Journal of Geophysical Research: Atmospheres

Cloud cover conditions and stability of the Western Ghats montane wet forests

Authors


Abstract

[1] Using remote sensing data from multiple satellites and numerical simulations we addressed the hypothesis that the Western Ghats tropical montane wet forests in southwest India are hydrometeorologically stressed and lowland deforestation has decreased cloud cover and precipitation. We generated water cloud and Leaf Area Index climatologies for a 6 year period and answered the following questions: (1) what are the diurnal and seasonal variations of water clouds, (2) what is the relationship between cloud cover and topography, (3) where are potential locations of the tropical montane wet forests, (4) what are the current hydrometeorological conditions over the identified tropical montane wet forests, and (5) what is the sensitivity of cloud cover and precipitation over the montane wet forests to lowland deforestation? The study found that (1) strong correlations between climatological precipitation and cloud frequency were present thereby indicating that higher spatial and temporal resolution cloud frequency is a good indicator of hydrometeorological conditions; (2) cloud frequency was lower in the dry season (10–25%) and higher in the monsoon seasons (40–45%); (3) a morning-afternoon pattern in cloud cover exists during both the dry and wet seasons; (4) morning cloud cover decreased from October to January with afternoon cloud cover similarly decreasing but with higher values than the morning cloud cover; (5) cloud cover generally increased with elevation; (6) 39%, 70%, 47% and 42% of the clouds intersected the mountains during March and May 2007 and February and April 2008, respectively; (7) our maps showing regions with conditions suitable to sustain tropical montane wet forests successfully identified known montane biodiversity hot spots; and (8) simulations showed an increase of afternoon cloud cover and dry season precipitation but reduced cloud immersion over the montane forests with lowland deforestation.

1. Introduction

[2] The Strategic Plan of the United Nation Environmental Program (UNEP) and the World Conservation Monitoring Centre (WCMC) is to reduce the rate of loss of the world's biodiversity by 2010 and to reverse the loss of environmental resources by 2015 [United Nation Environmental Program-World Conservation Monitoring Centre (UNEP-WCMC), 2006]. The goal is to put authoritative biodiversity knowledge at the center of decision making. However, accurate knowledge of current environmental conditions at several global biodiversity “hotspots” remains incomplete, an essential requirement for biodiversity conservation.

[3] The Western Ghats (in southwestern India) and Sri Lanka biodiversity hot spot are 1 of the 34 primary global biodiversity hot spots [Mittermeier et al., 2004]. However, the spatial extent and current environmental conditions of this fragile, endemic biodiversity hot spot [Hamilton et al., 1993; Doumenge et al., 1993; Myers et al., 2000] is not well defined. The total area of the Western Ghats and Sri Lankan biodiversity hot spots is estimated to have reduced from around 189,611 km2 to 43,611 km2 in recent times largely due to conversion of forests to tea, coffee, teak, eucalyptus and wattle plantations, shifting agriculture, and due to the creation of new reservoirs, construction of roads and railways [Mittermeier et al., 2004].

[4] The predominant biome type in the Western Ghats is tropical and subtropical broadleaf forest. Tropical rain forests represent primary centers of species richness and endemism within the Western Ghats, and are estimated to cover approximately 20,000 km2 [Mittermeier et al., 2004]. At the core of the tropical rain forests of the Western Ghats are tropical cloud forests [Nair et al., 1977], which are found in upland regions where orographically lifted cloud banks directly intersect the mountains [Bruijnzeel and Scatena, 2011]. The moisture input into tropical cloud forests consists of two parts: rainfall and direct cloud water interception, which is especially important during the dry season (Table 1). The Western Ghats biodiversity hot spot is home to hundreds of endemic species: 3049 plants, 18 mammals, 35 birds, 176 reptiles, 138 amphibians and 139 freshwater fishes. Of these 14 mammals, 10 birds and 95 amphibian species are vulnerable, endangered or critically endangered [Mittermeier et al., 2004].

Table 1. Reported Cloud Water Interception as a Percentage of Total Water Input From Several Locations Around the Worlda
LocationContribution of Cloud Water Interception to Total Water InputSource
  • a

    Reported values for the Western Ghats were not found in our literature search.

Central Cordillera of Panama2.4–60.6%Cavelier et al. [1996]
Columbia and Venezuela3.7–93.3%Cavelier and Goldstein [1989]
Galapagos Islands99%van der Werff [1978]
Central Veracruz, Mexico6–8% in the dry seasonHolwerda et al. [2010]
Northern Queensland, Australia4–30%McJannet et al. [2007]
Northern California17–34%Dawson [1998]
Dhofar Cloud Oasis, Oman∼67%Hildebrandt and Eltahir [2006]
Chile57.6%del-Val et al. [2006]
Guatemala12–35%Holder [2004]
Central Kalimantan, Indonesia6.2–11.4%Asdak et al. [1998]
Ibadan, Nigeria1.4–92.9%Opakunle [1989]
Ivory Coast4.6–42%Hutjes et al. [1990]
Colombia12.4–18.3%Veneklaas and van Ek [1990]
Philippines2.4%Mamanteo and Veracion [1985]
Tanzania21–23%Lundgren and Lundgre [1979]
Kona, Hawaii12–27%Brauman et al. [2010]
East Maui, Hawaii37–46%Scholl et al. [2007]
Canary Islands25–40%Ritter et al. [2008]

[5] Site-specific studies have shown that the tropical rain forests of the Western Ghats are limited in size and surrounded by an intervening matrix of land use that is under intense human pressure [Menon and Bawa, 1997; Jha et al., 2000], putting the protected regions at risk for continued deforestation. While a considerable amount of work has already been done on identifying areas of conservation value in the Western Ghats [e.g., Daniels et al., 1991; Ramesh et al., 1997; Prasad et al., 1998; Venkatraman et al., 2002; Das et al., 2006], with a biodiversity perspective, a regional-scale hydrometeorological investigation is necessary to (1) determine the environmental conditions of the Western Ghats and (2) identify potential locations that can hydrometeorologically sustain existing tropical rain forests and allow regeneration elsewhere if protected, as regrowth of secondary forests, which may occur rapidly under certain conditions, especially in hilly, mountainous and upland regions [Asner et al., 2009].

[6] In alignment with the goals of UNEP-WCMC [2006], this paper therefore focuses on (1) identification of current hydrometeorological conditions via cloud cover measurement over the Western Ghats and (2) identification of those regions of the Western Ghats with environmental conditions sufficient to sustain and regenerate secondary tropical rain forests if protected.

2. Study Area and Data

2.1. Study Area

[7] The Western Ghats of southwest India [Chandran, 1997] are an elongated mountainous region (8°N–20°N) [Das et al., 2006] running along the entire length of the west coast of peninsular southern India for approximately 1600 km except for the approximately 30 km break called the Palghat gap [Mittermeier et al., 2004]. The higher elevation regions are not continuous and using the United State Geological Survey (USGS) 1 km spatial resolution topography data three primary study regions (1, 2 and 3 from north to south) are delineated (Figure 1a). The entire Western Ghats mountains are spread over about 160,000 km2 [Das et al., 2006] and they form the catchment basins for the large number of rivers that drain about 40% of the Indian subcontinent. The average elevation of the Western Ghats is 1200 m, but there are numerous peaks, with the maximum elevation being 2695 m (e.g., Anaimudi). The lower Ghats have a humid and tropical climate, but elevated regions (>1500 m in the north and >2000 m in the south) have a temperate climate.

Figure 1.

(a) Topography from USGS. The three high elevation subregions that were studied are marked 1, 2, and 3. (b) Ecotypes were from a MODIS-derived land cover map. Both data sets were at 1 km spatial resolution.

[8] Annual rainfall varies from 3000 to 4000 mm on the western slopes and 500 to 1000 mm on the eastern slopes with occasional annual values up to 9000 mm. The higher rainfall on the western slopes is due to the predominant effect of the southwest summer monsoon that lasts from June to September. 80% of rainfall falls during the southwest monsoon period and the balance during the northeast monsoon (October–November) period [Mittermeier et al., 2004]. Mean temperature varies from 24°C to 20°C from north to south. January through March are the peak dry season months.

[9] Variations in topography and climate have created a highly diverse landscape with land cover that ranges from tropical evergreen forests to dry thorn forests. According to Kodandapani et al. [2004] roughly 31% of the Western Ghats are tropical evergreen and grassland/montane rain forests, while tropical dry deciduous and tropical moist forests account for 27% and 42% of the vegetation cover, respectively. Tropical montane cloud forests tend to form the core of the tropical rain forests and are found at elevations greater than 1800 m in south India [Nair et al., 1977]. Bruijnzeel et al. [2011] however report that in south India, Sri Lanka, Asian-Pacific islands and northern Australia the altitudinal limits of cloud forests ranges between 1000 and 3500 m. Characteristic patches of tropical montane stunted evergreen forests occur in the valleys and folds of the hills at these elevations. Current global land use categorizations do not identify tropical montane cloud forests.

