Abstract

This study details two unique methods to quantify cloud-immersion statistics for tropical montane cloud forests (TMCFs). The first technique uses a new algorithm for determining cloud-base height using Moderate Resolution Imaging Spectroradiometer (MODIS) cloud products, and the second method uses numerical atmospheric simulation along with geostationary satellite data. Cloud-immersion statistics are determined using MODIS data for March 2003 over the study region consisting of Costa Rica, southern Nicaragua, and northern Panama. Comparison with known locations of cloud forests in northern Costa Rica shows that the MODIS-derived cloud-immersion maps successfully identify known cloud-forest locations in the United Nations Environment Programme (UNEP) World Conservation Monitoring Centre (WCMC) database. Large connected regions of cloud immersion are observed in regions in which the trade wind flow is directly impinging upon the mountain slopes; in areas in which the flow is parallel to the slopes, a fractured spatial distribution of TMCFs is observed. Comparisons of the MODIS-derived cloud-immersion map with the model output show that the MODIS product successfully captures the important cloud-immersion patterns in the Monteverde region of Costa Rica. The areal extent of cloud immersion is at a maximum during morning hours and at a minimum during the afternoon, before increasing again in the evening. Cloud-immersion frequencies generally increase with increasing elevation and tend to be higher on the Caribbean Sea side of the mountains. This study shows that the MODIS data may be used successfully to map the biogeography of cloud forests and to quantify cloud immersion over cloud-forest locations.

1. Introduction

Tropical montane cloud forests (TMCFs) are ecosystems characterized by frequent and prolonged immersion within orographic clouds. TMCFs and associated montane ecosystems often lie at the core of the biological hotspots—areas of high biodiversity—whose conservation is crucial to preservation of global biodiversity. Because of the small scales associated with these islands of endemism and their dependence on cloud water interception, TMCFs are extremely susceptible to environmental and climatic changes at regional or global scales (Pounds et al. 1999, 2006; Still et al. 1999; Lawton et al. 2001; Nair et al. 2003; van der Molen 2003; Ray et al. 2006). TMCFs also are important water resources, because the vegetation directly intercepts water from orographic clouds, accounting for up to 14%–18% and 15%–100% of total precipitation during dry and wet seasons, respectively (Bruijnzeel and Proctor 1993; Hager 2006).

In view of the ecological and hydrological importance of TMCFs, it is important to understand the biogeographical distribution of these ecosystems. There are currently three major data sources for the biogeographical distribution of TMCFs. The first data source is a compilation of approximate TMCF locations on regional maps identified by the participants of the Puerto Rico Tropical Cloud Forest Symposium in 1993 (Hamilton et al. 1993). Even though this is one of the first compilations of global TMCF distributions, Hamilton et al. (1993) note that it is woefully incomplete and contains possible inaccuracies. The second source is the global atlas of potential cloud-forest locations compiled by the United Nations Environmental Program (UNEP), which identifies similar locations on a digital elevation model (DEM). Although the UNEP approach is more detailed, it also may be inaccurate. The geographical locations of the TMCFs are valuable information for conservation efforts, but quantification of the proportion of time for which these locations experience cloud immersion is needed for ecological and hydrological studies. The third source (Mulligan and Burke 2005, hereinafter MB2005) utilizes annual mean cloud frequency generated from a long time series of National Oceanic and Atmospheric Administration High-Resolution Infrared Radiation Sounder (HIRS) cloud data, combined with geographical distributions of monthly averaged lifting condensation level (LCL) and digital elevation model to determine pantropical frequencies of ground-level cloud immersion (http://www.ambiotek.com/cloudforests). MB2005 successfully identify 11 out of 13 of Costa Rica’s cloud forests and 79% of all of the UNEP World Conservation Monitoring Centre (WCMC) known cloud-forest points. MB2005 are one of the first approaches to utilize satellite-derived cloud cover information, and it has been used to study the impact of climate and land use change on TMCFs and to set priorities for conservation. However, note that this product does not explicitly use estimates of cloud-base height, but utilizes LCL, a surrogate for cloud base, derived from surface climatological information. Relating LCL to orographic cloud-base height is not straightforward, and MB2005 utilize statistical techniques for this purpose.

The study presented here quantifies cloud immersion in montane regions by using a unique application of remotely sensed data to estimate cloud-base height. An alternate method is considered that utilizes a combination of cloud-base height explicitly determined using mesoscale numerical model and geostationary satellite data. Quantification of cloud immersion provides a means for identifying cloud-forest locations as well as a valuable metric for ecological and hydrological studies of TMCFs.

The study area and data used in this study are described in sections 2 and 3, respectively, and the method is provided in section 4. The results from this study are detailed in section 5, and section 6 contains conclusions.

2. Study area

The area of interest for this study includes Costa Rica, areas of southern Nicaragua, and the northern region of Panama (Fig. 1). The prominent terrain feature in this region is the Continental Divide, stretching northwest from Panama to Nicaragua, consisting of the Cordillera de Talamanca, Cordillera Central, Cordillera de Tilarán, and the Cordillera de Guanacaste, which is a chain of isolated volcanoes. The dominant northeast trade wind flow interacts with the cordilleras, creating orographic cloud banks that support cloud forests at sites along the crest and upper windward slopes of these ranges, including the well-known Monteverde cloud forest (Nadkarni and Wheelwright 2000).

Fig. 1.

A map of the study area with topography overlay. Brighter regions represent higher elevations. The large white rectangle outlines the domain of the coarser, outer grid used in the RAMS simulations discussed in the text. The smaller black rectangle outlines the domain for the finer-spaced, nested grid centered on the Monteverde region.

Fig. 1.

A map of the study area with topography overlay. Brighter regions represent higher elevations. The large white rectangle outlines the domain of the coarser, outer grid used in the RAMS simulations discussed in the text. The smaller black rectangle outlines the domain for the finer-spaced, nested grid centered on the Monteverde region.

3. Data

This study utilizes the Moderate-Resolution Imaging Spectroradiometer (MODIS) standard data products in conjunction with the National Centers for Environmental Prediction (NCEP) global tropospheric final analysis fields (FNL) to estimate cloud-top heights. MODIS standard data products are used to estimate cloud thicknesses, which in combination with cloud-top heights yield cloud-base heights. Geostationary Operational Environmental Satellite-8 (GOES-8) data, along with the Colorado State University Regional Atmospheric Modeling System (RAMS), are used to generate an independent dataset of cloud-immersion frequency (CIF). These are used to validate the MODIS-derived spatial distributions of cloud immersion in the study region. Along with cloud-base height estimates, the U.S. Geological Survey (USGS) global DEM is used to determine CIF statistics.

