (Fig. 3a, middlepanel), the timing of the precipitation rate peak centerson 1700 local time (LT) and is similar in phase amongall of the datasets. The NERN and 3B42 products ex-hibit closely agreeing precipitation rates throughout theday while CMORPH, PERSIANN, and radar estimatesproduce significantly higher precipitation rates. Overthe coastal plain (Fig. 3a, lower panel), PERSIANNand CMORPH have much higher mean and peak diur-nal amplitudes of the estimated precipitation rate com-pared to the radar, 3B42, and NERN products.CMORPH also tends to estimate the hour of peak rainrate one bin earlier (1700 LT) than the other four prod-ucts (2000 LT).

c. Understanding radar–gauge discrepancies

To investigate variations in the estimated precipita-tion frequency from the radar composites and NERN(aside from the satellite estimates), we will examinethese datasets at higher resolution to bypass possibleerrors due to upscaling these natively high-resolutionestimates to 0.25° grids. Figure 4 shows maps depictingthe spatial pattern of the hourly frequency of occur-rence (expressed in %) of the nonzero precipitationfrom the original radar composite 5-km grid (shaded)and at each NERN gauge site (shaded circles) forhourly intervals between 1100 and 1600 LT. At highelevations, precipitation frequencies before noon aresmaller than later in the diurnal cycle. Note that theradar composites depict precipitation initiating and oc-curring more than 16% of the time over selected terrainfeatures, while only one gauge (near 26.2°N, 106.5°W)estimates precipitation more than 16% of the time.

Overall, absolute differences are small between theseradar and gauge estimates, and this likely explains whythe precipitation frequencies in the 1100 LT bin in Fig.3 over high elevations are small. Later, in the diurnalcycle over high terrain, precipitation frequencies arecomparable (Fig. 3b), with a slight tendency for theNERN gauges to have higher precipitation frequencies.This can also be seen in the 1500 and 1600 LT panels ofFig. 4. Precipitation ends for the most part by 2300 LT(not shown) in the high terrain, although radar-estimated rainfall diminishes before the NERN-estimated rainfall.

1) ISSUES IN THE SMO HIGH TERRAIN

Disparities between the radar and gauge estimates athigh elevation may be explained, in part, by the radarbeam dimension as a function of terrain geometry. Inthe high peaks of the SMO, despite the radar beambeing 150–200 km from the radar, the elevation of thebeam center is not more than 1–2 km above the moun-tain peaks (assuming an elevation angle of 0.8°, as wasused for S-Pol rain mapping; the elevation angle for theGuasave radar varied between 0.5° and 1.5°). This factallows the radar to detect relatively shallow precipita-tion over the high terrain compared to if the terrainwere lower in elevation for a given range to target.However, at 200-km range, the 1.0° (1.4°) beam widthof S-Pol (Guasave) is more than 3.2 (4.5) km wide.Provided that precipitation early in the diurnal cycle islikely isolated in nature (i.e., not well organized), therather large size of the radar beam would tend to smearthe echo over large distances, biasing radar-estimatedprecipitation frequencies high (but conditional rainrates low due to echo smearing) relative to gauges. Bi-ases in inverting reflectivity to rain rate, brightbandcontamination, and beam altitude above the freezinglevel all contribute to the uncertainty in estimating pre-cipitation from radars and comparing radars and raingauges at far ranges (Smith and Krajewski 1991).

2) ISSUES IN THE SMO FOOTHILLS

Over the midslopes of the SMO, radar-estimatedprecipitation frequencies and intensities are nearlytwice those of the NERN gauge estimates. One factor

TABLE 1. For each precipitation estimate, the fraction of variance explained by the mean diurnal cycle in each elevation band isshown.

Band NERN Radar CMORPH PERSIANN 3B42

Elevation �2250 m 0.37 0.32 0.36 0.31 0.38500 m � elevation �2250 m 0.32 0.13 0.27 0.29 0.31Elevation �500 m 0.22 0.18 0.16 0.23 0.15

AUGUST 2008 N E S B I T T E T A L . 733

Page 7: PDF (2978 KB)

that may contribute to this relative difference betweenthe radar and gauges in the foothills is the fact thatthere are comparatively few gauges in these regionscovered by the area where precipitation is shown to bemost frequent by the radars, say at 1600 LT. This wouldtend to cause the gauges to underreport the precipita-tion frequency (and amount) as represented by thelarger domain covered by the radars. This factor prob-ably combines with radar range effects (resolution andbeam elevation above the terrain), leading to higherfrequencies of precipitation over these regions. Com-bining the large vertical distance between the lowest-

elevation radar beam and the surface in this region andthe potential representativeness of the gauge locationslikely tends to radar estimates of precipitation rate andfrequency being higher than the gauges over the mid-slopes.

3) ISSUES IN THE COASTAL PLAIN

Over the coastal plain, the radars again tend to esti-mate the precipitation frequency as being higher thanfor the NERN gauges. One reason for this may be thatthe locations of the NERN gauges are not ideal forcapturing the sea-breeze convection depicted in the ra-

FIG. 4. Maps of the shaded frequencies of precipitation (expressed as percent) according to the NAME version 2 CSU/NCAR radarcomposites, as well as the frequencies of precipitation according to the NERN gauges (filled circles) over each hour beginning at thetime labeled in each panel. White areas indicate no radar data were collected in this region.