[10] Based on the 1 km spatial resolution Moderate Resolution Imaging Spectroradiometer (MODIS) imagery [Hansen et al., 2000], we found that circa 2000 forests make up nearly 45% of the land use categories in the study region, whereas woodlands represent 38.4% and the remaining 16.6% other land use categories (Figure 1b): 0.2% of the forests are evergreen needleleaf, 21.9% evergreen broadleaf, 0.5% deciduous needleleaf, and 21.5% deciduous broadleaf. Woodlands, in this satellite based land cover classification scheme, are grid cells containing herbaceous or woody understories with tree canopy cover greater than 40% but less than 60%, whereas forests are grid cells with greater than 60% tree coverage and height exceeding 5 m [Hansen et al., 2000].

[11] The location specific major ecosystems of the Western Ghats include tropical evergreen forests in Amboli and Radhanagari, montane evergreen forests in Mahabaleshwar and Bhimashanker, moist deciduous forests in Mulsi and scrub forests in Mundunthurai. Two main centers of biodiversity are the Agashyamalai hills and the Silent Valley. The Nilgiri Biosphere Reserve [Kodandapani et al., 2004] promotes conservation of endemic and endangered species. The combination of a multifaceted landscape and heavy rainfall has made some areas nearly inaccessible, which has facilitated in retaining the biodiversity of the region. For example, almost one third of all the flowering plant species in India are found in this region, with the wildlife showing an equally rich diversity [Das et al., 2006].

[12] The natural biota of the region exhibits a high level of endemicity [Subramanyam and Nair, 2001]. Some of the prominent endemic mammalian species are the bat Latidens salimalii, lion-tailed macaque (Macaca silenus), Nilgiri tahr (Hemitragus hylocrius) and the Malabar civet (Viverra civettina) [Mittermeier et al., 2004]. Thirty-five endemic bird species such as the gray-headed bulbul (Pycnonotus priocephalus), white-bellied tree pie (Dendrocitta leucogastra), and Malabar parakeet (Psittacula columboides) occur in the low-elevation forests, whereas the white-bellied shortwing (Brachypteryx major), Nilgiri flycatcher (Eumyias albicaudata), and broad-tailed grassbird (Schoenicola platyura) occur at higher elevations [Mittermeier et al., 2004]. There are 138 endemic amphibians, 176 endemic reptiles and 139 endemic species of freshwater fish found together with an unknown but significant number of endemic invertebrate species [Mittermeier et al., 2004].

2.2. MODIS Imagery

[13] To observe the current hydrometeorological conditions over the three study subregions of the Western Ghats and identify hydrometeorologically suitable tropical rain forest sites annual cloud climatology over the Western Ghats was developed from individual scenes of both the morning (Terra) and afternoon (Aqua) MODIS overpasses. Each satellite is in a sun-synchronous orbit and views the surface of the Earth every 1 or 2 days. In the present study, only daytime data from the Terra and Aqua platforms were used, which have a 1030 local time (LT) ascending node and 1330 LT descending node equator crossing times, respectively. Each MODIS scene is made of 36 spectral bands ranging from 0.4 μm to 14.4 μm. The first two visible bands are at 250 m spatial resolution, while bands 3–7 are at 500 m spatial resolution. The remaining bands (8–36) are at 1 km spatial resolution. Approximately 7200 MODIS scenes were acquired, covering 5 consecutive years (2003–2007) plus the later part of 2002 and early part of 2008, yielding a total of 6 years of data to develop the cloud cover climatology.

2.3. Leaf Area Index

[14] The method used to identify the tropical rain forest sites also uses Leaf Area Index (LAI) information. LAI is defined as the ratio of total upper leaf surface of vegetation and the surface area of the land upon which the vegetation grows. The vegetation surface is composed primarily of leaf area and in lesser part by twigs, branches and the stem surface. During the dry season in tropical areas, the woody parts determine vegetation surface area. LAI is a dimensionless value which ranges typically from zero for bare ground to values even greater than 6 for dense forests. A plant with one layer of leaves all placed next to each other has a leaf area index of exactly 1.0, because the leaf area equals the ground area covered. In evergreen tropical forests, such as those of the Western Ghats, the LAI values are higher than the deciduous forests throughout the year due to their “evergreen” nature. This information is used together with elevation and cloud cover values to identify the locations of current evergreen forests.

[15] We used the 1 km spatial resolution, 8 day average MODIS Team MOD15A2 (Terra) Leaf Area Index product [Knyazikhin et al., 1998]. One characteristic of the MOD15A2 product is a data quality indicator, allowing the investigator to determine if the data in question is appropriate for the targeted study. The present study area is a region of high cloud cover, so the MOD15A2 product is used to ensure that the LAI values are taken only from clear conditions. However, note that in a rapidly varying terrain determination of accurate LAI is more difficult than over flat locations; we used the data quality indicator to identify and use only the highest quality retrieved data.

2.4. CloudSat and CALIPSO Data

[16] On 28 April 2006 two active remote sensors, Cloud Satellite (CloudSat) and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) were launched by NASA [Stephens et al., 2002]. They were placed at 705 km altitude above Earth's surface in near identical orbit with the NASA Earth Observing System (EOS) satellites Aqua and Terra, the French satellite Polarization and Anisotropy of Reflectance for Atmospheric Sciences coupled with Observations from a Lidar (PARASOL) and the EOS chemistry satellite, Aura. CALIPSO combines an active lidar instrument with passive infrared and visible imagers to examine the vertical structure and properties of thin clouds and aerosols. CloudSat has the first satellite-based 94 GHz nadir-looking cloud profiling radar that is more than 1000 times more sensitive than existing weather radars. It has 500 m vertical resolution, 1.4 km cross-track resolution and 1.7 km along-track resolution. Together CALIPSO and CloudSat are highly complementary and provide 3-D perspectives of clouds and aerosols.

[17] The CALIPSO platform has three instruments that are coaligned and nadir pointing sensors: (1) the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), (2) the Imaging Infrared Radiometer (IIR), and (3) the Wide Field Camera (WFC). CALIOP is a two wavelength (532 and 1064 nm) polarization sensitive lidar that provides high resolution (∼30 m to 60 m) vertical profiles over a horizontal spatial scale of 1/3 to 5 km. Figure 2 shows two CloudSat/CALIPSO overpasses over the southern peninsula of India. Note the narrow swath of the CALIPSO instrument which prevents detailed spatial analysis. The individual lidar returns of the CALIOP instrument on board CALIPSO contains information about the cloud base and cloud top altitudes, aerosol layers and the altitude of Earth's surface.

Figure 2.

CALIPSO overpasses in the Western Ghats. The 2048:00 UT overpass is for morning, and the 0832:00 UT overpass is for the afternoon. The underlying figure is from Google Earth. Google Earth imagery ©Google Inc. Used with permission.

[18] The Level 1 total attenuated backscattered coefficient data at 532 nm were used to determine the vertical distribution of clouds over the Western Ghats for 13 overpasses during the dry seasons (December–April) of 2007 and 2008. These observations provided (1) average cloud base heights during the dry season, in support of the cloud climatology developed using MODIS observations, and (2) the distribution of cloud top and base heights over the Ghats. CloudSat Observations provided corresponding cloud liquid water content horizontal and vertical distributions.

2.5. FNL Data Sets

[19] The National Center for Environmental Prediction (NCEP) global tropospheric final analysis fields (FNL) provides horizontal and vertical profiles of meteorological conditions over the Ghats region. The FNL fields, available at 1° × 1° resolution every 6 h, contain meteorological variables including temperature and relative humidity at 26 pressure levels ranging from 1000 hPa to 10 hPa. Surface values of the meteorological fields are available from the FNL files. The FNL data set was used as an independent comparison data set of cloud base heights derived from the CALIPSO satellite.

2.6. RAMS Modeling Data Sets

[20] We used the Regional Atmospheric Modeling System (RAMS) [Pielke et al., 1992] to simulate the sensitivity of cloud cover and precipitation to lowland deforestation over the montane regions in subregion 3 (Figure 3). We conducted two simulations to estimate this sensitivity: (1) for current land cover conditions (derived from the MODIS ecosystem database [Hansen et al., 2000] (also see Figure 1b)) and (2) for forested land cover conditions (based on potential vegetation for the region [Puri, 1960]).

Figure 3.

Land use scenarios used for the RAMS model simulations. (a) The 4 km outer grid for (left) forested and (right) current land cover condition and (b) the telescoped inner grid for the location shown by the white rectangle in Figures 3a (left) and 3a (right).