This study focuses on March, the peak of the dry season for the study area. Cloud immersion during the dry season is critical to the maintenance of cloud forests, because orographic cloud and mist are the dominant water sources at this time of year.

a. MODIS satellite data

MODIS is the primary imager on the Earth Observing System’s Terra and Aqua platforms. Each satellite is in a sun-synchronous orbit and views the surface of Earth every 1–2 days. This study utilizes daytime data from the MODIS sensor on the National Aeronautics and Space Administration (NASA) Earth Observing System’s Terra platform, which has a 1030 local time (LT) descending-node equator-crossing time. The MODIS sensor has 36 spectral bands, ranging from 0.4 to 14.4 μm. The first two visible bands are at 250-m spatial resolution, and bands 3–7 are at 500-m spatial resolution. The remaining bands (8–36) are at 1-km spatial resolution. This study utilizes the following MODIS standard products: cloud optical thickness, effective droplet radius, and cloud-top temperature at 5-km spatial resolution.

b. Global tropospheric final analysis fields

The NCEP global tropospheric FNL fields, containing gridded meteorological variables that include temperature and relative humidity at 26 pressure levels ranging from 1000 to 10 hPa, are used in conjunction with MODIS products to estimate cloud-top heights. The FNL fields are available at 1° × 1° resolution every 6 h on an operational basis and are archived at the National Center for Atmospheric Research.

c. GOES satellite data

The GOES-8 has five spectral channels, with one in the visible (0.52–0.72 μm) and four in the infrared region (3.78–4.03, 6.47–7.02, 10.2–11.2, and 11.5–12.5 μm). Spatial resolution at nadir for is 1 km for channel 1, 4 km for channels 2, 4, and 5, and 8 km for channel 3. Only the visible channel is used in this study. The automated cloud classification algorithm of Nair et al. (1999), which utilizes a time series of 1-km-spatial-resolution visible-channel imagery, is used to detect clouds in the GOES-8 imagery.

d. The RAMS model output

This study uses a series of RAMS simulations of orographic cloud formation over the study area. RAMS (Pielke et al. 1992; Cotton et al. 2003) is a nonhydrostatic numerical atmospheric modeling system used for simulating a wide range of atmospheric phenomena. This study uses the RAMS “current” land use simulation (Ray et al. 2006) for 1–14 March 2003 over northern Costa Rica. The Ray et al. (2006) simulations used a nested-grid configuration (Fig. 1) consisting of a coarse outer grid at 4-km grid spacing over a domain of 400 km × 160 km with a finer nested grid of 1-km spacing over a domain covering 62 km × 42 km, both of which are centered approximately on the Monteverde region (10.25°N, 84.7°W). Details of the RAMS configuration used for simulations are given in Ray et al. (2006).

The mean height of the first model level is at 9.5 m above the ground; therefore the simulations provide cloud water mixing ratio at the vegetation level. Ray et al. (2006) validated the RAMS simulations by comparison with both in situ radiosonde observations acquired during the Land Use Cloud Interaction Experiment (LUCIE) field campaign and also cloudiness observed in GOES imagery. RAMS simulations show good agreement with LUCIE-observed boundary layer thermodynamic profiles and GOES-observed orographic cloud banks (Ray et al. 2006). The cloud mixing ratio field from the inner grid (Fig. 1) is used in this study.

e. The USGS global DEM

The USGS global DEM dataset (http://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html) provides elevation data on a 30 arc s (approximately 1 km) geographic projection grid.

f. Photographic observations of orographic cloud-base height

Photographs taken daily at the Monteverde Cloud Forest Reserve (Fig. 2) are used to estimate orographic cloud-base heights at two times per day at about 0600 and 1200 LT March 2003. A set of visually prominent surface features on the mountain slopes, whose altitudes are known from the 20-m-contour topographic maps of the Instituto Geografico de Costa Rica and from global positioning system readings, are used to estimate the orographic cloud-base heights manually from the photographs. The accuracy of cloud-base heights derived from photographic observations is estimated to be about 50–100 m (Welch et al. 2008, hereinafter Part I). Figure 2 shows representative photographs for documenting cloud-base heights on days of differing orographic cloud bank development along the Continental Divide in the central Cordillera de Tilarán, northern Costa Rica. These photographs were taken at midday at an elevation of 1540 m, near the headquarters of the Monteverde Cloud Forest Reserve. The view is southeast to the summit of Cerro Ojo de Agua (1800 m). The Continental Divide runs across the summit and through the Brillante saddle (1460–1510 m) in the middle ground, and then behind the ridge in the left foreground. The ravine descending away in the right foreground allows terrain features as low as 1250 m on the upper Pacific (lee) slope of the Continental Divide to be distinguished under most cloud conditions. In the leftmost photograph (25 March 2003), orographic cloud development is weak and patchy, and the cloud base is at ∼2100 m MSL, ∼300 m above the peak of Cerro Ojo de Agua. In the center photograph (5 March 2003), orographic clouds are better developed and form a continuous bank with its base at ∼1675 m MSL. In the photograph on the right (9 March 2003), the orographic cloud bank is well developed, enveloping Cerro Ojo de Agua and the Brillante saddle above ∼1400 m MSL and extending over the whole upper Caribbean slope of the cordillera.

Fig. 2.

Examples of photographic observations, from the Monteverde region in Costa Rica, used in to estimate orographic cloud-base height.

Fig. 2.

Examples of photographic observations, from the Monteverde region in Costa Rica, used in to estimate orographic cloud-base height.

Note that the MODIS overpass time (∼1030 LT) occurs approximately 90 min ahead of the 1200 LT photographic observations. This induces an additional error when comparing MODIS-derived cloud-base heights with photographic observations. However, prior numerical modeling studies (Ray et al. 2006) and visual observations [one of us (ROL) has conducted field studies in the study area over a period of more than 20 years] suggest that the orographic cloud-base heights do not often change rapidly during this time of the day. Therefore, errors resulting from mismatched observation times are expected to be not greater than 100 m, as determined by Part I.

g. UNEP WCMC database

The UNEP WCMC maintains a list of potential cloud-forest locations (http://www.unep-wcmc.org/forest/cloudforest/americasc.cfm) for the Americas. These maps shows distributions of forest cover in mountains within defined altitudinal ranges that are likely to include cloud forests. However, they also include montane rain forests and drier mountain forests, and therefore are a potential distribution of cloud forest that overestimates its occurrence. The UNEP WCMC Internet site also points out that cloud-base heights may be different on the wet and dry sides of mountains and that cloud forests tend to occur higher on higher mountains. They utilize a geographical information system base layer of mountain heights and overlay that with global forest coverage obtained from national sources up to 1997 and from MODIS data up to 2000. They “exclude the areas of mountain forest outside the altitudinal ranges for the likely occurrence of cloud forests,” which for Mesoamerica and South America are in the range of 2000–3500 m.