734 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Fig 4 live 4/C

Page 8: PDF (2978 KB)

dar analysis. This behavior is depicted as the observedcoast-parallel band of enhanced precipitation fre-quency just inland from the Gulf of California begin-ning around 1200 LT. Based on the figure, none of theNERN stations appears to be located in the radar-determined preferred location for sea-breeze precipita-tion. In addition, subcloud evaporation may be signifi-cant between the elevation of the lowest radar beamand the surface along the coastal plain, where subcloudrelative humidity tends to be comparatively lower thatat higher elevation (Rosenfeld and Ulbrich 2003). Low-end differences in sensitivity between precipitation es-timates may also contribute. For instance, 0.254 mm ofrain accumulation is required in order to register a tipin NERN gauges, which is equivalent to a radar echo of12 dBZ using the Z–R relationship listed above. Thissensitivity may be important since much of the precipi-tation over the coastal plain in the nighttime hours is inthe form of light stratiform rain from MCSs (Lang et al.2007).

d. Further examination of the satellite estimates

In terms of estimated precipitation frequency, 3B42has a similar time series compared with NERN, whilethe radar, PERSIANN, and CMORPH time series aresomewhat higher in magnitude. Precipitation frequencyestimates over the highest terrain (�2250 m; Fig. 3b,top panel) are quite similar between NERN and radarestimates while radar estimates of precipitation fre-quency are significantly higher than NERN over themid- and low elevations. The radar rainfall rate esti-mates in Fig. 3 tend to overestimate the precipitationaccumulation relative to the NERN gauges at all eleva-tion bands in this region. The fact that precipitationfrequency biases tend to follow in the same pattern asthe precipitation amount for these estimates shows thatthese algorithms tend to agree in terms of conditionalprecipitation rate (since the total precipitation is theproduct of the precipitation frequency and the condi-tional rain rate). At these time and space scales at least,the discrepancies tend to be largely the result of theprecipitation occurrence being different among theseestimates (with the radar, CMORPH, and PERSIANNhaving a larger spatial coverage of precipitation thanNERN and 3B42); however, the scaling properties ofrainfall from the satellite to the gauge scale need to bebetter understood in order to make such comparisonsmore quantitative.

Studies have shown that microwave algorithms suchas used in CMORPH [using the TRMM 2A12, NationalEnvironmental Satellite, Data, and Information Service(NESDIS) AMSU/B, and NESDIS SSM/I algorithms;Joyce et al. (2004)] and PERSIANN (using TRMM

2A12) often overestimate precipitation in deep convec-tive regimes (McCollum et al. 2002; Nesbitt et al. 2004).In this comparison, it appears that a similar positivebias exists in the PERSIANN and CMORPH satelliteproducts relative to the TRMM satellite and NERNgauge products. Issues with the backward interpolationin time may cause the early development of precipita-tion in CMORPH at high elevation, where precipita-tion rapidly develops around 1200 LT (Janowiak et al.2005); however, this hypothesis needs further testing.

Note the agreement between the NERN and 3B42 inboth the estimated precipitation frequency and theamount, particularly at low elevations. The agreementin precipitation amount and frequency between NERNand 3B42 at low elevations is likely due to the fact thatthe version 6 product is gauge adjusted using reportedmonthly accumulations (Huffman et al. 2007); however,the gauge analysis used in 3B42 only contains a fewoperational gauges on the coastal plain. Since noNERN (or other) gauges are used in the adjustment,this likely leads to the relative underestimate of 3B42relative to the other estimates at higher elevations (es-pecially relative to radar along the midslopes). Thenon-gauge-adjusted version of 3B42 (3B42RT) has pre-cipitation amounts that are more than twice those ofthe 3B42 version examined here (not shown), makingits peak afternoon accumulations greater than eitherPERSIANN or CMORPH. For the remainder of thispaper, the characteristics of the cloud systems withinthe diurnal cycle will be investigated to help explain thedisagreements in the precipitation estimates foundabove.

4. Diurnally varying convective vertical structurealong the SMO

a. Statistical evolution of IR-inferred cloud structurealong the SMO

As evidenced by the diurnally averaged precipitationtime series in the previous section, rainfall often beginsfalling on the high-elevation region of the SMO beforelocal noon. However, as shown in Fig. 3, the 3B42 andPERSIANN combined infrared and microwave precipi-tation estimation techniques tend to somewhat under-estimate the precipitation rates compared with theNERN observations during this period. In examining adifferent version of the PERSIANN algorithm (thePERSIAN-Cloud Classification Scheme or -CCS),Hong et al. (2007) hypothesized that the underestima-tion by microwave-adjusted algorithms early in the di-urnal cycle may be due to the shallowness of the pre-cipitating clouds during this time period. Overland mi-crowave retrievals rely on ice-scattering techniques at

AUGUST 2008 N E S B I T T E T A L . 735

Page 9: PDF (2978 KB)

85 GHz (Ferraro 1997; Kummerow et al. 2001); precipi-tation without significant ice (i.e., with no or decreasedmixed-phase microphysical processes occurring) willcause an underestimate of such retrievals. This lack ofsensitivity will at times cause a low bias in the relation-ship between both the IR and the microwave brightnesstemperatures and rainfall.

This leads us to the question: Does shallow precipi-tation exist over these high-elevation regions early in thediurnal cycle and, therefore, lead to an underestimate inprecipitation from the IR and microwave algorithms?To examine this issue, IR brightness temperature im-ages matched to the radar composite grid are examinedto characterize the diurnal variability in the estimatedheight of the clouds in this region. In Fig. 5, the per-centage of time that an IR brightness temperature pixelreaches below 275 K is shaded, while the percentage of

time pixels that reached a threshold of 208 K is con-toured in white. For reference, the 2250-m-elevationcontour is contoured in black. The 275-K threshold isselected to highlight convection reaching close to thefreezing level, while the 208-K threshold is selected tohighlight convection reaching near the tropopause.Note that according to the mean sounding at Mazatlán,calculated from all July and August sounding launchesduring 2004, 275 K corresponds to a mean height of 4.6km, while 208 K occurred at 13.0 km. The time-meantropopause (identified as the mean height at which aminimum temperature was identified) was slightlycolder at 195 K and was located at 15.4-km altitude.