[21] The NCEP reanalysis [Kalnay et al., 1996] and the upper air and surface observations from the University Corporation for Atmospheric Research (UCAR) provided model initialization and nudging lateral boundary conditions. Note that a comprehensive understanding of the relationship between lower elevation deforestation and montane hydrometeorological conditions necessarily requires extensive simulations spread over multiple years and for both dry and wet seasons over all the three subregions of the Western Ghats. We emphasize that the purpose of the RAMS simulations were not to exactly reproduce the details of the real-world conditions that existed during March 2003 but to gain some insight into the sensitivity of lowland deforestation on cloud cover and rainfall over the montane forests. We simulated the month of March 2003 for two reasons: (1) it was the peak dry season month for the first year when both the MODIS Aqua and Terra satellite data were present and (2) the land cover map used for current land conditions were closest in time to the current land cover maps from MODIS. If we simulated a significantly later or earlier date the land cover maps would require corrections as land cover itself is likely to have changed with time, requiring newer land cover observations or modeling using a land cover change model [e.g., Pijanowski et al., 2007; Ray and Pijanowski, 2010; Ray et al., 2010a]. Also, as we wanted to explicitly simulate clouds the simulations required very high spatial resolution and we were forced to choose spatial resolution over the length of the simulations due to the computational costs involved. Future studies should conduct comprehensive simulations.

2.7. Precipitation Data Set

[22] We compared the relationship between climatological cloud frequency and precipitation and the simulated RAMS precipitation for current conditions using the 1 km resolution “WorldClim” monthly total precipitation data set [Hijmans et al., 2005]. This data set was created using precipitation records from 47,554 stations for the period 1950 to 2000 together with information of the latitude, longitude and 90 m Shuttle Radar Topography Mission (SRTM) near-global elevation data of each station to generate second-order spline fitted interpolated precipitation surfaces at the 1 km spatial resolution [Hijmans et al., 2005]. This precipitation data set is a significant improvement over previous near-global precipitation data sets such as the 10 min (18.5 km at the equator) Climate Research Unit (CRU) precipitation data set [New et al., 2002]. We found no significant station bias toward higher elevation over the Western Ghats [i.e., Hijmans et al., 2005, Figure 2a] and the errors in precipitation were reported as less than 10 mm/month over the Western Ghats [i.e., see Hijmans et al., 2005, Figure 3] allowing us to use this data in a regional scale study such as this. However, at a local scale the complex topography of the Western Ghats and the absence of a dense network of station data is likely to make the WorldClim data set less reliable.

3. Methodology

3.1. Construction of MODIS Cloud Cover Climatology

[23] The ∼1 km cloud cover climatology developed using individual scenes from the MODIS Aqua and Terra Platforms (together with LAI and elevation information) was used to determine the locations of tropical rain forests in the Western Ghats, to estimate the current environmental conditions at these locations, and to identify currently deforested locations that could potentially sustain these forests. Calibrated radiances of the 36 channels of the MODIS Level 1B satellite data were converted to either temperature or reflectance, depending on the wavelength. The 36-channel information was fed to a trained neural network [Berendes et al., 1999] to classify each of the pixels. While clouds are detected more easily over a dark target (such as dense forest) the classifier used has been tested for cloud detection over a variety of bright underlying surfaces such as polar ice and deserts and had pixel classification accuracies between 95 and 96% over these surfaces [Berendes et al., 1999]. The classification resulted in the identification of each pixel as either a water cloud (cumulus or stratus), cirrus ice cloud, aerosol, clear land or clear water at 1 km spatial resolution and we believe no significant bias was introduced between clouds detected over forested and deforested land covers. The classifier was not trained with mixed classes, such as multilayered clouds and thus assigns each pixel to only one of the above classes. Algorithms are currently being developed that can detect multilayer clouds within the same pixel [e.g., Joiner et al., 2010; Wind et al., 2010]. Due to nonfunctional and noisy detectors in Aqua channel 6 (1.628–1.652 μm), this particular channel was not available in the classifier for the afternoon overpasses. An Aqua-specific trained neural network with channel 6 excluded was used instead, with no apparent loss in accuracy. All the approximately 7200 Terra and Aqua MODIS scenes used in this study were classified in this manner with accuracy greater than 90% [Berendes et al., 1999].

[24] Water cloud pixels were extracted from each of the classified MODIS scenes and then averaged monthly twice daily (at 1030 and 1330 LT). The result was the generation of monthly morning and afternoon cloud cover climatology of the Western Ghats for a 6 year period. Each pixel was geolocated to a grid with spatial scale of 0.01° × 0.01° covering the region 8°N–14°N and 72°E–82°E. Monthly cloud frequency was then determined, defined as the number of times in a month a particular map location had cloud compared to the total number of times the map location was viewed by MODIS. For example if a map location was viewed by the satellite 12 times in a month for each of the 5 years and water clouds were detected 6 times every year for this month then the monthly water cloud frequency for this location was computed as 50%, similar to the method used by Ray et al. [2003] using the Geostationary Meteorological Satellite-5 imageries. This approach was repeated for each month during the 6 year study, and then mean monthly values were derived.

3.2. Construction of Leaf Area Index Climatology

[25] Six years of the Terra MODIS LAI product were used to develop the ∼1 km LAI climatology, also at 0.01° × 0.01° spatial resolution. First, only those LAI pixels that were completely cloud free were used. Second, the MODIS LAI algorithm utilizes a radiative transfer algorithm as well as empirical methods to determine LAI. Only those LAI values that were retrieved by the science team using the primary radiative transfer algorithm were used in the current study. Since LAI data consisted of a 8 day composite product, data coverage occurred about four times a month for any grid cell, leading to a maximum 24 times of coverage during the 6 year study period for any given month and a maximum observation of 260 times per grid cell was possible (lower number of 260 was due to some missing LAI data sets). However, restrictions imposed regarding LAI data quality limited the number of high quality values available; in the worst case only 10 of the possible 24 sets of LAI data could be utilized in creating the monthly LAI climatology at certain grid cells for use in identifying tropical rain forest grid cells.

3.3. Relationship of Cloud Frequency and Precipitation

[26] Creation of cloud cover frequency climatology provides hydrometeorological information for the Western Ghats at a spatial resolution not generally available from precipitation data sets that are either from specific point locations (rain gauge measurements) or at coarser resolution (satellite derived). We evaluated whether climatological cloud frequency is a good indicator of climatological rainfall using the best available spatial precipitation data, called WorldClim [Hijmans et al., 2005]. The spatial resolution of the WorldClim precipitation data set is 30 arc s (∼0.008°) which is around 0.86 km2 at the equator and less elsewhere whereas our cloud frequency climatology was created at 36 arc sec (0.01°). Thus the spatial resolutions of both data sets were nearly identical. Both data sets are climatological. We extracted the rainfall information from the global WorldClim data set for the study region and collocated the rainfall and water cloud frequency using a nearest neighbor approach. Next we linearly regressed the cloud frequency with rainfall and determined the Pearson's correlation coefficient. Significant correlation coefficients would indicate that cloud cover is a good indicator of rainfall over the Western Ghats; fitted regressions would provide the exact relationship between the two variables.

3.4. Estimating Cloud Base Heights

[27] To determine the fraction of clouds intersecting the Western Ghats we used the National Center for Environmental Prediction Global Tropospheric Analysis fields (FNL) to estimate the Lifting Condensation Level (LCL). The LCL is the height at which unsaturated air lifted adiabatically becomes saturated. At this height the air temperature (T) becomes equal to the dew point temperature (Td) and clouds form. An estimate of cloud base height is obtained from determining the height at which T = Td from the FNL data set which most closely matches the satellite overpass time. Since the data sets are available only in 6 h increments and at 1° × 1° spatial resolution, there may be inaccuracies in the cloud base heights of 100–300 m [Welch et al., 2008; Nair et al., 2008].

[28] Improved estimates of cloud base heights are available using CALIPSO satellite observations, but these are limited to overpass times. Figure 2 shows two CALIPSO overpasses over the Western Ghats region. The morning overpass at 2048 UT crosses only a small region of the Ghats, whereas the afternoon overpass 0832 UT crosses a much more extensive region. This is fortuitous because the orographic cloud cover is much more extensive during the afternoon hours during the dry season. Also note the very small footprint of the CALIPSO observations and that the repeat cycle is sixteen days. Therefore, while CALIPSO data is an excellent resource, it is sparsely available over the region of interest.

[29] When using the CALIPSO observations to estimate cloud base heights over the Western Ghats, it is important to recognize that the CALIOP retrievals may be contaminated near the surface. The 1 km spatial resolution digital elevation data smoothes out many of the high contour regions, but the sharp mountain peaks extend above this height. Therefore, the backscatter signal was removed from all of the data below 1 km. Second, the Ghats region often has multilevel cloudiness. However, this study is primarily concerned with low level orographic clouds that intersect the mountains of the Western Ghats and provide direct moisture input to the ecosystems. Therefore the present study is focused only on the backscatter signal from clouds below 4 km. Note that the Western Ghats themselves average around 1200 in elevation (Figure 1b), but the highest peak is nearly at 2.7 km. Therefore, data close to ground (within 1 km of the local digital elevation height) is eliminated from processing as potentially contaminated by surface reflections.