Figure 3 shows these locations for the Costa Rica study region. Comparison of Figs. 1 and 3 shows that the UNEP WCMC map identifies potential cloud-forest sites as a continuous swath along the Cordillera de Talamanca, a few isolated locations in the Cordillera de Tilarán, and Cordillera Central, and no locations in the Cordillera de Guanacaste.

Fig. 3.

UNEP WCMC map of potential cloud-forest sites in the Costa Rica study area (Bubb et al. 2004).

Fig. 3.

UNEP WCMC map of potential cloud-forest sites in the Costa Rica study area (Bubb et al. 2004).

4. Methods

Two methods for quantifying cloud immersion are considered—one that uses the combination of the MODIS cloud mask and the orographic cloud-base height estimates derived from MODIS standard products and a second that uses the combination of a GOES cloud mask and cloud-base height estimates derived from the high-spatial-resolution RAMS mesoscale model.

a. Determination of cloud immersion using MODIS data

This study uses the cloud-base height estimation algorithm that is described in Part I. In brief, detection of cloud immersion at a particular location using MODIS data involves the following steps: 1) the MODIS cloud mask over the study area is examined to identify orographic clouds (if a cloud is present, then cloud-top height, cloud thickness, and cloud-base height are determined), 2) the surface elevation at each MODIS pixel location is determined using the USGS DEM, and 3) if the surface elevation is greater than or equal to the cloud-base height estimate then this location is flagged on the DEM as experiencing cloud immersion. The above-described process is applied to all cloudy pixels observed in a series of MODIS scenes over the study area that also satisfy the quality-control criteria described in section 4a(3).

The total number of cases within a given time period is the sum of the number of clear scenes nc, those contaminated by upper-level cloudiness nu, cloud-immersed scenes ni, and cloudy but nonimmersed scenes nni. CIF at a particular location is determined using the following relationship:

 
formula

where ns is the total number of uncontaminated cloud cases, which is the sum of ni and nni. Note that the calculation of CIF is not performed if cloud layers thicker than ∼2 km or cirrus clouds are present.

At a particular location, let nt be the total number of uncontaminated observations that are used in the computation of CIF [nt = nc + ns, i.e., the denominator in Eq. (1)], let pc be the fraction of total observations in which clear conditions are observed, let pi be the fraction of cloudy cases for which cloud immersion occurs at the location, and let ps be the fraction of cloudy cases for which the MODIS cloud-base retrieval algorithm is successful (includes immersed and nonimmersed cases). Then, nc = pcnt, ncld = (1 − pc)nt, ns = psncld = ps(1 − pc)nt, and ni = pips(1 − pc)nt, where nc is the total number of contaminated and uncontaminated cloudy cases. From Eq. (1), the CIF values determined from MODIS data, the actual value of CIF, and the ratio of MODIS-derived CIF at a given location are given by

 
formula
 
formula
 
formula

Because ps < 1, CIFMODIS < CIFACTUAL. Note that as ps increases, the MODIS estimate of CIF improves.

1) Terrain effects

Orientation of the terrain also plays a role in the spatial distribution of CIF, with cloud immersion being more prominent along regions in which the major axis of the terrain feature is approximately at right angles to the northeasterly direction of trade wind flow. To examine the relationship between the spatial distribution of topography and CIF, the dot products between unit vectors parallel to the gradient of surface elevation and the prevailing wind direction are computed. Unit vectors h, parallel to gradient of surface elevation H(x, y), and v, parallel to the prevailing northeasterly wind direction, along with the dot product of h and v, are given by

 
formula
 
formula
 
formula

The value of the dot product varies from −1.0 to 1.0, and a value of 1 indicates that the gradient of topography is parallel to the prevailing wind direction, meaning that the prevailing winds are directly impinging on the topography. A value of 0 for this dot product implies that the prevailing wind direction is perpendicular to the gradient of topography. In this case the wind vector is parallel to the elevation contours, and thus the topography is essentially channeling the airflow rather than lifting the air to higher elevations. When the value of the dot product is −1, the wind vector is directed away from the gradient of topography. In this case the airflow is directed down the slope.

2) Determination of cloud-base height from MODIS data

Cloud-base heights are estimated using the following MODIS cloud products: cloud mask, cloud-top temperature, cloud optical thickness, and cloud effective droplet radii. Details are given in Part I. Briefly stated, cloud-top height is estimated as the geopotential height at which the environmental dewpoint temperature is equal to the cloud-top temperature. Cloud-top temperature is determined from the 11-μm (channel 31) MODIS data. Because of saturated conditions that exist within a cloud, the dewpoint temperature is equal to the environmental temperature within the cloud and at cloud top.

Cloud-base height is computed by subtracting cloud thickness from cloud-top height. Part I examined three potential approaches for estimating cloud thickness: 1) the constant liquid water content (CLWC) approach (Hutchison 2002), 2) an empirical relationship (ER), which assumes an empirically derived relationship between cloud optical depth and geometrical thickness (Minnis et al. 1997), and 3) the adiabatic model (AM), based on the assumption of adiabatic evolution in a closed convective cloud system (Brenguier et al. 2000). Comparisons of the three approaches for estimating cloud-base height with photographic observations demonstrate that the ER and AM approaches produce superior performance. Root-mean-square errors associated with the ER, AM, and CLWC cloud-base retrievals for orographic clouds at Monteverde, Costa Rica, in March of 2003 are 54, 61, and 173 m, respectively (Part I). The AM method is chosen to compute cloud-base heights in this investigation.

3) Minimizing beam filling and edge effects

Erroneous retrievals of cloud-top temperature, cloud optical depth, and cloud particle size are probable along the edges of the orographic clouds. Along cloud edges, three-dimensional radiative transfer effects are significant because of the radiative flux entering and exiting the cloud sides, thereby invalidating the plane-parallel assumptions implicit in the retrievals. Mixing with environmental air also is enhanced along the cloud edges, leading to anomalous values of LWC relative to inner areas of the cloud decks. In this study, edge pixels are eliminated from computations of cloud-base height.

b. Determination of cloud immersion using GOES data and RAMS simulations

A blended CIF map is created for the Monteverde region based upon the assumption that, at a particular location, the RAMS-predicted cloud-base height is correct if the model forms orographic cloud and if cloud is detected at this location in the GOES imagery. For the simulations discussed in section 3d, Ray et al. (2006) conducted a detailed analysis of RAMS performance in simulating cloudiness by using a point-by-point comparison of the cloud field simulated within the 1-km grid with a 1-km-resolution cloud mask derived from GOES imagery. This analysis found that incorrect simulations of cloudiness (on average 20% of the time) occur mainly in the Caribbean lowlands and that RAMS demonstrates skill at simulating orographic cloud decks in montane and premontane regions.

