Investigation of negative cloud radiative forcing over the Indian subcontinent and adjacent oceans during the summer monsoon season

Abstract. Radiative properties of clouds over the Indian subcontinent and nearby oceanic regions (0–25° N, 60–100° E) during the Asian summer monsoon season (June–September) are investigated using the Clouds and Earth's Radiant Energy System (CERES) top-of-the-atmosphere (TOA) flux data. Using multiyear satellite data, the net cloud radiative forcing (NETCRF) at the TOA over the Indian region during the Asian monsoon season is examined. The seasonal mean NETCRF is found to be negative (with its magnitude exceeding ~30 Wm−2) over (1) the northern Bay of Bengal (close to the Myanmar–Thailand coast), (2) the Western Ghats and (3) the coastal regions of Myanmar. Such strong negative NETCRF values observed over the Indian monsoon region contradict the assumption that near cancellation between LWCRF and SWCRF is a generic property of all tropical convective regions. The seasonal mean cloud amount (high and upper middle) and corresponding cloud optical depth observed over the three regions show relatively large values compared to the rest of the Indian monsoon region. Using satellite-derived cloud data, a statistical cloud vertical model delineating the cloud cover and single-scattering albedo was developed for the three negative NETCRF regions. The shortwave (SW), longwave (LW) and net cloud radiative forcing over the three negative NETCRF regions are calculated using the rapid radiative transfer model (RRTM) with the cloud vertical model as input. The NETCRF estimated from CERES observations show good comparison with that computed using RRTM (within the uncertainty limit of CERES observations). Sensitivity tests are conducted using RRTM to identify the parameters that control the negative NETCRF observed over these regions during the summer monsoon season. Increase in atmospheric water vapor content during the summer monsoon season is found to influence the negative NETCRF values observed over the region.

forcing (NETCRF) at the TOA over the Indian region during the Asian monsoon season is examined. The seasonal mean NETCRF is found to be negative (with its magnitude exceeding ∼ 30 W m −2 ) over (1) the northern Bay of Bengal (close to the Myanmar-Thailand coast), (2) the Western Ghats and (3) the coastal regions of Myanmar. Such strong negative NETCRF values observed over the Indian monsoon region contradicts 10 the assumption that near cancellation between LWCRF and SWCRF is a generic property of all tropical convective regions. The seasonal mean cloud amount (high and upper middle) and corresponding cloud optical depth observed over the three regions show relatively large values compared to rest of the Indian monsoon region. Using satellite derived cloud data, a statistical cloud vertical model delineating the cloud cover 15 and single scattering albedo was developed for the three negative NETCRF regions. The shortwave (SW), longwave (LW) and net cloud radiative forcing over the three negative NETCRF regions are calculated using the Rapid Radiative Transfer Model (RRTM) with cloud vertical model as input. The NETCRF estimated from CERES observations show good comparison with that computed using RRTM (within the uncer-Section 3 examines the radiative transfer model used for the computation of TOA flux and different parameterization schemes. Section 4 examines the CERES observation of CRF over the Indian region. Section 5 describes the comparison of NETCRF from CERES and RRTM simulations and various sensitivity analysis performed. Sections 6 and 7 summarize the main results of the paper. The mean cloud radiative forcing over the Indian region during the summer monsoon season (June-September) of 2002-2005 was derived by analyzing the TOA flux data from CERES Aqua SRBAVG-GEO (Edition 2A) dataset, which contain the monthly 15 mean regional TOA total-sky and clear-sky radiative fluxes (LW and SW) in a 1 • × 1 • latitude/longitude grid. CERES SRBAVG-GEO data use narrow band radiance from geostationary meteorological satellites to account for changes in flux and cloud conditions between daily CERES observations, which reduce temporal sampling errors. The uncertainties in the estimated CERES TOA flux is relatively small compared to 20 that derived from the ERBE data  mainly due to better scene identification and incorporation of better angular distribution models (Smith et al., 2012 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | clear sky flux is possible only from geographical regions where clouds occur less frequently and cloud free regions have relatively large area. Because of this, clear-sky flux maps from CERES SRBAVG GEO contain many gaps (no clear sky flux data), especially over the Indian monsoon region. In order to circumvent this problem, TOA clear sky flux data from the CERES Terra Energy Balanced and Filled (EBAF) Edition 1A 5 product (Loeb et al., 2009a) were used in the present study. CERES EBAF data provides TOA clear-sky fluxes for many overcast regions that have no CERES clear-sky observations over the course of a month. CERES SSF TOA clear-sky fluxes require 99 % of MODIS pixels (with 1 km nominal area) within the CERES footprint (20 km nominal) to be classified as clear. However, in a overcast cloudy region, there maybe 10 1 km clear-sky patch present inside a 20 km footprint region. The CERES EBAF product uses this clear-sky MODIS pixel radiance to derive broadband radiances, which are constrained to the overall CERES footprint to derive the TOA clear sky flux. In the present study, monthly mean TOA total sky flux from CERES SRBAVG GEO product and monthly mean TOA clear sky flux from the CERES EBAF product are used in the 15 estimation of cloud radiative forcing over the Indian region during the summer monsoon season. More details regarding the CERES EBAF dataset are available online (http: //ceres.larc.nasa.gov/documents/DQ_summaries/CERES_EBAF_Ed2.7_DQS.pdf). CRF is used as a metric to assess the radiative impact of clouds on the climate system, which is defined as the difference between TOA clear and total-sky flux. Using 20 the CERES TOA flux measurements, the cloud radiative forcing is calculated by taking the difference between TOA clear sky and total sky flux. SWCRF = SW clear − SW total (1) LWCRF = LW clear − LW total (2) 25 Where the subscripts "clear" and "total" represent the TOA clear-sky and total-sky fluxes, respectively. The net cloud radiative forcing (NETCRF) at the TOA is estimated by adding the shortwave and longwave cloud forcing.