After reaching a minima in cloud cover around 0900LT, the frequency of occurrence of brightness tempera-tures �275 K begins to increase over the highest terrainaround 1100 LT, indicating the development of new

FIG. 5. Maps of the shaded relative frequencies of IR brightness temperatures less than 275 K, as well as the contoured frequenciesof IR brightness temperatures less than 208 K (between 0% and 20% by 4% increments) every 2 h for satellite observations during thehour beginning at the time labeled in each panel. The dark gray and black lines indicate the coastline and 2250-m elevation contours,respectively.

736 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Page 10: PDF (2978 KB)

convection; these clouds rapidly produce radar echoesand precipitation (Fig. 4). This time of initiation corre-sponds well with convective initiation times in the Col-orado Rockies reported by Banta and Schaaf (1987).Note that at 1100 and 1300 LT, no areas contain cloud-top brightness temperatures �208 K for more than 4%of the time, and the same is true at 235 K (not shown),except for one small region near Cerro Mohinora (el-evation 3250 m, near 26°N, 107°W). It is not until 1500LT that deep convection meeting the 208-K thresholdexists in any region for more than 4% of the time, andthis occurs almost exclusively in the SMO westernslopes below 2250-m elevation. The fact that the 208-Kcontours do not encroach on the high terrain (�2250m) indicates that convection rarely is of tropopausedepth at these high elevations, even though it is com-mon for shallow convection to be located in such loca-tions (�64% of the time).

After 1500 LT, near-tropopause depth convection isvery common over the western slopes of the SMO,tending, in most part, to die out by the time it reachesthe Gulf of California coast around local midnight, ex-cept 1) along the northern extent of the analysis domainwhere 208-K frequency contours persist until approxi-mately 0300 LT, and 2) the large region of 208-K fre-quency contours to the south of Mazatlán toward Puer-to Vallarta. Organized deep convection occasionallypersists offshore through the night and into the nextmorning (Lang et al. 2007). Cloudiness trends tend todecrease at both brightness temperature thresholdsthroughout the morning, until the diurnal cycle reini-tiates with new convection again by 1100 LT. Thus,while the SMO-perpendicular diurnal cycle of convec-tion can be largely represented as a symmetric compos-ite, there is significant along-range variability to thediurnal cycle from areas where convection can last intothe night and into the next morning.

To investigate this behavior in more detail, Fig. 6shows the hourly evolution of the diurnal cycle of the

IR brightness temperature. The shading in Fig. 6 indi-cates the relative fraction of the IR brightness tempera-ture counts (only for observations �275 K, to removepossible surface contamination) in 5-K brightness tem-perature bins as a function of each hour local time. Athigh elevations (Fig. 6a), there is a high fractional oc-currence (�0.7) of clouds in the bins with brightnesstemperature warmer than 265 K during two time peri-ods: one centered on 0900 LT and one centered on localnoon. The former maximum (see also Fig. 5; 0700–0900LT) could be due to two factors. Dissipating cirrus an-vils from the previous day’s convection, represented bythe gradual slope of relative frequency contours towardwarmer brightness temperatures in the evening andovernight hours, are consistent with gradually fallingand sublimating ice crystals produced by the previousday’s convection. Xie et al. (2005) noted a “flat tail” tothe IR brightness temperature distribution near theSMO crest, and attributed it to optically thin cirrusclouds. However, at high elevations low-level stratus/ground fog was repeatedly documented on servicingtrips to NERN gauges as well as in Alcantara et al.(2002). Nighttime cloud cover over the high terrain isimportant in determining the radiative fluxes in the re-gion, as well as influencing the strength of the radia-tively driven nocturnal downslope flows (Whiteman2000). Since downslope flows are necessarily dry, noc-turnal downslope flows could exhibit a strong influenceon the moisture availability at high terrain and alongthe SMO slopes, influencing the conditions for the nextday’s convection (Ciesielski and Johnson 2008).

The latter maximum around local noon at high el-evations (Fig. 6a) is the developing shallow convectionalong and just west of the SMO peaks. However, notingthe dearth of deep cloudiness over the high elevationsuntil well after noon, much of the cold cloudiness (�220K) may actually be eastwardly expanding anvils fromconvection, whose the updrafts are actually to the westat lower elevations. In the foothills (Fig. 6, middle

FIG. 6. (a)–(c) Shaded frequency diagrams of the relative fraction of the IR brightness temperature counts (only for observations�275 K) for three different elevation levels in 5-K brightness temperature bins as a function of each hour local time.

AUGUST 2008 N E S B I T T E T A L . 737

Page 11: PDF (2978 KB)

panel), there is evidence of shallow cloud developmentaround noon (with brightness temperatures �265 K).However, the primary feature of note is the high rela-tive fraction of deep cold cloud (with temperatures�210 K) present by midafternoon. Deep cloudiness isthe prominent feature near and just after sunset overlow-elevation regions (Fig. 6, right panel). There is arelative maximum of pixels with cold clouds �200 Kbetween 1800 and 2100 LT, and a transition to warmerbrightness temperatures thereafter, which is consistentwith gradually dissipating cirrus anvils overnight.

b. Case study of diurnal evolution along the SMO

Figure 7 illustrates a case study showing observed IRbrightness temperatures and radar reflectivity compos-ites every 4 h from 1200 LT 2 August 2004 to 1600 LT3 August 2004. By 1200 LT 2 August, isolated convec-tion developed over the high terrain of the SMO, withthe cloud temperatures in this region ranging from 240to 260 K; these are temperatures that are too cold to beassociated with the surface. Note the striated appear-ance of the IR brightness temperatures on the eastern(upwind) side of the SMO, indicating the breakup ofthis warm cloud-top feature (possibly via the develop-ment of convective rolls with solar heating and/or anupslope flow component). These circulations formedperpendicular to the east-southeasterly SMO ridge-parallel flow depicted by the North American Regional

Reanalysis (Mesinger et al. 2006) at this time (notshown).