The procedure for creating the blended CIF spatial distribution maps involves the following steps: 1) If the RAMS simulations predict the presence of orographic cloud, then the GOES cloud mask is examined to determine whether the model and observations agree. 2) If they agree, then it is determined whether the RAMS simulations produce cloud mixing ratio values >0.04 g kg−1 [this threshold is discussed in section 5c(2)] at the first model level above the surface. 3) If these values exist, then they predict the presence of cloud at ground level, thus indicating cloud immersion at this location. 4) Steps 1–3 are repeated, and the total number of incidences of cloud immersion observed in the model simulations for the time period considered is computed. 5) CIF statistics are computed as the percentage of days for which incidence of cloud immersion is observed at given location. and 6) Steps 1–5 are repeated for all gridpoint locations in the model domain.

Because GOES imagery and numerical model outputs are available several times per day, it is possible to examine the diurnal variation of CIF values using the blended products. For the 1–14 March 2003 period, CIF blended maps are generated using the method described above at 2-h intervals starting at 1215 UTC and ending at 2215 UTC. In the following sections, the method described in this section is referred to as the GOES–RAMS explicit cloud microphysics (GRECM). Note that because GRECM-derived CIF patterns are compared with MODIS-derived CIF patterns, only days for which MODIS overpasses are available over the study area are used in this analysis.

5. Results

a. Cloud-immersion frequency derived from MODIS data

Within the study area, a number of cloud-forest locations are listed in the UNEP WCMC database. Figure 4a shows 13 known cloud-forest sites along the cordilleras from northern Panama to northwestern Costa Rica, denoted with the red asterisk symbol overlaid on topography (the brighter the area is, the higher is the topography). Figure 4b shows the CIF spatial distribution for the study area, derived using the technique described in section 4a. These data are averages for March over 4 yr, 2003–06. The regions labeled “R3” in Fig. 4b are isolated pockets of cloud immersion in the southwestern regions of Nicaragua and northwestern regions of Costa Rica that are associated with a chain of isolated volcanoes in the Cordillera de Guanacaste. Note that these pockets are not included in the UNEP WCMC database. A near-continuous chain of montane regions frequently immersed in clouds is seen along the upper Caribbean slope and the crest of Cordillera de Tilarán and Cordillera Central, labeled as “R4” in Fig. 4b, and a more fractured spatial distribution of CIF values is observed over the Caribbean slopes of Cordillera de Talamanca, labeled as “R5.” There are several small, isolated areas immersed in clouds and a relatively large area of frequent cloud immersion on the Pacific side of the Cordillera de Talamanca, labeled “R6,” extending from Costa Rica to northern Panama. The spatial distribution of CIF shows that prominent areas of cloud immersion tend to be located along the upper Caribbean slopes of the continental divide, where orographic cloud formation is driven by the trade wind regime. The values of CIF range from 10% at lower elevations to >50% at higher elevations. Comparison of Figs. 4a and 4b shows that the locations of the known cloud-forest sites, obtained from the UNEP WCMC atlas, are included in the MODIS-derived CIF distribution.

Fig. 4.

(a) Known cloud-forest locations (UNEP WCMC compilation) for the study area, denoted by red asterisks, overlaid on topography for the study area (brighter regions represent greater elevation); (b) cloud-immersion frequency derived from MODIS for March 2003–06 overlaid on topography; (c) dot product of unit vectors parallel to gradient of topography and the prevailing wind direction (northeasterly); (d) as in (b), except that CIF is overlaid over the dot-product field given in (c) instead of topography. The three color/shading bars (from top to bottom) correspond to cloud immersion, topography, and dot-product fields depicted in the different panels; R1–R10 mark features discussed in the text.

Fig. 4.

(a) Known cloud-forest locations (UNEP WCMC compilation) for the study area, denoted by red asterisks, overlaid on topography for the study area (brighter regions represent greater elevation); (b) cloud-immersion frequency derived from MODIS for March 2003–06 overlaid on topography; (c) dot product of unit vectors parallel to gradient of topography and the prevailing wind direction (northeasterly); (d) as in (b), except that CIF is overlaid over the dot-product field given in (c) instead of topography. The three color/shading bars (from top to bottom) correspond to cloud immersion, topography, and dot-product fields depicted in the different panels; R1–R10 mark features discussed in the text.

Figure 4c shows the spatial distribution of dot product given by Eq. (7) over the study area. White areas in Fig. 4c indicate a dot product of +1, and black areas indicate a value of −1. Notice that the gradient of Caribbean slopes along the Cordillera de Guanacaste, Cordillera de Tilarán, and Cordillera Central are approximately parallel to the prevailing wind direction, indicated by “R7” in Fig. 4c. The spatial distribution of dot product shows a continuous stretch of Caribbean slopes along the Cordillera Central and Cordillera de Tilarán upon which the prevailing winds directly impinge. Along this region, the transition from the east-facing slopes on the Caribbean side to the west-facing slopes on the Pacific side is distinct and abrupt (R8). Also note that the slopes along the Caribbean side tend to be more gradual than those on the Pacific side. Farther south along Cordillera de Talamanca, the dot-product field does not show a continuous swath of slopes whose gradient is parallel to the prevailing wind vector (R9). This suggests that the topography along the Caribbean slopes of the Cordillera de Talamanca tends to channel the airflow rather than force the flow to higher elevations. This is in part due to the fact that the Cordillera de Talamanca rises to 3000–4000 m, extending well above the trade wind inversion. Therefore, it is more effective as a diversionary “dam” across the trade wind flow. Of interest is that, on the upper Pacific slope of the Cordillera de Talamanca, in the lee of the Continental Divide, there is a region of high CIF (R10), perhaps due to an upslope eddy behind the dam.