Which can be re written as, NETCRF = SW clear + LW clear − (SW total + LW total ) (4) To determine the cloud radiative forcing over the Indian region, monthly mean TOA total-sky flux from CERES SRBAVG GEO and clear-sky flux from CERES EBAF dataset during the summer monsoon season (June-September) of 2002-2005 are 5 used.

Uncertainty analysis
This section presents the methodology used to compute the total uncertainty in NETCRF values due to uncertainties in the CERES TOA flux measurement. Uncertainties in the CERES flux measurement can be broadly categorized into three main com-10 ponents; Sampling errors, calibration errors and algorithm errors. Sampling error refers to error associated with time sampling and spatial averaging of the data associated with the instrument normalization (Young et al., 1998), which corresponds to 0.3 W m −2 for the SW and LW flux (Loeb et al., 2009a). An additional error term equal to the standard error of the regional mean TOA flux is also added to the sampling error term. The 15 term "algorithm error" refers to the errors associated with data retrieval which in this case mainly stems from uncertainties associated with narrow-to-broadband conversion of radiance, Angular Distribution Models (ADM) and scene identification whereas calibration error refers to the instrument measurement error (Level 1 product error). More details regarding the of CERES TOA flux estimation and associated uncertain-20 ties can be found in Loeb et al. (2009a) as well as in the CERES EBAF data quality summary document. In the present analysis, it is assumed that sampling, calibration and algorithm errors associated with the CERES TOA flux are uncorrelated and independent. The total uncertainty in the TOA flux due to these different error sources can be expressed as (Chambon et al., 2012), Using the above equation, uncertainty in the CERES TOA clear and total sky flux (LW and SW) are estimated. The uncertainty computations are performed at the monthly mean scale using error estimates for a 1 • × 1 • latitude/longitude grid area. In order to estimate the uncertainty in NETCRF, effect of propagation of variable uncertainty on the uncertainty of a function is to be considered. From Eq. (4), it can be 5 seen that NETCRF is a function of SW and LW flux. Hence total uncertainty associated with the CERES NETCRF measurement is related to uncertainties in CERES SW and LW flux (Loeb et al., 2009b). Using the general law of error propagation, it is possible to analytically determine how measurement uncertainty propagates into quantities, which are functions of the measurement. For a multi-variable function y(x 1 , x 2 , x 3 , . . .x N ), the 10 total uncertainty in y due to uncertainty in the input variables x (assuming that error contributions are small compared to the absolute value of the variable) can be expressed as (Taylor, 1982;Lo, 2005) where δy is the total uncertainty in y, δx i and δx j are the uncertainties associated with 15 the input variables (x i and x j ) and R ij represent the correlation coefficient between the input variables. The uncertainty in y are governed by the (a) change in y for a given change in the variables x i and x j (partial derivatives), (b) uncertainties in the input variable δx i and δx j and (c) how the variables x i and x j are correlated. If x i and x j are not correlated and independent of each other, second term in Eq. (6) vanishes and the 20 equation takes the form of a Gaussian error propagation formula (Taylor, 1982;Evans et al., 1984). Depending on the correlation between individual input variables and sign of the product of partial derivatives in the second term of Eq. (6), uncertainty in y can increase or decrease. The uncertainties in the CERES TOA SW and LW (clear and total sky) flux represent the input variables in the Eq. (6) and R represent the 25 correlation between these two variables. However, if TOA flux from two different dataset 28902 is estimated.

Cloud data
The seasonal mean cloud parameters from the CERES SRBAVG2-GEO dataset for the summer monsoon season of 2002-2005 are used in the study. The SRBAVG2 GEO cloud data include layer averaged monthly mean cloud and aerosol retrievals 10 from MODIS and geostationary satellites (Remer et al., 2005;Menzel et al., 2008). The MODIS derived cloud fraction and cloud optical depth are believed to be more accurate than that from geostationary satellites owing to the higher quality of the MODIS data. Studies show that MODIS could detect cirrus clouds over tropics having cloud optical depth as low as 0.02 with an uncertainty factor of 2 (Dessler and Yang, 2003

ISCCP FD TOA flux data
The International Satellite Cloud Climatology Project (ISCCP) data is an archive of more than 20 yr of global cloud observations. ISCCP data utilizes the radiance information from a series of geostationary satellites to create 3 hourly global maps of cloudiness. The ISCCP-FD data is an improved version of a previous ISCCP radiative 5 flux product  and provides radiative fluxes at the TOA, surface, and several levels within the atmosphere (Zhang et al., 2004). ISCCP FD data provides global total-sky and clear-sky fluxes (at surface, 680 mbar, 440 mbar, 100 mbar and TOA) for every 3 h interval in the shortwave and longwave range. Inter comparison of monthly mean fluxes from ISCCP-FD with ERBE and CERES suggest that there exist 10 an uncertainty of the order of ∼ 5-10 W m −2 in the calculated TOA flux from ISCCP-FD dataset. Details regarding the data and methodology adopted in the estimation of ISCCP FD flux are provided in Zhang et al. (2004) and Rossow et al. (2005). In the present study, TOA SW and LW fluxes from the ISCCP radiative flux dataset (ISCCP-FD) for the June-September months of 2002-2005 are used to estimate the seasonal 15 mean CRF over the Indian region.