[30] For clouds identified as being below 4 km, cloud bases are computed and then compared with those LCL values obtained from the FNL data set. Due to the narrow swath of the CALIPSO instrument (333 m), the limited temporal coverage and the limited repeat cycle, detailed cloud base height climatologies are not yet possible using this data set.

3.5. Identification of Average Cloud Cover Over Locations of High LAI Values

[31] The Western Ghats region being highly cloudy all grid cells had some missing LAI data based on the data quality indicator; the minimum number of times that data was absent was 50 (out of a maximum of 260 observations possible per grid cell for the 6 year study period). In general we found that cloud frequency increased as the frequency of missing LAI observations increases (Figure 4). This effect was more clearly noticeable for nearly all the months at 1030 LT (Terra overpass time). What this observation signifies is that one cannot simultaneously determine LAI and cloud frequency without introducing a significant bias toward the LAI values of less frequently cloudy grid cells. In the absence of being able to determine an unbiased relationship between LAI and cloud frequency we prescribed LAI thresholds for the Western Ghats tropical rain forests on the basis of published results and expert knowledge.

Figure 4.

Monthly variation of average cloud frequency (y axis) as a function of the number of times the grid cell did not have a reported LAI value (x axis). Cloud frequency and LAI are both from 1030 LT Terra MODIS.

[32] Previous studies in wetter montane cloud forests from Costa Rica showed that during the rainy season, LAI values between 5 and 6 were common [Lawton and Putz, 1988]. Since the Western Ghats tropical rain forests are drier (R. O. Lawton, personal communication, 2005) LAI values of 4.0 for the tropical rain forests in the rainy monsoon season is reasonable for this region. Our analysis in fact showed that there were very few cells with LAI values above 5.0. In the dry season LAI values of 4.0 would then likely be the upper LAI limit for the tropical rain forests of the Western Ghats. Therefore, the threshold that locations above 1800 m in elevation with LAI values ≥4.0 for more than 75% of the time are locations where tropical rain forests are hydrometeorologically stable is a very restrictive criteria to identify these forest types. We choose this very restrictive season-invariant threshold to identify monthly cloud cover frequency over tropical rain forest grid cells. We also used two additional LAI thresholds of ≥3.75 and ≥4.25 for sensitivity analysis.

[33] Determination of the locations suitable for sustaining and regenerating secondary tropical rain forests in the Western Ghats first uses the information of elevation, cloud cover frequency and LAI (Figure 5). Average cloud cover frequency values from Terra and Aqua MODIS data were identified for grid cells above 1800 m and having LAI greater than 4.0. The 0.01° × 0.01° cells that satisfy these thresholds are denoted as monthly mean High LAI Cloud Cover (HLAICC) cells that have dense forests and cloud interception capabilities. Cloud interception by vegetation from orographic clouds leads to significant “horizontal precipitation” [Bruijnzeel and Proctor, 1995] and the absence of any dry season at these locations [Cavelier and Goldstein, 1989]. Note that while at low elevations (i.e., less than 1800 m) there are large regions with LAI ≥ 4.0, it is primarily those locations with LAI ≥ 4.0 above 1800 m that have the potential to receive moisture input directly as cloud water interception and form tropical rain forests.

Figure 5.

Methodology used to create cloud and LAI climatology and procedure to determine potential locations of tropical wet forests.

3.6. Identification of Potential Locations for Sustainable Tropical Rain Forests

[34] Tropical Rain Forests (TRF) are found in those cells which have the characteristic monthly mean HLAICC values that would provide sufficient intercepted rainfall to maintain these regions as tropical rain forests. In this study these values of intercepting cloud cover over locations of high (≥4.0) LAI values are used as a surrogate of sustaining moisture input. It is assumed that locations in the study region at elevations above 1800 m that have cloud cover greater than the mean monthly HLAICC values receive sufficient moisture input to sustain the TRFs. This is a reasonable assumption since the forests are found in these locations. Therefore, HLAICC values (that are a combined elevation and cloud cover value) are utilized as “threshold” cloud cover values for adequate moisture in TRF regions.

[35] Locations that have cloud cover above the “threshold” for each month of the year are identified as regions ideal for “hydrometeorologically stress free” TRFs. On the other hand, some locations above 1800 m seldom have the HLAICC “threshold” cloud cover. These locations then are classified either as being “hydrometeorogically stressed tropical rain forests” or already devoid of TRFs, having regenerated into another forest or vegetation cover category or, if deforested, unlikely to regenerate TRFs. Since HLAICC “threshold” cloud cover information was available twice daily during the MODIS overpasses, the variations in the stress levels can be estimated for the potential locations of tropical rain forests between morning and afternoon.

3.7. Simulations of Clouds and Precipitation

[36] In order to answer the question of sensitivity of hydrometeorological conditions over the montane rain forests in response to changed land cover at lower elevations similar to the Ray et al. [2006a, 2009, 2010b] study over Costa Rica's Monteverde cloud forests, we used the RAMS model to simulate conditions for March 2003 for the two land cover configurations, i.e., forested and current as of 2000, (Figure 3). The sensitivity experiments were done only for the southernmost subregion 3.

[37] Both the model simulations had identical telescoped nested grid configurations (Figure 3) with 4 km and 1 km grid spacing for the outer and inner nested grids, respectively. In the vertical, a stretched grid that varied from 20 m near the surface to 750 m higher up was used, with the top at 24 km using 48 vertical levels. A hybrid grid system in the vertical was used with terrain-following sigma z coordinates at lower atmospheric levels blended to isentropic coordinates at 6 km. Both simulations used identical initial soil moisture and soil temperature conditions using 24 soil layers going to a depth of 2.5 m.

[38] Only the lateral boundaries of the outer grid were nudged with an exponentially decreasing nudging strength and with a timescale of 900 s along five grid points. Because of the fine grid spacing, microphysical processes were explicitly represented [Walko et al., 2000] and allowed the simulation of clouds; the atmospheric radiative transfer scheme of Harrington and Olsson [2001] was used. Only the warm phase processes were prognosed from conservation equations that included determining advective, diffusive, and precipitation tendencies, and source terms resulting from interactions between rain and other forms of water substance. A deformation scheme was used to represent horizontal diffusion, while the vertical diffusion was parameterized using the Mellor and Yamada [1982] scheme.

[39] In the forested modeled scenario we forested the lowland regions to the west of the Western Ghats to evergreen broadleaf forests wherever the current land cover was not a forest. At each of these locations we prescribed a LAI value of 5.1 [Lawton and Putz, 1988] to create a dense forest cover. At all other locations we prescribed the LAI similar to those observed using the MODIS LAI data for these grid cells for the month of March 2003. In the current scenario we provided each grid cell the LAI value determined from the MODIS LAI product.

[40] The computation of vegetation fractions in each grid cell from the land cover map involved a couple of steps. First we cross walked between the land cover types given in the MODIS land cover map and the model's internal land cover map. Next we determined the geolocation of each MODIS land cover pixel with respect to a RAMS grid cell. For the inner grid that had a 1 km spatial resolution we did not figure out the vegetation fraction since both the MODIS map and the model grid cells were at 1 km spatial resolution. However, for the 4 km grid cell of the outer grid we first figured out the MODIS grid cells that were contained within each of the 4 km RAMS grid cells and then determined the vegetation fraction in each model grid cell. We allowed the model to use its in-built albedo corresponding to the land cover map that we updated. We however updated the rooting depths in the model (R. O. Lawton, personal communication, 2005) and provided the model with representative soil types [Webb et al., 1992; Gerakis and Baer, 1999].

[41] Associated with the land cover of each grid cell are land cover specific albedo, emissivity, roughness length and rooting depths that in turn determine the sensible heat and moisture fluxes from the vegetation to canopy, vegetation temperature and the exchange of heat and precipitation between vegetation and atmosphere. Other than the land cover related differences between the two models (i.e., one was mostly forested and the other had current land cover conditions), both simulations were provided exactly similar initial hydrometerological starting conditions. Both models were also identically nudged along their lateral boundaries (only five points) to match observed initial atmospheric and temporally varying lateral hydrometerological conditions at the model domains edges (and with exponentially decreasing strength) with the UCAR and NCEP reanalysis data. However, since they were allowed to simulate the hydrometerological conditions elsewhere according to the model physics, in the interior regions, where the inner nested grid was present, the effects of prescribed land cover was simulated.

[42] The RAMS model uses the LEAF2 land vegetation model [Walko et al., 2000] to represent the various land surface processes. The effect of land cover change on the hydrometeorology over the montane regions thus could be determined via subtracting the two simulations of hydrometeorological variables for the inner nested grid.