Radar echoes are apparent along and just west of thehigh SMO terrain by 1200 LT, and the NERN gaugestations at La Ermita (23.67°N, 105.72°W, 2716 m) andLas Rusias (23.74°N, 105.53°W, 2896 m) reported lightprecipitation (�1 mm) during the period 1100–1200LT. During the next 4 h, the convection organized andmoved westward down the western SMO slopes (withIR brightness temperature minima falling in the rangeof 204–220 K). Note that during this afternoon period,convective and stratiform precipitation propagatedwestward, but the cold anvil spread eastward back overthe high terrain (this depicts the cause of late afternooncold cloud tops at high elevations in Fig. 6 even thoughthe precipitation has moved to lower elevations). By0000 LT 3 August, most of the echo that was associatedwith the previous day’s convection had dissipated,and cloud-top temperatures warmed in the remnants ofthis convection over the higher terrain and coastalplain. However, a pair of mesoscale convective systemsthat formed to the southeast of the domain movednorthwest into the domain during the next morning.The aggregate of these types of systems is responsiblefor the enhanced probabilities for cold brightnesstemperatures offshore, near the mouth of the Gulf ofCalifornia, during the early morning hours shown inFig. 5.

FIG. 7. Time series of IR brightness temperatures (shaded, grayscale) and radar reflectivity (shaded, colors) for the period 1200 LT2 Aug–1600 LT 3 Aug 2004. The nominal range of the NAME radars is indicated at the upper left. The arrows indicate the locationsof inferred nocturnal stratus clouds.

738 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Fig 7 live 4/C

Page 12: PDF (2978 KB)

Also, note the depiction of low-level (warm IRbrightness temperature) stratus clouds over the highterrain by 0400 and 0800 LT (indicated with the whitearrows in Fig. 7). These stratus clouds appear to breakup into the roll-type circulations (and associated pre-cipitation) that were evident over the SMO slopes theprevious day at the same time (1200 LT). Despite thepresence of the mesoscale convective system near theGulf of California shoreline on this day, convection atthe higher elevations seemed to appear to form andorganize in a similar fashion on 3 August comparedwith 2 August (e.g., near 24°N, 106°W) apart from areaswhere solar radiation is shaded at the surface by theMCS anvil. Thus, as was commonly observed during thefield campaign, the daytime orographically forced con-vection existed simultaneously with organized, propa-gating convection at lower elevations.

5. Convective environments along the SMO

To interpret the differences in the convective low-level environments forcing cloud systems of differentcloud-top heights along the SMO, data from recordingsurface temperature and humidity sensors deployed byA. Douglas of Creighton University within instrumentshelters at the NERN gauge sites at La Cienega deNuestra Señora near the SMO crest (25.0°N 106.3°W,elevation 2483 m), El Palmito (23.6°N, 105.8°W, eleva-tion 1925 m), Mazatlán (23.2°N 106.4°W elevation 2 m),and Culiacán (24.8°N 107.4°W, elevation 66 m) wereanalyzed. Figure 8 shows the mean diurnal values (athalf-hour increments) of temperature, relative humid-ity, and estimated cloud-base height1 at each site. Tem-peratures at La Cienega and El Palmito are much

cooler than those at Mazatlán and Culiacán due to el-evation effects. As temperatures rapidly rise at LaCienega following sunrise, near-saturated surface con-ditions are quickly mixed out, on average, over a periodof approximately 2 h. At El Palmito (�600 m below),temperatures rise more slowly (from presunrise relativehumidity values exceeding 90%), and over the coastalplain, diurnal heating occurs much more gradually (andmean nighttime relative humidities are less than 90%).The observed trend of the slowing of the diurnal warm-ing with decreasing elevation could be due to factorssuch as shortwave or longwave radiative effects of in-creasing boundary layer specific humidity and evapo-transpiration (Durre and Wallace 2001), or cool sea-breeze flows, particularly at Mazatlán (Ciesielski andJohnson 2008). Note that the mean diurnal maximumsurface mixing ratio at La Cienega is 12.5 g kg�1, whileat Mazatlán it nears 20.3 g kg�1. Afternoon minimumrelative humidity values range along the coastal plainfrom roughly 70% at Mazatlán (very close to the Gulfof California) to less than 50% at Culiacán (in the cen-tral coastal plain). At the time when precipitating con-vection starts over the elevated terrain (1100 LT), esti-mated cloud-base heights are less than 1.0 km aboveground level (AGL). At Culiacán, they extend over 2km AGL at the time of convective initiation.

These environmental conditions have several key im-pacts on precipitating cumulus convection in the SMO.First, the efficiency of the warm rain processes in theconvection that occurs at high elevations, with cloudbase near the surface, will have reduced the evapora-tion and thus increased the precipitation efficiencycompared to low-elevation regions (Fig. 8b). Over thecoastal plain and foothills, it is presumed (based on ourlimited measurements) that significant evaporation ex-ists given that the mean afternoon relative humiditiesare near 50%.