This analysis of prevailing wind direction and topography provides an explanation for the locations of prominent montane cloud forests. Indeed, the pattern of cloud immersion is related to the orientation of slope along the cordilleras and the prevailing wind direction, as shown in Fig. 4d. Note the differences in CIF values between the northern and southern cordilleras. The slopes of the northern cordilleras are oriented such that the trade wind flow directly impinges on the slopes, creating a continuous swath of cloud immersion along the Cordillera Central and Cordillera de Tilarán (Fig. 4c, R8; Fig. 4d). Along the southern cordilleras, very large ridges extend toward the Caribbean coast, providing slopes that tend to channel the airflow, creating isolated spots of cloud immersion at the end of valleys (Fig. 4b), rather than forcing the flow to ascend. Because of this difference in terrain, regions of cloud immersion are more fractured along the Caribbean slope of the Cordillera de Talamanca than on the Cordillera Central, Cordillera de Tilarán, and Cordillera de Guanacaste, where there are continuous regions of high cloud immersion. Of special note is the fact that spatial distributions of cloud-forest locations derived from MODIS data differ significantly from the UNEP WCMC classification (Fig. 3). The UNEP WCMC classification identifies a continuous swath of potential cloud forest on the Caribbean slope of the Cordillera de Talamanca, and it suggests that there are only a series of isolated cloud-forest areas in the Cordillera de Tilarán. Furthermore, the UNEP WCMC approach does not identify any cloud-forest locations associated with Cordillera de Guanacaste in the northwestern region of Costa Rica. One reason for the differences between UNEP WCMC potential cloud-forest distribution and the MODIS-derived CIF regions is that the UNEP WCMC scheme identifies potential cloud-forest locations by examining the terrain associated with known cloud-forest sites and then identifying other locations in the vicinity with similar elevation ranges on a DEM. This UNEP WCMC approach clearly makes a significant advance over previous efforts to map cloud forests, but the analysis presented in this paper suggests that it makes both false positives and omissions. Perhaps more important, the UNEP WCMC results do not explicitly consider cloud immersion, a defining characteristic of TMCFs, whereas the technique used in this study explicitly considers cloudiness at vegetation level, as discussed next.

Comparison with the cloud-forest distribution of MB2005 (not shown) reveals several similarities but also some differences. Unlike the UNEP WCMC distribution, MB2005 identify most of the isolated cloud-forest locations in the Cordillera de Guanacaste. However, as compared with the MODIS-derived product, they do not identify isolated areas of cloud immersion on the volcanoes in Lake Nicaragua, located in the southwest Nicaraguan region. MB2005 also identify a continuous region of cloud immersion in the Caribbean slopes of Cordillera de Talamanca as compared with a more fractured distribution found in the MODIS-derived product.

b. Cloud-immersion frequency using GRECM

Figure 5a shows the spatial distribution of GRECM-derived CIF values at 1615 UTC 1–14 March 2003, over the Monteverde region of Costa Rica. Note the nearly continuous region of frequent cloud immersion along the crest of the Cordillera de Tilarán. Along the Caribbean slopes, GRECM-derived CIF fields show cloud-immersion frequency ranging from less than 10% at about 1200 m to more than 80% along the crest of the cordillera. However, on the Pacific slopes cloud immersion abruptly drops from 50% to near zero at about 1400 m (Fig. 5a). Figure 5a also shows that within this continuous region of cloud immersion three prominent, local maxima of CIF are present.

Fig. 5.

Cloud-immersion frequency over Monteverde derived using GRECM, valid at (a) 1615, (b) 2015, and (c) 2215 UTC and the (d) 1215–2215 UTC average. CIF and topography are shown using color and gray shades, respectively.

Fig. 5.

Cloud-immersion frequency over Monteverde derived using GRECM, valid at (a) 1615, (b) 2015, and (c) 2215 UTC and the (d) 1215–2215 UTC average. CIF and topography are shown using color and gray shades, respectively.

One of the advantages of the GRECM technique is that, unlike the MODIS-based method, it is possible to examine the diurnal variation of CIF values. The GRECM-derived CIF distribution at 1615 UTC (Fig. 5a) has the same general spatial pattern as the GRECM CIF distribution that is observed for early morning hours (not shown). Comparison of Figs. 5a and 5b shows that the core regions are rapidly and significantly reduced in extent by 2015 UTC, the time of minimum CIF values. Then, during the next 2 h, the CIF spatial distribution pattern begins to expand, and by 2215 UTC (Fig. 5c) the CIF spatial distribution pattern is returning to that seen in the morning hours (Fig. 5a).

The diurnally averaged spatial distribution of GRECM CIF shown in Fig. 5d is dominated by the behavior of the orographic cloud deck during the morning and evening hours. This diurnal behavior is modulated by land surface processes, with the warming of the air in the lowlands through the course of the day leading to higher orographic cloud-base heights and reduction in the area immersed in cloud. (Ray et al. 2006). During the late-afternoon and evening hours, the air over the lowland region cools, leading to a reduction in cloud-base heights and an increase in areal extent of cloud immersion.

c. Comparison of MODIS and GRECM retrievals

Figure 6a shows the MODIS-derived CIF spatial distributions near Monteverde that correspond to GRECM-derived CIF spatial distributions shown in Fig. 5a. There are clearly similarities and differences between these two estimates of cloud-immersion frequency. Both show similar orientations and shapes of the CIF features, with near-continuous cloud immersion along the continental divide, with three distinct local maxima.

Fig. 6.

Cloud-immersion frequency over Monteverde derived using MODIS data for (a) March 2003 and (b) March 2003–06. Note that the color shades show cloud-immersion frequency and gray shades depict topography, with the brighter values indicating greater elevations; X1 and X2 indicate the locations of the cross sections shown in Fig. 8. The red rectangle in (a) shows the location of the averaging window used for MODIS and RAMS CIF shown in Fig. 7.

Fig. 6.

Cloud-immersion frequency over Monteverde derived using MODIS data for (a) March 2003 and (b) March 2003–06. Note that the color shades show cloud-immersion frequency and gray shades depict topography, with the brighter values indicating greater elevations; X1 and X2 indicate the locations of the cross sections shown in Fig. 8. The red rectangle in (a) shows the location of the averaging window used for MODIS and RAMS CIF shown in Fig. 7.

Most notable in the comparison between Figs. 5a and 6a are the differences in magnitude of CIF values, with local maxima ranging from about 20% to 90% in the MODIS-derived results and from about 60% to 85% in the GRECM results. Also, there are differences in the areal extents of the nonzero CIF regions, with the GRECM regions being considerably larger. There are two reasons for the differences in magnitudes of CIF between the GRECM- and MODIS-derived CIF fields. Limitations in the MODIS retrievals lead to underestimations of cloud-immersion frequency, especially at lower elevations, whereas errors in the model tend to create overestimates of CIF. These limitations, analyzed by comparisons with photographic observations, are detailed in sections 5c(1) and 5c(2). However, note that this analysis is subject to errors associated with difference in timing between MODIS overpass and photographic observations, but such errors are expected to be minimal as noted in section 3e.