Rapid Radiative Transfer Model (RRTM)
The rapid radiative transfer model is a band model for the calculation of longwave and shortwave atmospheric radiative fluxes and heating rates (Mlawer et al., 1997;Iacono et al., 2000;Clough et al., 2005). RRTM use the correlated-k method, which is a ac-20 curate and computationally fast radiative transfer scheme. TOA LW fluxes calculated by RRTM agree with those computed by line-by-line radiative transfer model within ∼ 1 W m −2 range, while SW fluxes agree within ∼ 1.5 W m −2 range (Clough et al., 2005;Morcrette et al., 2008). During the Spectral Radiance Experiment (SPECTRE) program, results from the RRTM flux simulations are validated against other radiation mod-Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | and Fouquart, 1991;Ellingson and Wiscombe, 1996). An important feature of RRTM is that it incorporates Monte-Carlo Independent Column Approximation (McICA) to represent sub-grid cloud variability Pincus et al., 2003). A 1-D column version of RRTMG is used in the present study to compute the TOA fluxes. RRTM input data typically consists of surface emissivity, cloud/aerosol optical depth, altitude 5 profile (60 atmospheric pressure levels) of temperature, pressure, cloud fraction and single scattering albedo, atmospheric mixing ratio profiles of water vapor, ozone, CO 2 , methane and other trace gases, etc.
In the present study, a tropical model atmosphere incorporating above mentioned parameters are used as input in RRTM simulations. An idealized altitude profile of 10 atmospheric water vapor mixing ratio is built and used in the model simulations. For this purpose, altitude profile of relative humidity (RH) in the atmosphere is constructed following the methodology adopted in Roca et al. (2004). In order to construct the vertical profile, first it is assumed that the relative humidity is constant in the atmospheric boundary layer (between surface and 850 hPa) and in the stratospheric lay-15 ers (< 100 hPa). RH value of 85 % is assumed for the boundary layer and 10 % for the stratospheric layer in all the profiles. The free tropospheric (between 850 hPa and 100 hPa) humidity is then varied from 5 % to 100 % in steps of 5 to create a number of RH altitude profiles. By converting the relative humidity to corresponding water vapor mixing ratio, a number of idealized water vapor mixing ratio profiles are created. 20 The precipitable water (PW) concentration for each profile is estimated by integrating the water vapor mixing ratio from surface to top of the model atmospheric layer. In the present model simulations, changing the free tropospheric humidity (from 5 % to 100 %) in the model atmospheric profile leads to a corresponding change in the PW from 33 to 68 mm. 25 Using RRTM, TOA flux is estimated for every half hour interval for full day, which is then integrated to obtain the daily mean TOA SW and LW flux. The total sky SW and LW flux are estimated by incorporating cloud parameters (cloud cover, optical depth and single scattering albedo) in to the model computation, whereas clear sky flux estima-Introduction tion do not include any cloud data. In the computation of total sky LW flux using RRTM, altitude profile of mean cloud cover and cloud optical depth (from CERES data) for the study region is used. For the total sky SW flux computation, in addition to cloud cover and optical depth, altitude profile of single scattering albedo is also required (which is not available from the satellite measurements). In addition, the model also requires 5 information regarding vertical overlap between different cloud layers. However, satellite derived cloud cover data do not provide information about overlap among different cloud layers in the atmosphere without which modeling of TOA flux is difficult. To circumvent this deficiency, parameterization schemes are used to determine the vertical overlap between different cloud layers and cloud single scattering albedo (ice and wa-10 ter clouds). Details regarding the cloud overlap scheme and single scattering albedo parameterizations are described in the subsequent sections.