1 Cloud-base height (km) is calculated from Stull (1995) asCBH � 0.125(T � Td).

FIG. 8. For surface stations at Ceniega de Nuestra Señora, El Palmito, Mazatlán, and Culiacán, the mean diurnaltime series of (a) temperature (°C, solid), relative humidity (%, dashed), and (b) estimated cloud-base height (km).

AUGUST 2008 N E S B I T T E T A L . 739

Page 13: PDF (2978 KB)

Second, convection over the high terrain is shown tobe comparatively shallow (i.e., it is rarely of tropopausedepth) with minimum cloud-top IR brightness tem-peratures of 240–220 K (Fig. 6a), and CMORPH andPERSIANN satellite estimates have a delayed precipi-tation onset and less of an overall positive fractionalbias compared with lower elevations (Hong et al. 2007).The lack of moisture available at cloud base in the highterrain relative to lower elevations (almost one-half themoisture in terms of specific humidity) likely limits theobserved maximum height of convection since signifi-cantly less convective available potential energy(CAPE) and moisture (i.e., supercooled water) areavailable for latent heat release in convection. We hy-pothesize that for these reasons, the lack of moisturelimits the vertical extent of the convection at high el-evations, and leads to a low bias in IR rainfall retrievalsin such locations. This effect will also reduce, on aver-age, the depth and intensity of the mixed-phase micro-physical processes, reducing the passive microwave icescattering, and leading to an underdetection and lowbias from the microwave precipitation retrievals in el-evated terrain where IR retrievals have a low bias.

Over the coastal plain, high adiabatic liquid watercontents are hypothesized to lead to very intense tropo-pause-penetrating convective updrafts (Figs. 5 and 6)that produce copious precipitation-sized ice particlesaloft, and strong depressions in IR and microwavebrightness temperatures. Since the low levels of the at-mosphere along the coastal plain typically have lowerrelative humidities than do the higher elevations (Fig.8), subcloud evaporation is thought to decrease surfacerainfall as a function of microwave ice scattering. Inaddition, warm rain processes may be very efficient(despite mean cloud bases of 1.25–2 km) along thecoastal plain due to the large vertical distance between-the cloud base and the freezing level where warm mi-crophysical growth can occur (Rowe et al. 2008). All ofthese factors are hypothesized to lead to positive biasesin current infrared and microwave precipitation re-trieval techniques.

6. Summary

A conceptual figure illustrating the hypothesizedmonsoon-averaged diurnal convective regime along theSMO is shown in Fig. 9. By local noon, relatively shal-low and rapidly precipitating convection is triggerednear the high peaks of the SMO. This convection has itscloud base near the SMO peaks and has relatively littlemoisture available at its disposal (Fig. 8b). Whilepropagating westward during the early afternoon, theconvection remains shallow for 3–4 h before it extends

FIG. 9. Schematic of observed diurnal mechanisms along theSMO at 25°N. Cloud type indicates relative height attained byclouds, shading indicates specific (not relative) humidity contrasts,asterisks (*) indicate mixed-phase microphysical processes, andthe density of the vertical streaks indicates the locations and rela-tive intensities of the precipitation.

740 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Page 14: PDF (2978 KB)

to tropopause depth in the SMO foothills by 1500 LT.It is hypothesized that this time delay in the onset ofdeep convection along the SMO corresponds with thetime within the diurnal cycle that it takes the convec-tion to propagate downslope to locations with sufficientmoisture—and thus convective available potential en-ergy (CAPE)—to penetrate the tropopause. Shallowconvection at high elevations will have warm IR bright-ness temperatures, which are less likely to trigger sig-nificant precipitation in infrared-based rainfall esti-mates, since these methods either use hard thresholds(e.g., the GOES Precipitation Index; Arkin and Meis-ner 1987), derived linear relationships between surfaceprecipitation accumulation and IR brightness tempera-tures [e.g., as in the 3B42 algorithm; Huffman et al.(2001), (2007)], or neural network matchups betweenmicrowave-estimated precipitation and IR brightnesstemperatures as is used by PERSIANN (Hsu et al.1997; Sorooshian et al. 2000). Compared with thedeeper convection observed later in the day, shallowconvection located in high terrain will more likely

1) have weaker mean updraft speeds that are less ableto penetrate through the depth of the troposphere;

2) produce a shallower depth of precipitation-sized ice,or exist completely below the freezing level and thusconstitute a complete warm rain process that willscatter proportionally less 85-GHz microwave radia-tion out of the footprint;

3) have lower total liquid water contents availablesince the lifted parcel is originating at a high altitude(�2250 m), which in turn will limit the latent heatrelease and the vigor and/or depth of the mixed-phase microphysical processes; and

4) be of a smaller horizontal scale (i.e., less organized)as a result of weaker convective circulations andweaker cold pool dynamics, as well as increased en-trainment. These processes will likely increase non-uniform beamfilling effects (Harris et al. 2003; Kum-merow 1998) within the footprint of a microwavesensor [5 km � 7 km in the case of the TRMMMicrowave Imager (TMI)].

These factors likely lead to less ice scattering at 85 GHzfor a given surface rain rate and, perhaps, lead to de-creased detectability of precipitation by current micro-wave methods over the high terrain of the SMO. Com-bined with IR underdetection of surface precipitationin shallow cloud regimes, and even potential space-borne radar beamfilling, ground clutter, and sidelobecontamination in such clouds, this underdetection prob-lem likely compounds problems in remote sensing re-trievals, resulting in an estimate of delayed onset and

reduced intensities of precipitating systems over com-plex terrain.