1) Limitations of the MODIS retrievals

The MODIS cloud algorithm is not always successful at retrieving cloud properties, and even when it is successful the values of cloud particle size and optical depth may not fall within the range in which cloud-base height can be accurately estimated (Part I). The MODIS cloud-base height retrieval algorithm also fails when cirrus or other upper-level clouds are present. It is not possible to compute cloud immersion at locations in which either of these conditions occurs.

Given the assumptions that the percentage of clear cases is relatively constant and that the fraction pc of total observations in which clear conditions are observed does not vary drastically from year to year, Eq. (4) shows that an increase in the fraction of cloudy cases ps for which the MODIS cloud-base height retrieval is successful leads to more accurate estimates of MODIS-derived CIF values. Because ps < 1, the CIF values derived from MODIS always will be underestimated. However, note that simply having more cases does not necessarily produce higher accuracies; rather, higher accuracy is determined by larger numbers of cases in which cloud-base height can be determined. This result implies that regions of small cloud cover tend to have lower accuracies, because in those cases there are a larger number of cloud edge pixels, for which the retrieval scheme cannot be applied. Likewise, the value of ps usually is smaller at lower elevations, because the edges of orographic cloud banks generally are located in these areas, especially on the Caribbean side. Indeed, the largest discrepancies between Figs. 5a and 6a are in areas dominated by edges of the orographic cloud banks.

One strategy to address this problem is to increase the observation period, thereby increasing the probability of successful retrievals, especially along the areas dominated by cloud edges. To test this hypothesis, results were generated for the time period of March of 2004, 2005, and 2006 and are shown in Fig. 6b. Comparison of Figs. 6a and 6b shows that the areal extent of the MODIS-derived CIF spatial distributions expands greatly and compares somewhat better to the GRECM-derived CIF patterns (Fig. 5a). As expected, CIF patterns within the core regions (local maxima) remain relatively unchanged but important changes are observed along the edges of these regions.

Figure 7a shows cloud-immersion frequencies derived from MODIS, GRECM, and photographs averaged over the 1–14 March 2003 time period. The MODIS and GRECM CIF values were averaged over a 7 km × 8 km area centered on the site of photographic observations, shown by the box in Fig. 6a. The MODIS and GRECM CIF values have lowest cloud-base heights of 1150 and 1250 m, respectively, and both show CIF values increasing with elevation, reaching values of 61% and 50%, respectively, at elevations of 1450 and 1650 m. Note that the decline in RAMS-derived CIF values between 1350 and 1550 m is an artifact of averaging, resulting from the terrain and the placement of the window over which spatial averaging occurs. A larger proportion of upper-elevation terrain in the window lies commonly beyond the lee edges of the orographic cloud bank, and therefore when the orographic cloud-base height is in this range the proportion of area in this elevation band immersed in cloud necessarily declines. However, photograph-based CIF values have a plateau region followed by a steady increase after 1450 m, reaching a maximum value of 100% at elevations of 1850 m, which is approximately the elevation of the highest peak in the area.

Fig. 7.

(a) Average cloud-immersion frequency as a function of elevation for the time period 1–14 Mar 2003. Triangle, diamond, and circle markers show average CIF derived from MODIS, GRECM, and photographs, respectively. (b) Average cloud-immersion frequency for March 2003 and 2004. Triangle and square markers show MODIS-derived average CIF for 2003 and 2004, respectively, and circle and cross markers show CIF values derived from photographs for 2003 and 2004. An elevation bin size of 100 m is used for averaging CIF values.

Fig. 7.

(a) Average cloud-immersion frequency as a function of elevation for the time period 1–14 Mar 2003. Triangle, diamond, and circle markers show average CIF derived from MODIS, GRECM, and photographs, respectively. (b) Average cloud-immersion frequency for March 2003 and 2004. Triangle and square markers show MODIS-derived average CIF for 2003 and 2004, respectively, and circle and cross markers show CIF values derived from photographs for 2003 and 2004. An elevation bin size of 100 m is used for averaging CIF values.

Photographic observations at the Continental Divide on the central Cordillera de Tilarán show significant interannual variations in altitudinal distributions of CIF (Fig. 7b). For March of 2003 and 2004, monthly averaged values of CIF derived from photographs show a steady increase in CIF from 13% to 53% and from 58% to 83%, respectively, for a change in elevation from 1250 to 1650 m. For the corresponding range of altitudinal change, MODIS-derived values of CIF vary from 6% to 36% and from 8% to 33% for March 2003 and 2004, respectively. Because the photographic observations do not show a significant change in the percentage of cloudy days from March of 2003 to March of 2004, the increase in CIF values at lower elevations indicate a downward movement of the orographic cloud banks in March of 2004. Soil moisture observations from northern Costa Rica show soil moisture values in regions directly upwind of observational area were unusually high (30% volumetric soil moisture; near field capacity) in March of 2004 relative to the more normal values of March of 2003 (18% volumetric soil moisture). Indeed, modeling studies of Lawton et al. (2001), Nair et al. (2003), and Ray et al. (2006) showed that high soil moistures in the Caribbean lowlands leads to lower cloud-base heights. The results presented here are an observational confirmation of those previous results.

2) Limitations of the GRECM approach

Comparison of averaged cloud-immersion frequencies derived from GRECM and photographs (Fig. 7a) for 1–14 March 2003 shows that GRECM overestimates CIF values at elevations below 1350 m and underestimates CIF values at elevations greater than 1450 m. The GRECM-derived CIF values show an abrupt increase from 0% to ∼27% at elevations of 1150 m (0% for photograph based) and a further increase to 38% by 1350 m (28% for photograph based), followed by a decrease, (reasons discussed above) to 33% by 1550 m (71% for photograph based), and then an increase to 50% by 1650 m (85% for photograph based). Figure 7 shows that domain-averaged CIF values derived using GRECM have a pattern of variation very similar to that found for MODIS-derived and photographic observations. However, note that the lower limit of photographic observations is at 1200 m. Also, photographic observations are point observations while the GRECM behavior shown in Fig. 7a is an average over a 7 km × 8 km area centered over the area of photographic observations, shown by the box in Fig. 6a.

Several sources of error associated with the GRECM technique may lead to overestimates of CIF values. Errors in simulated orographic cloud-base height may result from improper specification of the spatial distribution of soil moisture (Nair et al. 2003; Ray et al. 2006). Another source of error in the GRECM technique results from the use of RAMS-predicted nonzero liquid water contents at the first model level as an indicator of cloud immersion (if cloud is detected in the GOES imagery at the same location). The technique does not currently discriminate whether this cloud liquid water is a part of a thicker cloud layer or is just light fog that sometimes persists near the surface during morning hours. To minimize this problem, a threshold of 0.04 g kg−1 cloud water mixing ratio is used, based on cloud liquid water content measurements obtained from Monteverde (Part I). This value is the mean of the lower 10% of LWC observations for March, assumed to be representative of light-fog events. However, application of this arbitrary threshold does not completely eliminate this problem, and model-generated light surface fog may be responsible for nonzero CIF in a number of valleys, such as that of Rio Peñas Blancas, and also in the lowland areas.