Cloud overlap
Modeling of radiative flux due to clouds are complicated by difficulties in parameter- 15 izing its single-scattering properties (Liou, 1986) and cloud vertical structure (Weare, 1999;Rossow et al., 2005). Satellite observations of multi-layered clouds from space only provide information about the topmost cloud layer encountered with lower level clouds being either fully or partially observed. If there is no overlap among different cloud layers in an atmospheric column, then cloud amount at each level observed by 20 the satellite is the actual cloud amount. However, this assumption does not hold true in most cases involving partially cloudy skies. When there is overlap among different cloud layers, information about actual amount of cloud in the lower levels are not fully recorded by satellites. Even with enhanced satellite and surface observation capabilities, information about cloud vertical structure is rather limited. Presence of partially 25 filled cloud layers in the atmosphere creates problem in the model computation of the radiative fluxes because of the nonlinear relation between cloud properties and TOA  , 1998). This, along with lack of information about the vertical cloud overlap in the model can lead to large errors in the estimated radiative fluxes. In order to circumvent this problem, most models employ cloud overlap schemes for computing the radiative fluxes (Collins, 2001;Zhang et al., 2004;Rossow et al., 2005;Cole et al., 2011). Cloud overlap schemes are also used by satellite simu-5 lators in comparing the simulated cloud data with observations from passive or active remote sensing instruments (Klein and Jakob, 1999;Webb et al., 2001;Zhang et al., 2005;Bodas-Salcedo et al., 2011). Most radiative transfer models use vertical cloud overlap schemes like maximum overlap, random overlap or combination of maximum and random overlap between cloud layers. In the present study, a cloud vertical model 10 is developed using a type of maximum/random cloud overlap scheme. Geleyn and Hollingsworth (1979) theorized that if clouds appear in two adjacent atmospheric layers, such cloud layers are usually vertical parts of the same cloud and there should be maximum overlap between them. Maximum overlap between two cloud layers can be expressed mathematically as, Where C n and C n−1 represents the cloud fraction of two adjacent cloud layers and C max represent total cloud amount due to the overlap of two cloud layers. The random overlap assumption holds good only when the cloud layers are separated by at least one clear-sky layer. The random cloud overlap scheme assumes that the cloudiness 20 in any given cloud layer is independent of the cloudiness of other layer (Warren et al., 1985). The total cloud amount in a vertical column assuming random overlap between cloud layers can be expressed mathematically as (Stephens et al., 2004), Where C total is the total cloud amount, C n is cloud fraction for a given cloud layer n. Tian 25 and Curry (1989)  small horizontal cloud area whereas they follow the random overlap for large horizontal area (≥ 500 km 2 ). But in reality, there are no completely random or maximum cloud overlap occurrences in nature but rather specific combinations of cloud types associated with specific meteorological conditions (Hahn et al., 2001;Rossow et al., 2005). Whether observed from satellites or surface, there exists a specific overlap relationship 5 among different cloud types for each meteorological situation. In the present study, we are trying to develop a cloud overlap scheme that represent the altitude structure of a convective cloud system characterized by contiguous cloud layers. Since contiguous cloud layers can be expected to possess fairly high degree of vertical correlation, a combination of random and maximum overlap schemes are used to represent the 10 cloud vertical structure. Here it is assumed that cloud layers belonging to a particular cloud block (eg., all the cloud layers in the high cloud group) in a convective system are maximally overlapped, whereas adjacent cloud blocks are randomly overlapped (eg., between high and upper middle cloud group). Therefore, effective cloud fraction for all the cloud layers belonging to a particular cloud block will remain same (due to maxi-15 mum overlap) whereas it will change from one cloud block to another (due to random overlap between two adjacent cloud blocks). Chou et al. (1998) also adopted a similar type of maximum/random assumption with maximum cloud overlap in each of three cloud regions (lower, middle, and upper troposphere) and random overlap between these cloud regions. Using this methodology (Eqs. 7 and 8), altitude profile of cloud 20 cover is constructed using the cloud fraction data. A graphical representation of typical altitude structure of contiguous cloud layers calculated using the cloud overlap scheme is shown in Fig. 1. The cloud vertical model is built as follows: the CERES GEO layer averaged cloud properties are defined for four cloud groups mainly; high, upper middle, lower middle 25 and low level clouds. The mean cloud top pressure for each cloud group defines the boundary of cloud blocks in the model atmosphere; ie, cloud layers belonging to high cloud block in the model atmosphere are defined between the CERES mean high cloud top pressure (usually between 180-250 hPa in the model) and upper middle cloud top Introduction pressure (between 300-400 hPa in the model). In the present cloud overlap scheme, the effective cloud fraction for all high cloud layers in the model will be equivalent to the CERES high cloud fraction since all cloud layers with in a particular cloud block are maximally overlapped. Effective cloud fraction for the upper middle cloud block in the model is estimated assuming random overlap (using the Eq. 7) between the CERES 5 high cloud and upper middle level cloud fraction. This newly estimated cloud fraction (using Eq. 8) is assigned to all the layers in the upper middle cloud block (maximum overlap) in the model atmosphere defined between the upper middle cloud top pressure and lower middle cloud top pressure. Using this methodology, effective cloud fraction for lower middle and low level clouds are also estimated. Base of the newly constructed 10 cloud vertical profile is fixed at the top of boundary layer (850 hPa) while the cloud top coincides with that of the high level cloud.