4. Results

4.1. Spatial Variations in Monthly Cloud Cover

[43] Figure 6a shows the monthly averaged water cloud climatology for subregion 1. The monthly water cloud climatology at 1030 LT (i.e., developed from Terra MODIS) and similar results at 1330 LT (i.e., developed using Aqua MODIS) are shown. Several salient features of the water cloud formation over this section of the Western Ghats stands out. First, there is a clear tendency for higher cloud cover over the higher elevations as shown by the contour lines. Significant and high positive correlation between cloud cover frequency and elevation at all the three regions for all the months and at both times of the day signifies this effect (Table 2).

Figure 6.

(a) Monthly variations of cloud cover frequency (%) at subregion 1 at 1030 LT and 1330 LT. Cloud frequency over water is set to values below 0. (b) Same as Figure 6a but for subregion 2.

Figure 6.

(continued)

Figure 6.

(continued)

Figure 6.

(continued)

Table 2. Correlations Between Cloud Cover and Elevation in the Three Study Regions for the Terra (Morning) and Aqua (Afternoon) Overpasses
MonthRegion 1Region 2Region 3
TerraAquaTerraAquaTerraAqua
Jan0.170.560.280.640.250.47
Feb0.450.600.490.710.420.54
Mar0.300.460.410.590.590.45
Apr0.220.350.390.480.560.45
May0.250.460.170.420.410.48
Jun0.330.520.380.660.480.67
Jul0.410.450.360.500.430.44
Aug0.470.590.300.570.470.61
Sep0.420.550.500.660.520.63
Oct0.610.570.630.740.580.67
Nov0.560.560.590.730.350.52
Dec0.370.500.260.590.280.44

[44] The spatial cloud frequency patterns show that this is especially true for the months of January through May irrespective of the time of the day (Figure 6a). In particular consider the high elevation regions on the south and west sides. At 1330 LT the high elevation regions have cloud covers of about 15% during January and February, increasing to 40–60% in April. The increase in cloud cover frequency from January to June is due to the transition from dry season to the wet monsoon season that normally starts in the last week of May and ends normally in August. Indeed, note the widespread values of cloud cover in the range of 40–50% in August. There is another notable peak in cloud cover at higher elevations of 40–60% in October. This study fills in a major gap in the understanding of the cloud cover variability over the Western Ghats. Note the contrast in the morning cloud cover frequency from the afternoon cloud cover frequency for each month. In January and February morning cloud cover in most high elevations is less than 20%. There is higher cloud cover in the afternoon in all the dry season months of January through May, but during the monsoon months of June onwards there are no clear differences. This leads to the important conclusion that cloud cover increases from morning to afternoon during the dry season over the Western Ghats, while being relatively constant throughout the entire region during the monsoon season at 20–30%. The montane forests in the Western Ghats receive an important major daytime source of moisture in the dry season during the afternoon hours from the combination of convective precipitation and “horizontal precipitation.” Note, however that the 1030 LT and 1330 LT MODIS overpasses may underestimate cloud cover due to diurnal convection which may peak in intensity well after 1330 LT.

[45] Figure 6b shows results for subregion 2 which has a more complicated and steeper topography. While subregion 1 is almost a continuous stretch of high elevation, subregion 2 is made up of intervening mountains and valleys. The high elevation regions have cloud covers similar to those in subregion 1, but there are several instances in which lower elevations also have high cloud cover frequency. In the dry season this occurs primarily on the east side of the Ghats, whereas in the wet season this occurs primarily on the west side. Similar to subregion 1, cloud cover increases from the dry season to the wet season. Finally, higher cloud cover also occurs during the afternoon in the dry season months. Subregion 3 shows similar variations in cloud cover (not shown). Correlations between cloud cover and elevation for the morning and afternoon overpasses are given in Table 2 averaged over each of the three regions. Correlations while not very different between the three subregions are overall lowest for subregion 1 and highest for subregion 2, with values ranging from about 0.2 to about 0.7.

4.2. Relationship Between Climatological Cloud Cover Frequency and Precipitation

[46] Data collected from various cloud forest locations around the world (Table 1) shows that horizontal precipitation from clouds can account from 1.4% to 99% of the total precipitation with significant variations even for the same site. Bruijnzeel and Proctor [1993] note values for up to 14–18% and 15–100% of the total precipitation during the wet and dry seasons, respectively. While over the Western Ghats there is no known direct measurement of horizontal precipitation, Bruijnzeel et al. [2011] computed that between 10% and 15% of the mean annual total water input was from horizontal precipitation using a statistical approach. When the contribution from horizontal precipitation is added there is no dry season in the cloud forests leading to their evergreen nature (the reasoning behind using our seasonal invariant LAI threshold value). At some locations horizontal interception increases with elevation [Cavelier and Goldstein, 1989; Veneklaas and van Ek, 1990; Holder, 2004] while at others it decreases between sites at the same elevation [Clark et al., 2000]. The variations are obviously strongly dependent on other factors as well such as windward versus leeward side and distance from the continental divide [Lawton et al., 2011].

[47] While higher cloud cover indicates higher cloud water interception the Western Ghats also receive direct rainfall. The Pearson's correlation coefficient was determined between monthly climatological precipitation and climatological cloud frequency to estimate the strength of linear dependence between the two variables for morning and afternoon hours (Table 3). In general the correlation between precipitation and cloud frequency is positive and significant at p = 0.0000. Since we know that cloud cover is the necessary cause for precipitation, significant correlation coefficient also shows that our assumption of cloud frequency as a good indicator for total (intercepted/horizontal + direct/rainfall) moisture input to tropical rain forests is correct. Overall, region 3 had higher correlation coefficient followed by region 1 and last by region 2. Part of the reason for this variation between the regions are due to the differences in topographic complexity and size of the study regions (number of sample points are denoted in parenthesis in Table 3); region 2 is the largest and region 3 the smallest study region. For region 3 the correlation values were almost always higher at 1030 LT indicating that the 1030 morning hour monthly cloud frequency was a better predictor of precipitation. The only exception was October when the afternoon correlation coefficient was slightly higher. In region 2 the correlations were higher between May and October and for the month of December at 1030 LT, whereas for region 1 any pattern was absent.

Table 3. Correlation and Linear Regression Between Precipitation And Cloud Frequency for the Three Study Regions at 1330 and 1030 LT
MonthCorrelationaRegressionb
Region 1Region 2Region 3Region 1Region 2Region 3
  • a

    The total number of points used are 18,695 to 18,718 data points for region 1, 22,954 to 22,983 data points for region 2, and 3524 to 3535 data points for region 3.

  • b

    P, precipitation (in mm); C, cloud frequency (in %).

  • c

    Here p is not significant.

Around 1330 LT (Aqua Overpass Time)
Jan0.490.330.49P = 0.50C + 1.40P = 0.24C + 17.33P = 0.41C + 23.04
Feb0.450.100.29P = 0.36C + 6.83P = 0.13C + 25.89P = 0.22C + 26.86
Mar0.420.29−0.01cP = 0.34C + 12.46P = 0.53C + 39.37P = −0.01C + 50.46
Apr0.460.480.43P = 0.83C + 68.87P = 1.42C + 87.50P = 0.61C + 93.62
May0.250.550.57P = 1.45C + 139.06P = 4.45C + 90.94P = 1.56C + 88.87
Jun0.270.240.57P = 10.14C + 93.02P = 5.92C + 176.27P = 3.58C + 100.44
Jul0.500.510.70P = 26.78C – 174.06P = 13.26C + 0.51P = 4.72C + 35.62
Aug0.320.420.71P = 5.77C + 72.51P = 6.17C + 74.35P = 2.52C + 29.91
Sep0.350.230.54P = 3.19C + 67.96P = 2.13C + 117.67P = 1.55C + 56.69
Oct0.420.140.48P = 2.03C + 154.29P = 0.77C + 229.26P = 1.08C + 189.53
Nov0.41−0.250.27P = 1.56C + 73.37P = −0.80C + 193.75P = 0.25C + 197.90
Dec0.590.620.64P = 1.44C + 15.66P = 0.90C + 50.64P = 0.90C + 65.01
 