Comparisons between GRECM-, MODIS-, and photographic observation–derived values of CIF show that 1) the spatial patterns of MODIS-derived CIF patterns are robust while the absolute magnitudes of MODIS-derived CIF values are an underestimate, 2) GRECM-derived values of CIF compare reasonably well to observations, although the CIF values derived using this technique are overestimated at lower elevations and underestimated at higher elevations, and 3) photographic observations show considerable interannual variations in CIF, but MODIS-derived CIF values do not always capture the interannual variations at all elevations.

d. Cross-sectional view of MODIS-derived CIF

A west-to-east cross section of MODIS-derived CIF values along the transect X1 indicated in Fig. 6a is shown in Fig. 8a. A similar cross section along a ridge is indicated by transect X2 in Fig. 6a and is shown in Fig. 8b. Note that maximum CIF values occur on the upper Caribbean slope near the crest of the cordillera. Figures 8a,b show that cloud immersion starts with CIF values of about 4% at elevations of ∼1100 m on the Caribbean slopes and then rapidly increases to about 47%–50% at elevations of 1500–1650 m. The CIF values decrease along the descending Pacific slopes, reaching a value of about 4% near 1300 m. As expected, orographic cloud-base heights are about 200 m higher on the Pacific side. The cross sections shown in Fig. 8 also show good correlation between CIF values and topography in the core regions of cloud immersion, with local CIF minima and maxima being well correlated with topography.

Fig. 8.

MODIS-derived cloud-immersion frequency (a) along west-to-east cross section X1 and (b) along northeast-to-southwest cross section X2 shown in Fig. 5b. The thick black line shows topography; the thin line with diamonds shows variation of cloud-immersion frequency along the cross section.

Fig. 8.

MODIS-derived cloud-immersion frequency (a) along west-to-east cross section X1 and (b) along northeast-to-southwest cross section X2 shown in Fig. 5b. The thick black line shows topography; the thin line with diamonds shows variation of cloud-immersion frequency along the cross section.

6. Conclusions

Tropical montane cloud forests rely on frequent, persistent cloud cover at the vegetation level. Mapping cloud-forest locations is important for conservation purposes, and characterizing regional microclimates is important for understanding both ecological and hydrological processes. At present there are three sources that compile biogeographical distributions of cloud forests, of which two do not explicitly consider cloud immersion, the most defining characteristic of TMCFs. The third uses cloud immersion but uses LCL as a surrogate of cloud-base height to determine cloud immersion.

The study presented here examines the use of satellite remote sensing and numerical modeling to determine cloud-immersion statistics, which can be used both for identifying cloud-forest locations and also for characterizing the most important microclimatic variable relevant to TMCFs—namely, the proportion of time a given location experiences cloud immersion. A unique aspect of this study is the use of explicit measures of cloud-base height. Two techniques are considered in this study to derive the spatial distribution of CIF values over the study region that includes Costa Rica, southern Nicaragua, and northern Panama: 1) CIF values determined using cloud-base heights derived from MODIS satellite data using the algorithm of Part I and 2) CIF values determined from cloud-base height estimated from RAMS-simulated cloud fields and GOES satellite imagery. This study compares these techniques with photographic observations of orographic cloud-base heights. The spatial distributions of CIF from these techniques are also compared with the known cloud-forest locations in the UNEP WCMC database and also the potential cloud-forest locations in the same database. The major findings from this study are as follows:

  1. Spatial patterns of cloud immersion over cloud-forest areas may be successfully derived using MODIS satellite data. The MODIS-derived CIF spatial patterns correlate well with GRECM-derived CIF patterns and the locations of known cloud-forest sites in the region and exhibit superior skill in comparison with the UNEP WCMC algorithm.

  2. The MODIS-derived CIF pattern is similar to that of MB2005 in the northern Costa Rican region. However, the MODIS-derived CIF pattern is more fractured in the southern Costa Rican region and also identifies isolated regions of cloud immersion in the southern Nicaraguan area.

  3. Even though MODIS is successful at identifying the spatial patterns of CIF, it underestimates the magnitude of CIF values. The reasons for this behavior are that algorithms used for retrieving cloud-base height are not always successful, and thus MODIS-derived CIF values do not account for all cases of cloud immersion. This problem is most severe along the cloud edges, and thus the CIF analysis needs to use multiple years of data to obtain reliable CIF patterns, especially at lower elevations.

  4. Photographic observations show considerable interannual variations in CIF, but the MODIS technique may not be reliable for capturing such variations at all sites.

  5. The major features observed in the MODIS-derived CIF distributions over the study area for March may be explained on the basis of interactions between the prevailing trade wind regime and topography. Large connected regions of cloud immersion are observed in regions in which the trade wind flow is directly impinging upon the mountain slopes, whereas in areas in which the flow is parallel to the slopes a fractured spatial distribution of CIF values is observed.

  6. The CIF values derived using the GRECM technique show good agreement with observations. Photographic observations suggest that GRECM-derived CIF values are an overestimate at lower elevations and an underestimate at higher elevations.

  7. The GRECM technique allows diurnal variations of CIF values to be determined. Over the Monteverde region, GRECM-derived CIF values show a pattern in which the areal extent of cloud immersion is higher during the early-morning hours, decreases in the afternoon, and then increases again in the evening.

Both of the techniques considered in this study appear to be promising for biogeographical mapping of TMCFs. The first technique (MODIS-derived CIF) is computationally less expensive and can be applied globally. However, the MODIS technique provides CIF maps for only two times of the day, and multiple years of data must be used to create robust statistics. MODIS-derived CIF values should be applied for studying the average patterns of CIF distribution. The MODIS approach has limited ability to resolve diurnal variations, and it may not be capable of capturing some of the patterns of interannual variability in CIF values. Although the GRECM method is capable of capturing spatial and diurnal patterns of cloud immersion, numerical modeling utilizing explicit cloud microphysics is computationally expensive and currently cannot be applied except in regional settings. When using the GRECM technique, it is difficult to distinguish between ground-fog events and orographic cloudiness.

The MODIS-derived CIF products, including seasonal variations of CIF, are currently being developed for the Central American region by the Regional Visualization and Monitoring System (SERVIR)1 for utilization within Mesoamerican environmental ministries. It is expected that these CIF products will be used for improved mapping and monitoring of TMCFs throughout the region.