Cloud single-scattering albedo (SSA)
For estimating TOA SW flux using RRTM, altitude profile of cloud SSA is required along with cloud cover information. Single scattering properties of clouds are governed 15 by the cloud particle size, shape and water content. They also vary over a large range of values depending on the wavelength band under consideration. In this section, parameterization schemes used for deriving the single scattering albedo (SSA) of ice and water clouds are explained. SSA parameterization provides a mathematical relationship between the cloud properties (particle size, optical depth, water content) 20 and single scattering albedo. Several attempts were made to parameterize the SSA of a cloud system solely based on the cloud water content alone (Sun and Shine, 1994;Platt, 1997). However, it was observed that cloud water content alone is insufficient and information about the cloud particle size is also required in the parameterization of SSA (Wyser and Yang, 1998). For water clouds, the single-scattering properties can 25 be effectively parameterized either in terms of their average size (Slingo, 1989;Hu and Stamnes, 1993) or based on the water content/optical depth (Fouquart and Bonnel, 1980;Fouquart, 1985;Räisänen, 1999 Fouquart (1985), single scattering albedo (ω) for water clouds are estimated for different values of cloud optical depth. The parameterization equation can be expressed as (Fouquart, 1985), Where µ is the solar zenith angle and τ is the cloud optical depth. This equation is 5 based on calculations for a lower-tropospheric cloud with a specific drop size distribution having effective radius of 9.9 µm. Using the above formula, SSA values were calculated for water clouds with varying optical depth and used in the present study. Unlike water clouds, spherical particle assumption is not valid in determining the single scattering properties of ice clouds since they take on a variety of shapes like plates, hexagonal crystals, bullet rosettes (Schmidt et al., 1995). Because of this, single-scattering albedo of ice clouds are defined mainly by the effective cloud particle size (Hu and Stamnes, 1993;Fu, 1996). The effective particle size of a ice cloud can be defined mathematically as, Where L is the dimension of an ice crystal, V (L) is the volume of the crystal, A(L) is the projected area, and n(L) is the size distribution. The parameterization scheme takes into account the effective size of ice crystal, which removes the ambiguity regarding the particle shape and size from the SSA estimation. The single scattering albedo of an ice crystal can be expressed as (Key et al., 2002), Where b 0 , b 1 , b 2 and b 3 are the empirical coefficients determined through regression for different SW spectral bands (i) and R is the mean ice particle size (estimated from 28910 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | CERES SRBAVG2 GEO data). The bulk single scattering albedo (ω) of ice crystals for the entire SW band is computed by integrating the individual SSA values computed for different shortwave bands and expressed mathematically as (Slingo and Schrecker, 1982;Chou et al., 1998), Where β is the mean extinction coefficient and S is normalized irradiance in each spectral band. The above parameterization scheme for ice cloud single-scattering albedo is an extension of the Streamer radiative transfer model (Key and Schweiger, 1996), which has been validated for different ice crystal size distributions and habits. Using this method, SSA values are computed for ice cloud particles and used in the computation of TOA flux. Using these SSA parameterization schemes, a cloud vertical model delineating the effective cloud cover and SSA are developed and used to model the TOA flux and CRF.  Fig. 3, it can be seen that seasonal mean NETCRF show large variation over the Indian region with values ranging from +30 W m −2 to −80 W m −2 . No specific pattern exists in the regional variation of NETCRF over the Indian region. Based on the seasonal mean variation of NETCRF, the Indian region can be categorized into three distinct NETCRF regimes; (a) posi-5 tive NETCRF regime (> 20 W m −2 ) over the south Indian land mass as well as over the Srilankan region (b) near zero NETCRF (between +20 and −20 W m −2 ) over the oceanic regions to south of Indian land mass as well as over northern India and (c) negative NETCRF regime (< −20 W m −2 ) over the north Bay of Bengal close to Myanmar coast, Bangladesh, Myanmar, Inland china and over the northeast Arabian Sea as 10 well as over the Western Ghats. In the present study, we focus mainly on the negative NETCRF regimes over the Indian region (delineated by black boxes in Fig. 3). They are designated as, (1) the Bay of Bengal (10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22) • N, 85-100 • E) region representing oceanic regime, (2) Myanmar (15)(16)(17)(18)(19)(20) • N, 92-100 • E) and (3) the Western Ghats (10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20) • N, 72-77 • E) representing land regime. input to Eq. (6), total uncertainty in the NETCRF for the three regions are computed and presented in Table 3. In general, the total uncertainty in the estimated NETCRF from CERES TOA flux measurements varies between 3 and 6 W m −2 . Estimated uncertainty values are smallest for the Bay of Bengal region and largest for the Western Ghats. It must be noted that total uncertainty in the CERES NETCRF measurements 5 are much lower compared to that from ERBE NETCRF measurements Cess et al., 2001). Figure 4a and b depict the seasonal mean variation of cloud area fraction (in percentage) for the high and upper middle level clouds over the Indian region during the sum-10 mer monsoon season estimated from CERES cloud data. Only high and upper middle level cloud fractions are shown in Fig. 4, since fractional coverage of lower middle and low level clouds are relatively lower (∼ 10 % or less) over the most of the negative NETCRF regions. From Fig. 4a, it can be seen that high cloud fraction is relatively large (> 40 %) over the North Bay of Bengal and land areas over eastern India compared to 15 that observed (∼ 20-30 %) over oceanic regions to the south and the Western Ghats. In contrast, middle level cloud fraction shows large values (∼ 40 %) compared to high cloud fraction over Coastal regions of Myanmar, Thailand and Cambodia as well as over the Western Ghats over Indian peninsula. From Fig. 4, it can be observed that the three negative NETCRF regions (in Fig. 3 The seasonal mean values of cloud parameters (cloud fraction, particle size, optical depth) and environmental variables (rain rate, Free tropospheric humidity (FTH), precipitable water) over the three negative NETCRF regions are calculated for the summer monsoon season of 2002-2005 and are shown in Table 4. From Table 4, it can be seen that high and upper middle cloud fraction dominates the total cloudiness over 5 the three regions. High cloud amount (∼ 51 %) and optical depth (∼ 14) are largest over the Bay of Bengal while upper middle cloud amount shows large values (∼ 34 % and 43 %) compared to high cloud amount over the land areas. Amount of lower level (lower middle and low) clouds are below 10 % over the Bay of Bengal while, it shows values between 10-15 % over land regions. Earlier studies showed that high clouds account for ∼ 65 % of all cloud grids observed over the Indian monsoon region during the monsoon season (Tang and Chen, 2006;Meenu et al., 2007). Relatively lower values of low level clouds (compared to high and middle level clouds) are observed over the three regions, which could be due to underestimation of low level clouds by MODIS and geostationary instruments over the Indian monsoon region (Tang and Chen, 2006). 