Around 1030 LT (Terra Overpass Time)
Jan0.530.320.59P = 0.59C + 1.95P = 0.23C + 18.26P = 0.30C + 24.36
Feb0.52−0.230.53P = 0.54C + 4.53P = −0.38C + 31.04P = 0.32C + 24.99
Mar0.450.040.33P = 0.67C + 11.95P = 0.13C + 44.77P = 0.48C + 44.74
Apr0.430.310.67P = 1.12C + 69.19P = 1.50C + 93.08P = 1.32C + 87.55
May0.420.650.68P = 3.37C + 122.05P = 7.40C + 64.97P = 2.57C + 82.94
Jun0.460.600.65P = 15.60C – 30.14P = 14.33C + 28.87P = 4.62C + 80.74
Jul0.470.630.72P = 26.30C – 114.29P = 17.21C – 41.71P = 5.02C + 23.60
Aug0.440.760.75P = 8.08C + 30.36P = 9.02C + 18.56P = 3.51C + 25.51
Sep0.540.530.65P = 4.79C + 30.48P = 4.82C + 62.78P = 2.42C + 39.08
Oct0.270.290.47P = 1.48C + 172.84P = 1.75C + 206.32P = 1.31C + 188.99
Nov0.13−0.320.43P = 0.50C + 96.47P = −0.98C + 197.18P = 0.31C + 195.82
Dec0.550.550.79P = 1.45C + 17.55P = 0.86C + 56.24P = 0.67C + 70.18

[48] In the next step we did a linear regression between cloud frequency and precipitation using a minimization of the chi-square merit function [Press et al., 1992]. The coefficients to two decimal points are given in Table 3. Figure 7 shows the WorldClim total monthly precipitation for the two regions for the dry season months of January to March and the monsoon season months of July to September. The effect of spline interpolation used by Hijmans et al. [2005] to create this precipitation data and the coarse resolution of the data are clearly visible. However, there is no bias toward higher elevation in Figure 7. The precipitation patterns and cloud frequency patterns clearly match in most of the months.

Figure 7.

Total monthly precipitation (in mm) for region 1 and region 2 (see Figure 1) for the dry season months of January, February and March and the monsoon season months of July, August and September. Note that each month has a different precipitation scale for clearly showing variations in precipitation patterns. Precipitation over water is set to values below 0.

Figure 7.

(continued)

[49] For region 1 the cloud frequency is higher from the southeastern quadrant in January and over the southeastern parts of the Western Ghats whereas in February higher cloud frequency is mostly over the higher elevations and in March the cloud frequency pattern changes to higher values from the west and over the higher elevations. The precipitation patterns is also higher from the southeast and over the higher elevations in January and in February the higher values are over the Western Ghats whereas in March the high precipitation is over the Western Ghats as well as from the west with a distinct rain shadow in the northeast which is also present in the 1030 LT cloud frequency.

[50] For region 2 in January cloud frequency is distinctly higher to the east and over the Western Ghats with a cloud frequency ‘shadow’ to the west. In February, the cloud frequency is higher over the Western Ghats and in March the cloud frequency is higher over the Western Ghats and at several locations to the west of the mountains. Precipitation in January for region 2 is higher at the southeastern edge and over some locations of the Western Ghats. In February the precipitation is higher over the mountains as well as to the west and southwest of the mountains with values decreasing to the north and northeast. In March precipitation is distinctly higher to the west and southwest of the Western Ghats with lower values over the mountains and rain shadow to the northeast.

[51] In the rainy season months of July, August and September both cloud frequency and precipitation are distinctly higher from the west and rain/cloud shadow seen in all these months but more distinctly in July and August. The similarity in cloud and precipitation patterns clearly points to the close relationship between cloud cover frequency climatology and precipitation climatology and while rainfall patterns clearly show the effects of coarse resolution and provide a monthly total precipitation value, cloud frequency has higher spatial resolution and provides a morning and an afternoon moisture input information to the tropical rain forests. This extra piece of information provides better understanding of the hydrometeorological processes occurring in the study region and hydrometeorological stability of the TRF of the Western Ghats.

[52] Since our classifier cannot detect multilayered clouds in the same pixel, water clouds underlying thick cirrus clouds are however likely missed. Techniques to detect multilayered clouds are not fully developed and are currently being investigated [e.g., Joiner et al., 2010; Wind et al., 2010]. An analysis of Figure 10 of Wind et al. [2010] however shows that for October 2008, 0.6 to 0.8 fraction of 5 km2 grid cells over southern India were cloudy, and the fraction of these cloudy pixels that had multilayered clouds was only around 0.2. Joiner et al. [2010] also showed similar values over southern India in July 2007 [see Joiner et al., 2010, Figure 14] with a detection accuracy of 83.4%. These results show that only small errors may have been introduced in our regression from nondetection of water clouds underlying thick cirrus clouds over southern India.

4.3. CALIPSO Observations

[53] The previous results (section 4.1) showed that high cloud cover tends to be associated with the higher elevation regions. We analyzed both the CloudSat and CALIPSO data initially and found that while both satellites agreed at locations where data were present in both, the CALIPSO data had significantly fewer missing data. Also since CloudSat and CALIPSO retrieved similar cloud base heights as shown in Table 4 we did not study the CloudSat data any further. Results from the CALIPSO instrument confirm that clouds in the Western Ghats have base heights that intersect the mountains. Figure 8 shows the vertical distribution of the clouds for 3 February 2008, 20 March 2007, 7 April 2008 and 7 May 2007. The late launch and 16 day repeat cycle of CloudSat/CALIPSO limited the number of scenes available during the study period. In Figure 8a the white background indicates clear air, whereas the red color shows cloud signal attenuation, with the strength of the attenuation signified by the darkness of the color. Ground level backscatter has been removed. Note that many low level clouds have cloud bases within the range of 1500 to 1800 m (above sea level). In some cases cloud base heights are found as low as 1200 m. Cloud base heights above 2500 m are unlikely to intersect most of the mountains (Figure 1a). These results are consistent for all the cases observed from CALIOP. A total of 38 individual clouds were identified with cloud bases at or below 1800 m in February 2008 out of a total of 81 clouds within 3 km. On this basis ∼47% clouds are expected to intersect the mountains. Approximately 39%, 42% and 70% of the clouds had cloud bases at or below 1800 m in March, April and May 2007, respectively. Also shown in Figure 8b are the various categories of the classified backscatter that were identified, from which cloud locations (light blue) below 4 km were extracted. The derived cloud bases from CALIOP were compared with the lifting condensation level (LCL) computed from National Center for Environmental Prediction (NCEP) Global Tropospheric Final Analysis fields (FNL) data [Kalnay et al., 1996]. The results from the limited amount of CALIOP observations and the FNL data set were comparable (black lines in Figure 8a), suggesting that the FNL data set is reliable for cloud base height determinations in the Western Ghats region. This is an important result, since Nair et al. [2008] showed that LCL estimates were unreliable for many regions in Costa Rica. The FNL fields showed that the direction of prevailing winds in the Ghats during the dry season is easterly, from the Bay of Bengal. Traveling across the extensive land surface from the east coast toward the west, large convective systems are generated which produce significant convective precipitation over the Ghats. During the monsoon or wet season (June to August) the wind becomes westerly from the Arabian Sea. Further analysis of the utility of FNL data in determining LCL is required as more CALIOP becomes available.

Figure 8.

(a) Vertical distribution of the clouds for 3 February 2008, 20 March 2007, 7 April 2008 and 7 May 2007 detected by CALIOP over the Western Ghats of India as cloud attenuated signal. The white background represents the clear air and the red color represents the cloud signal attenuation. The black lines are the LCL from the 1° FNL data. (b) The various categories of atmospheric features identified using the CALIOP instrument.

Table 4. Cloud Base Height Retrieved From CloudSat and CALIPSO on 7 May 2007
Latitude (deg)Longitude (deg)CloudSat-Retrieved Cloud Base Height (m)CALIPSO-Retrieved Cloud Base Height (m)
9.1277.1110001010
10.0276.514781480
10.376.4115101510
10.3976.2319982000
11.2076.1520182010

4.4. Identification of Potential Tropical Wet Forest Locations and Its Stability

[54] Figure 9 shows the variations in morning-afternoon and monthly cloud cover frequency thresholds associated with a priori chosen climatological LAI thresholds representative of conditions at elevations greater than 1800 m and conducive to supporting current tropical rain forests. We determined the cloud frequency corresponding to three LAI thresholds ± standard error (SE): 3.75 ± SE, 4.00 ± SE and 4.25 ± SE to look at the sensitivity of cloud frequency threshold to the a priori chosen LAI threshold values and its impact on the maps of potential tropical rain forest locations and their stability. Note that not only does the number of grid cells reduce for determining cloud frequency thresholds as LAI thresholds are raised, but cloud frequency itself also varies with the number of times LAI values are not reported (Figure 4). We determined the minimum and maximum values of cloud frequency (CF) thresholds by looking at cloud frequencies across the 3 above mentioned LAI thresholds:

equation image
equation image

where T stands for time (terra overpass or aqua overpass time) and M stands for month of the year. The resulting envelop of values (Figure 9) shows that CF thresholds vary both as a function of month and time of day (i.e., 1030 and 1330 LT). The solid lines correspond to CF associated with LAI ≥ 4.0. The envelope of the shaded areas while showing that CF expectedly shows some sensitivity to different LAI thresholds they do not generally overlap between the morning and afternoon overpass times and that over stable LAI regions (irrespective of the threshold) higher cloud cover was in the afternoon and in the wet season.