Large differences in the total precipitation amountexist among the estimates examined in these studies,and separating biases from sampling and representa-tiveness errors requires the development of improvedquantitative means of comparing precipitation mea-surements of different spatial and temporal resolutions(Tustison et al. 2001, 2003). Uncertainties in the scalingproperties of rainfall will have significant implicationsin using spaceborne precipitation estimates as inputinto hydrologic prediction models. Understanding theuncertainties involved in determining the scaling prop-erties of rainfall, both from observational and theoret-ical approaches, should be of key importance, since it iscurrently a scientific barrier in comparing rainfall data.Observing the properties of rainfall (rain rate, particlesize distribution) at many sites at subsatellite and radarpixel time and space scales would be one approach toaddressing such importance scale issues.

A key question for future research involves deter-mining the processes by which the shallow to deep,organized diurnal convective transition occurs alongthe SMO, including the roles of the boundary layerstructure and the moisture transport within topographicflows. The organizational mechanism for these systemsalong the SMO slopes has yet to be attributed, but itscharacteristics appear to be similar to a slope-flowMCS-generation mechanism (Tripoli and Cotton1989a,b). These rainfall events have significantly higherintraseasonal variability than do events in the foothills,which is likely due to variations in environmental flow;also, moisture availability, convective instability, and“gulf surges” (Stensrud et al. 1997) need to be exam-ined in close context with the life cycle of orogenicallyforced MCSs in the core monsoon region.

In addition, stratus and fog are commonly observedovernight in the high terrain, which likely plays an im-portant role in the radiation balance over the SMO. Wehypothesize that nocturnal clouds could limit radia-tional cooling and potentially limit nocturnal down-slope flows, allowing moisture to persist along theslopes rather than being swept to lower elevations over-night by such flows and remaining to invigorate thenext day’s convection once the stratus deck is broken.Evaporation of canopy and surface soil moisture mayalso provide accessible reserves for feeding boundarylayer moisture the following day (Durre and Wallace2001). Given the relatively complex evolution of thediurnal cycle of convection, its forcing, and its verticalstructure, it is not surprising that convection-param-eterizing models representing these process have sig-nificant difficulty representing the timing, intensity,

AUGUST 2008 N E S B I T T E T A L . 741

Page 15: PDF (2978 KB)

and structure of the precipitation and surface hydrologyin the region (Gochis et al. 2002, 2003; Gutzler et al.2005). Therefore, the continued collection and analysisof meteorological data (including state parameter andcloud microphysical data) alongside land surfacedatasets will be essential to improving our understand-ing of warm season convection initiation and evolutionin complex terrain.

Acknowledgments. Thanks go to Rob Cifelli, SteveRutledge, Dave Ahijevych, Rit Carbone, Paul Ciesiel-ski, Bob Maddox, Art Douglas, and Bob Joyce for fruit-ful scientific discussions. The comments of four anony-mous reviewers were very helpful. We wish to thank thestaff of the NCAR/Earth Observing Laboratory, Boul-der, Colorado for maintaining, operating, and collect-ing data from the NAME radars, and ChristopherWatts, Jaime Garatuza-Payán, Julio César-Rodríguez,and the Comisión Nacional del Agua of Mexico forcontinued collection and maintenance of the NERNstations. NERN, radiosonde, radar, surface, and light-ning data were obtained from the NAME data archiveat NCAR/EOL (information online at http://www.eol.ucar.edu). The IR satellite data were obtainedfrom the NOAA/Climate Diagnostic Center, CampSprings, Maryland, via the NASA Goddard Earth Sci-ences Data and Information Services Center, Green-belt, Maryland (information online at http://daac.gsfc.nasa.gov). Data analysis, NERN data collec-tion and analysis, data collection from the SMN radars,and the creation of the radar composites were providedby the NOAA/Climate Prediction Program for theAmericas (CPPA) under Drs. Jin Huang and AnnaritaMariott i (Grants NA07OAR4310214 to SN,NA04OAR4310166 to DG, and NA04OAR4310153 toTL); the deployment of the S-Pol radar was funded byNSF Grant ATM-0340544 to CSU and NCAR.

REFERENCES

Adams, D. K., and A. C. Comrie, 1997: The North Americanmonsoon. Bull. Amer. Meteor. Soc., 78, 2197–2213.

Alcantara, A. O., I. Luna, and A. Velaquez, 2002: Altitudinaldistribution patterns of Mexican cloud forests based uponpreferential characteristic genera. Plant Ecol., 161, 167–174.

Arkin, P. A., and B. N. Meisner, 1987: The relationship betweenlarge-scale convective rainfall and cold cloud over the West-ern Hemisphere during 1982–84. Mon. Wea. Rev., 115, 51–74.

Banta, R. M., and C. B. Schaaf, 1987: Thunderstorm genesis zonesin the Colorado Rocky Mountains as determined by trace-back of geosynchronous satellite images. Mon. Wea. Rev.,115, 463–476.

Berbery, E. H., 2001: Mesoscale moisture analysis of the NorthAmerican monsoon. J. Climate, 14, 121–137.

Bordoni, S., and B. Stevens, 2006: Principal component analysis ofthe summertime winds over the Gulf of California. A gulfsurge index. Mon. Wea. Rev., 134, 3395–3414.

——, P. E. Ciesielski, R. H. Johnson, B. D. McNoldy, and B.

Stevens, 2004: The low-level circulation of the North Ameri-can monsoon as revealed by QuikSCAT. Geophys. Res. Lett.,31, L10109, doi:10.1029/2004GL020009.

Bringi, V. N., T. W. Tang, and V. Chandrasekar, 2004: Evaluationof a new polarimetrically based Z–R relation. J. Atmos. Oce-anic Technol., 21, 612–623.