15 The mean ice and water cloud particle size for the three regions show similar values while precipitable water vapor values (from GMAO GEOS database) show a variation between 48 to 58 mm for the three regions. From this analysis, it is not easy to comprehend the influence of various parameters on the negative NETCRF over these regions. In order to obtain a better understanding regarding the influence of these parameters 20 on the negative NETCRF over the three regions, TOA flux and CRF values for the three regions are computed using RRTM simulations. The regional mean values of various cloud and environmental variables estimated for the three regions are used as input in the model. Using the cloud cover and cloud particle size data from CERES, altitude profile of cloud cover and SSA for each region is developed employing the parame- The seasonal mean TOA flux estimated for the three negative NETCRF regions from CERES, ISCCP-FD and RRTM simulations are shown in Table 5. The ISCCP FD TOA flux data belonging to the study period is used in the estimation of seasonal mean TOA flux and CRF. In general, TOA flux from CERES observations and RRTM simulations  Table 5), SW, LW and NET CRF values for the three regions are calculated and presented in Table 6. From used in the ISCCP FD flux estimation. However, it can be argued that the ISCCP CVS scheme is a general overlap scheme applicable for a broad range of meteorological conditions, while the present cloud overlap scheme is tailored for a specific convective cloud condition. The disparity in SW and LW CRF values computed by the RRTM and CERES over the land regimes (Myanmar and the Western Ghats) can be attributed 5 to the underlying uncertainty associated with the parameterization of cloud vertical structure and its microphysical properties. After successfully simulating the negative NETCRF regimes over the Indian region using RRTM, the next step is to identify the parameters that control the negative NETCRF over these regions. This is carried out by analyzing the sensitivity of cloud radiative forcing to various cloud and environmental 10 variables (using RRTM) and explained in the subsequent sections.

Sensitivity calculations
In this section, sensitivity of CRF to various cloud macro/micro physical properties and environmental variables are analyzed by studying their relative contribution to the NETCRF. In the earlier section, it was shown that negative NETCRF over the three re-15 gions could be modeled with good accuracy using RRTM and cloud parameterization schemes. Using the same simulation methods, it is possible to quantify the dependence of NETCRF on various cloud micro/macro physical and environmental variables over the three regions. This also provides an opportunity to test the veracity of theoretical hypothesis propagated by various investigators on the occurrence of negative 20 NETCRF over tropical convective regions. In this analysis, influence of cloud amount, cloud particle size, single scattering albedo and atmospheric water vapor on the negative NETCRF over the three regions are examined.

Influence of cloud macro-physical properties on CRF
In  Fig. 5. From Fig. 5, it can be seen that LWCRF decreases monotonically 5 with increase in the cloud top pressure (from ∼ 100 and 80 W m −2 respectively to ∼ 45 and 40 W m −2 , respectively) for the two regions while the SWCRF shows very little variation (< 5 W m −2 ). This indicates that change in cloud top altitude of high clouds cause relatively large variation in LWCRF (∼ 50 W m −2 ) compared to SWCRF, which results in causing an imbalance between the two. The sharp wedge observed in the SWCRF variation in Fig. 5 is due to the change in cloud cover type from high level to middle level. These results are in agreement with that of Kiehl (1994) who showed that decrease (increase) in cloud top altitude causes system to shift towards a negative (positive) NETCRF regime. However, seasonal mean cloud top altitude (high clouds) over the Indian region during monsoon season shows a variation between 180-240 hPa. For 15 such variation in cloud top altitude, corresponding variation in LWCRF observed for the three regions are less than 13 Wm −2 . This shows that small variation in cloud top altitude of high clouds cannot significantly influence the magnitude of NETCRF over the Indian region.
In the second sensitivity analysis, we try to quantify the competing influence of cloud 20 top and cloud amount on the LWCRF and NETCRF. In this analysis, high cloud amount in the cloud vertical model representing the Bay of Bengal regime is varied from its original value of 50 % to 20 % in steps of 10 %, keeping cloud amount of other groups in the cloud vertical model same. By stepwise reduction in high cloud amount from cloud vertical model, fraction of upper middle level clouds exposed to the TOA (and to 25 the incoming solar incoming radiation) increases from 12 % to 42 %. Running RRTM with this modified cloud vertical profile shows a decrease in LWCRF from 78 W m −2 to 66 W m −2 while the SWCRF show very little variation since the total cloud cover remains the same (top level clouds are sliced off while bottom level clouds remain intact). Impact of this reduction in high cloud amount is to increase the absolute magnitude of the NETCRF (by 12 W m −2 ) and shift NETCRF the region towards a stronger negative regime due the decrease of LWCRF. Similar analysis was performed for the Myanmar and the Western Ghats region where it showed similar result as that of the Bay of Bengal region. Cess et al. (2001) observed that change in cloud vertical structure as-5 sociated with El Nino over the tropical Pacific Ocean tend to cause substantial radiative cooling. It was observed that CRF over Pacific warm pool is partially governed by high and middle-level clouds during the El Nino year compared to high level clouds during normal years.This analysis indicates that in a multi layered cloud system, both cloud vertical structure and cloud amount influences the NETCRF, even though magnitude 10 of variation in NETCRF is not significantly large in this case.