Figure 9.

Monthly morning and afternoon climatological cloud cover over locations with stable LAI. Values above 4.0 are shown in solid blank lines with the vertical bars showing the standard errors. The shaded envelope shows the minimum to maximum variations in cloud frequency associated with LAI ± SE variations of thresholds ≥3.75 ± SE, ≥4.00 ± SE and ≥4.25 ± SE.

[55] Corresponding to the LAI threshold of ≥4.0, 1030 LT cloud cover increases steadily during the morning hours from values of about 10% in January to peak cloud frequency values of around 40% in October, and then rapidly decreases for the remainder of the year. During the afternoon hour, around 1330 LT, the cloud cover frequency values increase in a similar fashion, from around 25% in January to 45% in October followed by a similar rapid decrease for the remainder of the year. The vertical bars of standard error associated with the MODIS morning overpass time do not generally intersect with those of the afternoon overpass time, demonstrating that there is a distinct morning and afternoon pattern of cloud cover at these locations of stable high LAI values. This was generally true even after considering the envelop of CF values associated with different LAI thresholds. Afternoon cloud cover is consistently larger than morning cloud cover by 5–15% for LAI threshold value of 4.0, with the largest differences in April and May and the smallest differences in September and October.

[56] It is assumed that all locations that have cloud frequency equal to or higher than those at these targeted locations also can receive sufficient moisture input for sustaining a similar tropical rain forest. For example, locations above 1800 m that have cloud frequency equal or above 10% in the morning in January and with afternoon cloud frequency 25% or above can sustain tropical rain forests in January. This assumption is based upon the observation that forests survive under these conditions as determined from the climatology of cloud cover over locations of stable high LAI values. However, forests require moisture input more than 1 month of the year. Locations that have cloud frequencies above the threshold for a large number of months are more likely to have sustainable tropical wet forests than those regions that have high cloud frequencies only in fewer months. For each month the thresholds are the values shown in Figure 9. As we change the threshold to CFmin(T, M) from CFLAI≥4.0(T, M) we relax the threshold to identify a grid cell as a tropical rain forest, whereas changing the threshold to CFmax(T, M) makes it more stringent.

[57] For each grid cell location above 1800 m the number of times that a location had cloud cover above the 3 monthly threshold frequency was tabulated and converted into an annual cloud frequency of occurrence map (Figure 10). Note the surprising result that the annual afternoon frequency of occurrence above the thresholds is lower (Figure 10a) than the morning values (Figure 10b) over the high elevation regions for LAI threshold of ≥4.0. What this means is that while cloud cover often is higher during the afternoon (Figure 6), it is much more variable and more infrequent above the required frequency for tropical rain forest hydrometeorological stability overall. Morning hours generally provide more stable hydrological conditions with lower vegetation stress than during the afternoon hours. Similar results were reported for Costa Rica [Lawton et al., 2001; Nair et al., 2003; Ray et al., 2006a]. For the higher CF threshold (i.e., CFmax(T, M)) the number of times a grid cell's CF was above the threshold was lower which implies more stressed tropical rain forest potential locations and vice versa for the CFmin(T, M) threshold.

Figure 10.

(a) Annual cloud frequency of occurrence in the afternoon from the Aqua overpass and (b) annual cloud frequency of occurrence in the morning from the Terra overpass. In each case close ups of subregions 1 and 2 are shown together with the locations of UNEP biodiversity hot spots identified with a plus symbol and for the three cloud frequency thresholds shown in Figure 9.

Figure 10.

(continued)

Figure 10.

(continued)

Figure 10.

(continued)

[58] We also tested the accuracy of our maps in detecting biodiversity hot spots by comparing our maps with 3 known biodiversity hot spot locations from the UNEP: Nilgiri Hills (11.37°N, 76.73°E), Palani Hills (10.23°N, 77.83°E) and Annamalai Hills (10.13°N, 77.23°E). These three locations are shown with a ‘+’ symbol in Figure 10. While the Nilgiri Hills biodiversity hot spot location was in subregion 1 the Annamalia Hills was in subregion 2 and the Palani Hills at an elevation lower than 1800 m was excluded from our mapping technique. The Nilgiri and Annamalia sites were identified as suitable locations of stable tropical rain forests both during the morning (Terra) and afternoon (Aqua) times using our mapping technique.

4.5. RAMS Simulations of the Effects of Land Cover Changes

[59] Dynamically downscaled [Ray et al., 2010b] sensitivity simulations were performed for the month of March 2003, to determine the effects of lowland deforestation. Figure 11 shows that for the current land cover conditions higher total March precipitation occurred than for a completely forested land cover environment condition. We averaged the precipitation over the entire outer coarser grid as well as the nested 1 km spatial resolution inner grid. Averaged over the RAMS coarse grid, the precipitation in the dry season month of March 2003 was about 33 mm of rainfall for current land cover condition. If the widespread deforestation currently experienced in southern India were replaced with forests, rainfall would decrease in the region to about 23 mm. Rainfall however was higher over the higher elevations as expected and consistent with our observations of higher cloud cover. At these higher elevations are the forests and next we averaged the precipitation over these forests using the simulations at the 1 km inner nested grid. We found that currently these forests receive around 121 mm of precipitation (March 2003) but if the entire region was converted to forests the precipitation over these montane forests would decrease to about 91 mm. In both the land cover scenarios we found that the cloud banks occasionally intersected the higher elevation regions, but cloud thickness generally was larger for the scenario of current land cover scenario. A qualitative comparison with the WorldClim March precipitation shows that similar to the simulated precipitation, the precipitation pattern shows higher values at higher elevations, though the amounts are much lower. We did not proceed any further and conduct quantitative comparisons because the simulated precipitation was for a specific year whereas the WorldClim data is a climatology.

Figure 11.

Total precipitation simulated for the month of March 2003 for 4 km (a) outer grid and (b) inner grid for (left) forested conditions and (right) current conditions. (c) Climatological precipitation is given. Note the different scale bars for the simulated precipitation and the climatological precipitation.

[60] To determine why large-scale lowland deforestation would result in higher domain averaged monthly precipitation (for March 2003), we analyzed the diurnal variation of several variables that connects the land surface with the atmosphere above (Figure 12). The model simulations show that deforestation increases 2 m air temperature and sensible heat fluxes but decreases the evapotranspiration rate and latent heat fluxes. The differences were largest during peak daytime conditions around 8 UTC (1330 LT). The domain averaged 2 m air temperature and sensible heat fluxes were nearly 4°C and 150 W m−2 higher respectively at this time due to lowland deforestation. On the other hand evapotranspiration rates and latent heat fluxes were only about half the forested condition values of 0.002 mm hr−1 and 150 W m−2 respectively. The diurnal distribution of these four land surface related variables thus shows that deforestation results in a larger proportion of absorbed solar radiation getting partitioned into surface sensible heat flux. The higher sensible heat flux and lowered evapotranspiration rates associated with deforestation in turn results in higher vertical velocity between 0 and ∼1 km above the ground and a deeper development of the planetary boundary layer. The net result is slightly higher late afternoon cloudiness and convective afternoon precipitation due to lowland deforestation, as shown by the higher afternoon peak in precipitation rates. The development of higher planetary boundary layer due to higher sensible heat flux and lower evapotranspiration rates is a well-known mechanism [Pan and Marht, 1987; McCumber and Pielke, 1981]. The changes in temperature and surface energy fluxes due to deforestation are also consistent with our previous studies from Costa Rica [Ray et al., 2006a, 2009, 2010b] and SW Australia [Ray et al., 2003]. The slightly higher precipitation rates associated with deforestation results in the maintenance of slightly higher soil moisture conditions though both simulations were started off with identical soil moisture and temperature conditions. Note that the overall evapotranspiration rate however decreases due to deforestation. Also note that during early morning, late afternoon and nighttime, higher cloud cover is present for the forested conditions but there is little rainfall at these times.

Figure 12.

Diurnal variations of monthly averaged physical variables for the two model scenarios: forested (solid lines) and current (dashed lines). Note that the x axis shows time of day in UTC units. The local time is UTC + 5 h 30 min. Thus 1000 UTC is 1530 LT or 0330 P.M.