Brito-Castillo, L., A. V. Douglas, A. Leyva-Contreras, and D.Lluch-Belda, 2003: The effect of large-scale circulation onprecipitation and streamflow in the Gulf of California water-shed. Int. J. Climatol., 23, 751–768.

Ciesielski, P. E., and R. H. Johnson, 2008: Diurnal cycle of surfaceflows during 2004 NAME and comparison to model reanaly-sis. J. Climate, 21, 3890–3913.

Douglas, M., R. A. Maddox, K. Howard, and S. Reyes, 1993: TheMexican monsoon. J. Climate, 6, 1655–1667.

Durre, I., and J. M. Wallace, 2001: The warm season dip in diurnaltemperature range over the eastern United States. J. Climate,14, 354–360.

Farfan, L. M., and J. A. Zehnder, 1994: Moving and stationarymesoscale convective systems over northwest Mexico duringthe Southwest Area Monsoon Project. Wea. Forecasting, 9,630–639.

Ferraro, R. R., 1997: Special Sensor Microwave Imager derivedglobal rainfall estimates for climatological applications. J.Geophys. Res., 102, 16 715–16 735.

Fiorino, S. T., and E. A. Smith, 2006: Critical assessment of mi-crophysical assumptions within TRMM radiometer rain pro-file algorithm using satellite, aircraft, and surface datasetsfrom KWAJEX. J. Appl. Meteor. Climatol., 45, 754–786.

Gebremichael, M., E. Vivoni, C. J. Watts, and J. C. Rodriguez,2007: Submesoscale spatiotemporal variability of NorthAmerican monsoon rainfall over complex terrain. J. Climate,20, 1751–1773.

Gochis, D. J., W. J. Shuttleworth, and Z.-L. Yang, 2002: Sensitiv-ity of the modeled North American monsoon regional cli-mate to convective parameterization. Mon. Wea. Rev., 130,1282–1298.

——, J. C. Leal, W. J. Shuttleworth, C. J. Watts, and J. Garatuza-Payan, 2003: Preliminary diagnostics from a new event-basedprecipitation monitoring system in support of the North Ameri-can Monsoon Experiment. J. Hydrometeor., 4, 974–981.

——, A. Jimenez, C. J. Watts, J. Garatuza-Payan, and W. J.Shuttleworth, 2004: Analysis of 2002 and 2003 warm-seasonprecipitation from the North American Monsoon Experi-ment Event Rain Gauge Network. Mon. Wea. Rev., 132,2938–2953.

——, L. Brito-Castillo, and W. J. Shuttleworth, 2006: Hydrocli-matology of the North American monsoon region in north-west Mexico. J. Hydrol., 316, 53–70.

——, C. J. Watts, J. Garatuza-Payan, and J. C. Rodriguez, 2007:Spatial and temporal patterns of precipitation intensity asobserved by the NAME Event Rain Gauge Network from2002 to 2004. J. Climate, 20, 1734–1750.

Gutzler, D. S., and Coauthors, 2005: The North American Mon-soon Model Assessment Project: Integrating numerical mod-eling into a field-based process study. Bull. Amer. Meteor.Soc., 86, 1423–1429.

Harris, D., E. Foufoula-Georgiou, and C. Kummerow, 2003: Ef-fects of underrepresented hydrometeor variability and par-tial beam filling on microwave brightness temperatures forrainfall retrieval. J. Geophys. Res., 108, 8380, doi:10.1029/2001JD001144.

Higgins, W., and Coauthors, 2006: The NAME 2004 field cam-

742 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 9

Page 16: PDF (2978 KB)

paign and modeling strategy. Bull. Amer. Meteor. Soc., 87,79–94.

Hong, Y., K. Hsu, S. Sorooshian, and X. Gao, 2004: Precipitationestimation from remotely sensed imagery using an artificialneural network cloud classification system. J. Appl. Meteor.,43, 1834–1853.

——, ——, ——, and X. G. Gao, 2005: Improved representationof diurnal variability of rainfall retrieved from the TropicalRainfall Measurement Mission Microwave Imager adjustedPrecipitation Estimation from Remotely Sensed InformationUsing Artificial Neural Networks (PERSIANN) system. J.Geophys. Res., 110, D06102, doi:10.1029/2004JD005301.

——, D. J. Gochis, Y. Cheng, K.-L. Hsu, and S. Sorooshian, 2007:Evaluation of PERSIANN-CCS rainfall measurement usingthe NAME Event Rain Gauge Network. J. Hydrometeor., 8,469–482.

Hsu, K. L., X. G. Gao, S. Sorooshian, and H. V. Gupta, 1997:Precipitation estimation from remotely sensed informationusing artificial neural networks. J. Appl. Meteor., 36, 1176–1190.

Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, S.Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Globalprecipitation at one-degree daily resolution from multisatel-lite observations. J. Hydrometeor., 2, 36–50.

——, and Coauthors, 2007: The TRMM Multi-satellite Precipita-tion Analysis: Quasi-global, multi-year, combined-sensorprecipitation estimates at fine scale. J. Hydrometeor., 8, 38–55.

Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2001: A real-timeglobal half-hourly pixel-resolution infrared dataset and itsapplications. Bull. Amer. Meteor. Soc., 82, 205–217.

——, V. E. Kousky, and R. J. Joyce, 2005: Diurnal cycle of pre-cipitation determined from the CMORPH high spatial andtemporal resolution global precipitation analyses. J. Geo-phys. Res., 110, D23105, doi:10.1029/2005JD006156.

Johnson, R. H., P. E. Ciesielski, B. D. McNoldy, P. J. Rogers, andR. K. Taft, 2007: Multiscale variability of the flow during theNorth American Monsoon Experiment. J. Climate, 20, 1628–1648.

Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. P. Xie, 2004:CMORPH: A method that produces global precipitation es-timates from passive microwave and infrared data at highspatial and temporal resolution. J. Hydrometeor., 5, 487–503.

Kummerow, C., 1998: Beamfilling errors in passive microwaverainfall retrievals. J. Appl. Meteor., 37, 356–370.

——, and Coauthors, 2001: The evolution of the Goddard Profil-ing Algorithm (GPROF) for rainfall estimation from passivemicrowave sensors. J. Appl. Meteor., 40, 1801–1820.

Lang, T. J., D. A. Ahijevych, S. W. Nesbitt, R. E. Carbone, S. A.Rutledge, and R. Cifelli, 2007: Radar-observed characteris-tics of precipitating systems during NAME 2004. J. Climate,20, 1713–1733.

Liu, C., E. J. Zipser, and S. W. Nesbitt, 2007: Global distributionof tropical deep convection: Different perspectives fromTRMM infrared and radar data. J. Climate, 20, 489–503.

McCollum, J. R., W. F. Krajewski, R. R. Ferraro, and M. B. Ba,2002: Evaluation of biases of satellite rainfall estimation al-gorithms over the continental United States. J. Appl. Meteor.,41, 1065–1080.

Mesinger, F., and Coauthors, 2006: North American Regional Re-analysis. Bull. Amer. Meteor. Soc., 87, 343–360.

Negri, A., R. F. Adler, E. J. Nelkin, and G. J. Huffman, 1994:Regional rainfall climatologies derived from Special SensorMicrowave Imager (SSM/I) data. Bull. Amer. Meteor. Soc.,75, 1165–1182.

Nesbitt, S. W., and E. J. Zipser, 2003: The diurnal cycle of rainfalland convective intensity according to three years of TRMMmeasurements. J. Climate, 16, 1456–1475.

——, ——, and C. D. Kummerow, 2004: An examination of ver-sion-5 rainfall estimates from the TRMM Microwave Imager,Precipitation Radar, and rain gauges on global, regional, andstorm scales. J. Appl. Meteor., 43, 1016–1036.

Rosenfeld, D., and C. W. Ulbrich, 2003: Cloud microphysicalproperties, processes, and rainfall estimation opportunities.Radar and Atmospheric Science: A Collection of Essays inHonor of David Atlas, Meteor. Monogr., No. 52, Amer. Me-teor. Soc., 237–258.

Rowe, A. K., S. A. Rutledge, T. J. Lang, P. E. Ciesielski, and S. M.Saleeby, 2008: Elevation-dependent trends in precipitationduring NAME. Mon. Wea. Rev., in press.

Saleeby, S. M., and W. R. Cotton, 2004: Simulations of the NorthAmerican Monsoon System. Part I: Model analysis of the1993 monsoon season. J. Climate, 17, 1997–2018.

Smith, J. A., and W. F. Krajewski, 1991: Estimation of the meanfield bias of radar estimates. J. Appl. Meteor., 30, 397–412.

Sorooshian, S., K. L. Hsu, X. Gao, H. V. Gupta, B. Imam, and D.Braithwaite, 2000: Evaluation of PERSIANN system satel-lite-based estimates of tropical rainfall. Bull. Amer. Meteor.Soc., 81, 2035–2046.

Steiner, M., and J. A. Smith, 2004: Scale dependence of radar-rainfall rates—An assessment based on raindrop spectra. J.Hydrometeor., 5, 1171–1180.

Stensrud, D. J., R. L. Gall, and M. K. Nordquist, 1997: Surges overthe Gulf of California during the Mexican monsoon. Mon.Wea. Rev., 125, 417–437.

Stull, R. B., 1995: Meteorology Today: For Scientists and Engi-neers. West Publishing 88 pp.

Tripoli, G. J., and W. R. Cotton, 1989a: Numerical study of anobserved orogenic mesoscale convective system. Part I: Simu-lated genesis and comparison with observations. Mon. Wea.Rev., 117, 273–304.

——, and ——, 1989b: Numerical study of an observed orogenicmesoscale convective system. Part II: Analysis of governingdynamics. Mon. Wea. Rev., 117, 305–328.

Tustison, B., D. Harris, and E. Foufoula-Georgiou, 2001: Scaleissues in verification of precipitation forecasts. J. Geophys.Res., 106, 11 775–11 784.

——, D. E. Foufoula-Georgiou, and D. Harris, 2003: Scale-recursive estimation for multisensor quantitative precipita-tion forecast verification: A preliminary assessment. J. Geo-phys. Res., 107, 8377, doi:10.1029/2001JD001073.

Vivoni, E. R., and Coauthors, 2007: Variation of hydrometeorogi-cal conditions along a topographics transect in northwesternMexico during the North American mosoon. J. Climate, 20,1792–1809.

Whiteman, C. D., 2000: Mountain Meteorology: Fundamentalsand Applications. Oxford University Press, 355 pp.

Xie, P. P., and P. A. Arkin, 1997: Global precipitation: A 17-yearmonthly analysis based on gauge observations, satellite esti-mates, and numerical model outputs. Bull. Amer. Meteor.Soc., 78, 2539–2558.

Xie, P., Y. Yarosh, M. Chen, R. Joyce, J. J. Janowiak, and P. A.Arkin, 2005: Diurnal cycle of cloud and precipitation associ-ated with the North America monsoon system: Preliminaryresults for 2003 and 2004. Preprints, 16th Conf. on ClimateVariability and Change, San Diego, CA, Amer. Meteor. Soc.,4.22. [Available online at http://ams.confex.com/ams/pdfpapers/82939.pdf.]

AUGUST 2008 N E S B I T T E T A L . 743