ACPD
In the third test, influence of low levels clouds on the NETCRF is examined over the three regions while keeping all other parameters in the cloud vertical model constant. In this analysis, un-obscured portions (visible from TOA) of the low levels clouds are removed from the cloud vertical profile and LWCRF and SWCRF values are recom-15 puted using the modified cloud vertical model. The modified LW and SW CRF values estimated from the analysis are presented in Table 7. From Table 7, it can be seen that the low level cloud cover over these regions has very little influence in controlling the observed LWCRF and SWCRF compared to high level clouds. The SWCRF over the Western Ghats shows a decrease (∼ 7 W m −2 ) when the modified cloud vertical profile 20 (low level clouds removed) is used. For the Bay of Bengal and Myanmar region, the maximum decrease in SWCRF is of the order of < 3 W m −2 . Corresponding variation in the LWCRF is considerably small compared to that of SWCRF. Large low level cloud fraction (∼ 14 %) was observed over the Western Ghats while it is < 6 % for the other two regions. Present analysis shows that low level cloud amount has little influence in 25 modulating the NETCRF values over the three study regions. 28918

Influence of cloud microphysical properties on CRF
In the present analysis, influence of cloud single scattering albedo and ice cloud particle size on NETCRF is investigated using RRTM. The layer averaged cloud particle size (from SRBAVG2 GEO data) for ice clouds over the Indian region show variation of effective particle size between the range 18-30 µm. Cloud ice particle size is directly related to the cloud single scattering properties (Eq. 11), which in turn modulates the SWCRF and NETCRF of the cloud system. In the present analysis, an attempt is made to quantify the sensitivity of SWCRF to the ice cloud particle size. This is done by computing the SWCRF over the three regions by varying the ice particle size of high and upper middle level clouds (there by varying the SSA) in the cloud vertical model 10 while keeping all other parameters constant. The ice particle size is varied from 20 to 40 µm and SWCRF values are computed each time using RRTM by incorporating the modified SSA in the input cloud model. SWCRF estimated from the model simulations are presented in Table 8. From Table 8, it can be seen that the SWCRF values show a monotonic decrease with increasing ice particle size for the three regions. A max- 15 imum decrease in SWCRF value of ∼ 10 W m −2 for an increase in ice particle size from 20 to 40 µm is observed over the Bay of Bengal region while it is much less over the other two regions. The study indicates that increase in particle size leads to moderate decrease in SWCRF and NETCRF over the study region. Studies have shown changes in cloud particle (ice) size can modify the net radiative forcing of cirrus clouds 20 to a cooling or warming regime depending on the direction of change (Zhang et al., 1999). However, it is seems unlikely that small variation in cloud ice particle size (between 18-30 µm) over the Indian region alone can bring about the negative NETCRF values observed over the region. 25 In this section, sensitivity analysis carried out using RRTM to understand the influence of atmospheric water vapor on the LWCRF over the three negative NETCRF regions  (Sohn and Schmetz, 2004;Sohn and Bennartz, 2008). Water vapor being an important greenhouse gas absorbs the atmospheric LW radiation and decreases the LWCRF (while having little effect on SWCRF). Roca et al. (2004) proposed that damping of LWCRF by large amount of wa-5 ter vapor present in the atmosphere over the Bay of Bengal region during the summer monsoon season could be a reason for the observed negative NETCRF. In this analysis, CRF over the three negative NETCRF regions are simulated by varying the PW vapor content in the atmosphere from their original values (between 48-58 mm) while keeping all other parameters in model simulation constant. This is achieved by varying 10 the relative humidity (thereby water vapor mixing ratio) of free troposphere in the model atmosphere while keeping the boundary layer and stratospheric RH constant. This results in the formation of several model atmospheric profiles with distinctly different PW values (between 33 to 68 mm). Using this methodology, variation of TOA LW flux and LWCRF with PW for the three regions is examined. 15 Variation of TOA LW flux (clear and total sky) and LWCRF with PW for the Bay of Bengal region is shown in Fig. 6. In general, clear and total sky LW flux shows a monotonic decrease with increase in PW. TOA LW clear sky flux over the Bay of Bengal region shows a decrease of ∼ 45 W m −2 for an increase in PW value from 33 to 68 mm while the total sky LW flux show a decrease of ∼ 20 W m −2 . The LWCRF 20 also show considerable decrease over the Bay of Bengal region from 106 W m −2 to 74 W m −2 (decrease of ∼ 32 W m −2 ) for a corresponding increase in PW from 33 mm to 68 mm. Over Myanmar and the Western Ghats, LWCRF shows a decrease of ∼ 30 and ∼ 25 W m −2 respectively, for a similar variation in PW from 33 mm to 68 mm. Sohn et al. (2006) demonstrated that water vapor in the upper troposphere can contribute 25 12 W m −2 to the LWCRF over convectively active tropical regions. This analysis shows that atmospheric water vapor can cause a relatively larger variation in LWCRF (and in NETCRF) compared to other variables discussed in earlier sections. For a through understanding, seasonal mean variation in PW over the Indian region during different 28920 (DJF). This indicates that between dry (winter) and wet (summer) seasons, PW content in the atmosphere over the Indian region increases by ∼ 25-35 mm. RRTM simulations shows that such increase in PW can cause a substantial decrease in clear sky TOA flux (∼ 30-40 W m −2 ) and LWCRF (∼ 20-30 W m −2 ). From Fig. 6, it can be seen that LWCRF over Bay of Bengal increase by ∼ 23 W m −2 for a decrease in PW content from 10 58 mm to 30 mm. Over the two land regimes, a similar increase in LWCRF is observed though magnitude of increase in LWCRF is lower than that observed over the Bay of Bengal region. From this analysis, it can be seen that the atmospheric water content over these regions significantly modifies the observed LWCRF and NETCRF values. Hence, relatively large amount of water vapor in the atmosphere over the three negative 15 NETCRF regions during the summer monsoon season is a major factor controlling the imbalance between SWCRF and LWCRF (and the negative NETCRF regimes).