[61] Due to the significance of cloud base heights to cloud forests we also analyzed in greater details the simulated cloud base heights in the two models (Figure 13). We found that the cloud base heights were different in the southwestern part of the two models with a complicated pattern at other locations. Till around 0800 LT, at many higher elevation locations with peaks at around 1100 m, in both the models, there were clouds with bases comparable to the height of the mountains (i.e., clouds were intersecting the mountains). By late morning around 1000 LT the cloud base heights for the current land cover condition was generally higher than the topography, whereas for the forested land cover scenario the cloud base heights were still intersecting the mountains at many locations. Between 1000 LT and 1300 LT cloud base heights continued to rise and stopped intersecting the mountains at most locations in both the models expect in the southwestern part of the forested land cover modeled scenario. On the eastern and western slopes of the model domain the pattern was more complex at all times but often cloud bases were lower for the forested land cover scenario. However, note that the cloud bases were not sufficiently low to intersect the terrain in these locations between 1000 LT and 1500 LT and had cloud base heights between 1400 m to 2100 m comparable to the CALIPSO observed cloud base heights for March 2007 (Figure 9). From 1600 LT cloud base heights again began to lower for the forested land cover scenario and intersection with the topography gradually increases as the evening progresses. For the current land cover scenario the bases get low enough to intersect the topography only from around 1900 LT.

Figure 13.

Spatial variations of cloud base heights (m above sea level) at 1000, 1200 and 1500 LT for forested and current land cover conditions. Only clouds with bases ≤2750 m are shown.

Figure 13.

(continued)

5. Discussions and Conclusions

[62] The UNEP-WCMC strategic plan seeks to reduce the loss of biodiversity worldwide. However, for many unique ecosystems, such as the montane cloud forests and tropical rain forests, the locations and current conditions over these ecosystems are not well known. Unlike the hydrometeorologically well-studied montane cloud forests such as found in Monteverde Costa Rica there has been no detailed study that characterizes the cloud cover over the Western Ghats cloud forests. Rain forests require high water inputs.

[63] One purpose of the present paper is to reduce the uncertainty in the potential locations of these remote forests by conducting the first detailed study of the cloud cover climatology over the Western Ghats. Potential locations of tropical rain forests in the Western Ghats are identified using MODIS satellite cloud cover and LAI products together with elevation information. The locations of highest cloud cover are potential and likely sites of cloud forests. This was accomplished by developing a 6 year cloud climatology at two times of the day. CALIPSO data over the Western Ghats shows that a significant portion of the cloud cover in the dry season intersects the higher mountain elevations, in agreement with the assumptions made in the cloud climatology.

[64] The spatial distributions of cloud cover shows that (1) cloud cover increases from the dry season to the wet season, (2) cloud cover increases with elevation, (3) cloud cover is strongly correlated with rainfall, and (4) there is higher cloud cover in the afternoon. This is consistent with the findings of Sahany et al. [2010] who found that over the Western Ghats the rainfall peaks during late afternoon/early evening during the summer season using 0.25° × 0.25° Tropical Rainfall Measuring Mission 3 hourly, rainfall product. Note however that in this study since our afternoon cloud cover information is from the Aqua-MODIS with overpass times around 1330 LT the late afternoon cloud cover cannot be determined though our RAMS simulations suggests that the peak rainfall is around 1500 LT (0930 UTC) for the innermost 1 km spatial resolution domain. Future studies should investigate late afternoon and nighttime cloud cover.

[65] Locations with high stable LAI values above 4.0 were identified at elevations above 1800 m. These are potential locations that sustain stable tropical rain forests. Next using the thresholds of cloud cover determined at these locations other similar high cloud cover locations were identified that can sustain similar tropical rain forests. Not all locations had the required threshold cloud cover over all the months. Therefore the frequency of occurrence of cloud cover above the threshold was determined. The primary assumption is that higher the cloud cover frequency of occurrence above the threshold, higher is the possibility that the location has sufficient water input to sustain the tropical rain forests. On this basis we generated maps showing the locations which are most likely to sustain the tropical rain forests from the hydrometeorological point of view. The technique successfully identified the two UNEP reported biodiversity hot spots in the study region.

[66] Finally we addressed the issue of sensitivity of lowland deforestation to cloud cover and precipitation over the montane forests using numerical simulations. Our results unlike those from Costa Rica [Ray et al., 2010b] and Central America [Ray et al., 2006b] shows that lowland deforestation has increased precipitation over the montane forests similar to the results of Van der Molen [2002] from Puerto Rico. Our simulations suggest that lowland deforestation triggers stronger convective rainfall events though our analysis of cloud base heights suggests reduced horizontal precipitation. However, note that the numerical simulations were conducted only over the southernmost study region and only for the month of 2003. Follow up validated studies of the entire Western Ghats and for a longer time frame is necessary to establish whether this would be true for other locations as well as for other times, i.e., wet season months. Interestingly however, Murugan et al. [2009] observed an increasing trend in March precipitation over three of four sites they studied in the Western Ghats for the month of March which they suggest could be due to lowland land cover changes (i.e., deforestation). Unfortunately there are no in situ measurements of horizontal precipitation over the Western Ghats for comparison. Also note, that the accuracy of dynamical downscaling of land cover change impacts could be severely affected from the quality of the atmospheric information used especially for simulations in remote locations [Ray et al., 2009, 2010b]. However, since the region of South West India had regular launches of radiosondes in March 2003 we believe that the atmospheric profile was sampled well and thus our simulations were in agreement with the observations of Murugan et al. [2009]. The surface temperature, surface energy fluxes, evapotranspiration rates and cloud base heights results are however similar to our previous results from Costa Rica [Ray et al., 2006a, 2009, 2010b].

[67] There are several limitations and assumptions in our cloud cover–LAI method for tropical rain forest identification that needs to be addressed for more precise estimation of the potential locations of the tropical rain forests. First, cloud cover climatology was used as a surrogate for moisture input, lacking other direct means of moisture availability in these remote locations at high spatial and temporal resolution. However, we have shown using the CALIPSO observations and the FNL data that clouds do indeed intersect the mountains of the Western Ghats and provide an important source of moisture via interception similar to other locations (Table 1). Bruijnzeel et al. [2011] modeled between 10 and 15% annual cloud water interception over the Western Ghats region. We also showed that cloud cover is strongly correlated with precipitation and linear regressions between morning and afternoon cloud cover were also developed similar to Ray et al. [2006b]. Second, nighttime cloudiness is not included in the present analysis at all and yet nocturnal inversions could significantly change cloud cover with important implications for moisture input into the montane forests. The simulations however show that at least for March 2003 there is little nighttime precipitation. Third, the tropical rain forests were not ground truthed. However, we used independent biodiversity hot spot location and found that our method successfully identified these locations. Locations with stable LAI values of around 4.0 over the 6 year period of the study were assumed to be likely locations of the primary tropical rain forests. Indeed, previous studies have shown that within 5–6 years after complete deforestation, if left alone a region may develop secondary forests with LAI values of the primary forests [Lawton and Putz, 1988]. Asner et al. [2009] have shown that in the humid tropics secondary growth of forests is large.

[68] Nagendra and Gadgil [1999] demonstrated that using only satellite derived NDVI, ecotypes in the Western Ghats could be discriminated. Future studies might utilize a time series of classified high resolution satellite imagery, multilayered clouds and nighttime clouds to identify the primary tropical rain forests. The method presented here to determine the stability of potential locations for tropical rain forests is based only on hydrometeorological conditions derived from cloud cover statistics though our maps successfully identified UNEP identified biodiversity hot spots. However, in reality the stability of a region is determined from multiple other factors, such as fire hazards [Kodandapani et al., 2004], deforestation pressure [Jha et al., 2000; Nagendra et al., 2009] and changes in global climate [Malcolm et al., 2006].

[69] The RAMS simulations for the meteorological conditions of March 2003 demonstrates that partly deforested conditions (i.e., current land cover conditions) produces more precipitation than a completely forested condition though forested conditions had more frequent occurrences of clouds intersecting the topography (i.e., horizontal precipitation). These results are consistent with many observational studies which report increased cloud cover and precipitation over deforested regions in Amazonia during the dry season [Rabin et al., 1990; Cutrim et al., 1995; Negri et al., 2004; Costa et al., 2007] but contrary to the findings from Costa Rica [e.g., Ray et al., 2010b]. Modeling studies by Wang et al. [2000] also find enhanced cumulus cloud cover and enhanced deep convection over deforested patches in Amazonia. Our study seems to suggest that if the environment is convectively unstable tropical montane regions may receive higher rainfall. Further studies to determine whether similar results hold for drier initial conditions in the other subregions, as well as for other months and more years, is required before we can come to any general conclusion that deforestation leads to increasing precipitation over the entire Western Ghats. What we have shown by our simulations is only that the rainfall amounts and horizontal precipitation over the montane regions are highly sensitive to large-scale deforestation at lower elevations.

Acknowledgments

[70] We thank the three anonymous reviewers and the associate editor whose comments for further analysis led to a significant improvement in the quality of this manuscript. We also thank Denise Berendes at UAH for arranging the use of computer resources for model simulations and data storage. The author contributions are as follows: D.K.R. designed the research; D.K.R. ran the models; D.K.R., V.S.M. and R.M.W. collected and analyzed the satellite data; and D.K.R., V.S.M. and R.M.W. wrote the paper.