Acknowledgments

This research was supported by NASA Grants NNG04GH51G, NNX06AB68G, and NNM05AA22A. CIF products for 2003–06 were developed for SERVIR by Jason A. Arnold, Universities Space Research Association (USRA), at the National Space Science and Technology Center in Huntsville, Alabama. We thank UNEP WCMC for providing the global cloud-forest point dataset, which is available from UNEP WCMC, Cambridge, United Kingdom. We are thankful for the helpful comments and suggestions of the reviewers.

REFERENCES

REFERENCES
Brenguier
,
J-L.
,
H.
Pawlowska
,
L.
Schüller
,
R.
Preusker
,
J.
Fischer
, and
Y.
Fouquart
,
2000
:
Radiative properties of boundary layer clouds: Droplet effective radius versus number concentration.
J. Atmos. Sci.
,
57
,
803
821
.
Bruijnzeel
,
L. A.
, and
J.
Proctor
,
1993
:
Hydrology and biogeochemistry of tropical montane cloud forests: What do we really know?
Tropical Montane Cloud Forests, L. S. Hamilton, J. O. Juvik, and F. N. Scatena, Eds., East–West Center, 21–46
.
Bubb
,
P.
,
I.
May
,
L.
Miles
, and
J.
Sayer
,
2004
:
Cloud forest agenda. UNEP-WCMC Biodiversity Series 20, 36 pp
.
Cotton
,
W. R.
, and
Coauthors
,
2003
:
RAMS 2001: Current status and future directions.
Meteor. Atmos. Phys.
,
82
,
5
29
.
Hager
,
A.
,
2006
:
Einfluss von Klima und Topographie auf Struktur, Zusammensetzung und Dynamik eines tropischen Wolkenwaldes in Monteverde, Costa Rica (Influence of climate and topography on structure, composition, and dynamics of a tropical cloud forest in Monteverde, Costa Rica). Ph.D. thesis, Georg August Universitat Gottingen, 275 pp
.
Hamilton
,
L. S.
,
J. O.
Juvik
, and
F. N.
Scatena
,
1993
:
The Puerto Rico tropical cloud forest symposium: Introduction and workshop synthesis.
Tropical Montane Cloud Forests, L. S. Hamilton, J. O. Juvik, and F. N. Scatena, Eds., East–West Center, 1–16
.
Hutchison
,
K. D.
,
2002
:
The retrieval of cloud base height from MODIS and three-dimensional cloud fields from NASA’s EOS Aqua mission.
Int. J. Remote Sens.
,
23
,
5249
5265
.
Lawton
,
R. O.
,
U. S.
Nair
,
R. A.
Pielke
Sr.
, and
R. M.
Welch
,
2001
:
Climatic impact of tropical lowland deforestation on nearby montane cloud forests.
Science
,
294
,
584
587
.
Minnis
,
P.
, and
Coauthors
,
1997
:
Cloud optical property retrieval (subsystem 4.3). CERES Algorithm Theoretical Basis Document, 60 pp
.
Mulligan
,
M.
, and
S. M.
Burke
,
2005
:
Global cloud forests and environmental change in a hydrological context, December 2005. DFID FRP Project ZF0216 Final Tech. Rep., 74 pp. [Available online at http://www.ambiotek.com/cloudforests.]
.
Nadkarni
,
N.
, and
N.
Wheelwright
,
2000
:
Monteverde: Ecology and Conservation of a Tropical Cloud Forest.
Oxford University Press, 620 pp
.
Nair
,
U. S.
,
J. A.
Rushing
,
R.
Ramachandran
,
K. S.
Kuo
,
R. M.
Welch
, and
S. J.
Graves
,
1999
:
Detection of cumulus cloud fields in satellite imagery. Earth Observing Systems IV, W. L. Barnes, Ed., International Society for Optical Engineering (SPIE Proceedings, Vol. 3750), 345–355
.
Nair
,
U. S.
,
R. O.
Lawton
,
R. M.
Welch
, and
R. A.
Pielke
Sr.
,
2003
:
Impact of land use on Costa Rican tropical montane cloud forests: Sensitivity of cumulus cloud field characteristics to lowland deforestation.
J. Geophys. Res.
,
108
.
4206, doi:10.1029/2001JD001135
.
Pielke
,
R. A.
, and
Coauthors
,
1992
:
A comprehensive meteorological modeling system—RAMS.
Meteor. Atmos. Phys.
,
49
,
69
91
.
Pounds
,
J. A.
,
M. P. L.
Fogden
, and
J. H.
Campbell
,
1999
:
Biological response to climate change on a tropical mountain.
Nature
,
398
,
611
614
.
Pounds
,
J. A.
, and
Coauthors
,
2006
:
Widespread amphibian extinctions from epidemic disease driven by global warming.
Nature
,
439
,
161
167
.
Ray
,
D. K.
,
U. S.
Nair
,
R. O.
Lawton
,
R. M.
Welch
, and
R. A.
Pielke
Sr.
,
2006
:
Impact of land use on Costa Rican tropical montane cloud forests: Sensitivity of orographic cloud formation to deforestation in the plains.
J. Geophys. Res.
,
111
.
D02108, doi:10.1029/2005JD006096
.
Still
,
C. J.
,
P. N.
Foster
, and
S. H.
Schneider
,
1999
:
Simulating the effects of climate change on tropical montane cloud forests.
Nature
,
389
,
608
610
.
van der Molen
,
M. K.
,
2002
:
Meteorological impacts of land use change in the maritime tropics. Ph.D. thesis, Free University of Amsterdam, 274 pp
.
Welch
,
R. M.
,
S.
Asefi
,
J.
Zeng
,
U. S.
Nair
,
Q.
Han
,
R. O.
Lawton
,
D. K.
Ray
, and
V. S.
Manoharan
,
2008
:
Biogeography of tropical montane cloud forests. Part I: Remote sensing of cloud-base heights.
J. Appl. Meteor. Climatol.
,
47
,
960
975
.

Footnotes

Corresponding author address: Udaysankar S. Nair, Earth System Science Center, National Space Science and Technology Center, The University of Alabama in Huntsville, Huntsville, AL 35806. Email: nair@nsstc.uah.edu

1

SERVIR was created at the behest of the governments of Central America and is recognized as a premier implementation of the Global Earth Observation System of Systems (GEOSS) concept in providing Mesoamerica with a first-of-its-kind automated, integrated, Internet-based system for near-real-time monitoring and forecasting of the environment. The system is implemented jointly by NASA, the Central American Commission for the Environment and Development (CCAD), the Water Center for the Humid Tropics of Latin America (CATHALAC), the U.S. Agency for International Development (USAID), the World Bank, and other partner institutions.