Inter-comparison of the oceanic NETCRF regimes: Bay of Bengal vs. West Pacific
We now consider two specific oceanic convective regions for a better understanding Pacific during this period lies mainly in the range of ±20 W m −2 . Negative NETCRF values (∼ 20 Wm −2 ) are observed over small areas close to the Indonesian islands while rest of the region show positive or near zero NETCRF values. A similar regional variation of NETCRF over the Bay of Bengal (Fig. 3) region during the monsoon season shows negative NETCRF values ranging from −10 to −70 W m −2 . For a more quantitative assessment, mean TOA flux and CRF over the two regions are estimated and presented in Table 9 Table 9.
Table 10 presents the regional mean cloud fraction and cloud optical depth estimated using CERES data over the Bay of Bengal and the West Pacific. From Table 10, it can be seen that high level cloud fraction (∼ 51 %) observed over the Bay of Bengal is relatively large compared to that observed over the West Pacific (∼ 38 %) while, lower 20 level cloud types show almost similar variation over the two regions. A comparison of cloud optical depth and cloud top altitude between the various cloud groups also show similar variations. This indicates that cloud properties over the two convective regions show a lot of similarities except for the high level cloud fraction. However, the interesting question here is "whether the variation observed in the cloud amount (13 % and 5 % in 25 high and upper middle cloud cover respectively between these two regions) alone can cause the NETCRF to shift between near zero and negative values"? Rajeevan and Sreenivasan (2000) postulated that large high cloud amount observed over the Bay of Bengal region is the reason behind the large negative NETCRF compared to rest of 28922 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | tropical convective regions. However, RRTM simulations incorporating the mean cloud and environmental variables over the West Pacific region show that such variability in cloud cover alone is not enough to drive the system from a near zero to large negative NETCRF regime.
The other important parameters that can influence the NETCRF over the two regions 5 are cloud microphysical properties (cloud particle size, shape etc) and atmospheric water content. Comparison of cloud particle size (∼ 25 µm for ice and ∼ 12 µm for water clouds, respectively) over the two regions shows very little variation. The other variable that can influence the NETCRF over the two regions is the atmospheric water vapor content. In Sect. 5.1.3, influence of atmospheric water vapor on the TOA flux and CRF 10 over the Indian region was discussed. A similar analysis is carried out to quantify the influence of atmospheric water vapor on the NETCRF over the West Pacific. For this purpose, monthly mean atmospheric water vapor content over the two regions from special sensor microwave/imager (SSM/I) data (Wentz, 1997) was estimated for the April (West Pacific) and July (Bay of Bengal) months (2002)(2003)(2004)(2005) and is shown in 15 Fig. 9. The atmospheric water vapor content over the Bay of Bengal during the July month is extremely large (> 58 mm) compared to that observed over the West Pacific region (∼ 45-55 mm) during April. Over the negative NETCRF regions (< −30 W m −2 ) of Bay of Bengal, the PW reaches as high as ∼ 68 mm. The water vapor information over the land is not available from SSM/I data. However, from Fig. 9 gions during the summer monsoon season was studied. Using the cloud vertical model as input, NETCRF for the three regions were estimated using RRTM. The NETCRF values calculated from RRTM simulations found to agree well with CERES observations, while that from ISCCP FD data showed large differences. Sensitivity of the negative NETCRF values to various cloud micro/macro physical and environmental variables 20 were carried out using model simulations. Sensitivity of ice particle size to the NETCRF was evaluated by varying the ice particle radius (from 20 to 40 µm) in the parameterization of SSA, which produced a maximum variation of ∼ 10 W m −2 on the NETCRF values. Decrease in cloud particle size found to increase the SWCRF with very little variation in LWCRF. However, variations in CRF due to changes in cloud particle size 25 alone is not sufficient enough to cause the formation of negative NETCRF regimes observed over the Indian region.