Introduction
The Global Energy and Water Cycle Experiment Cloud System Study (GCSS; Browning 1993) has identified a number of areas of cloud research that should be given priority. Among these areas are the effect of surface topography on the initiation, structure, and evolution of cloud systems; and interactions between surface fluxes and clouds. Both interactions are important over a range of scales that extend from the mesoscale to the climate scale. Although GCSS is specifically concerned with issues relating to the parameterization of these effects in the physics algorithms used by numerical models of the atmosphere, there are many other reasons to study clouds at high spatial resolution over regions of complex topography. High-resolution analyses of cloud distributions in such areas have the potential to be useful in the validation of numerical weather prediction and regional climate model outputs (Karlsson 1996) and will be required for the initialization of mesoscale weather prediction models (e.g., Saunders and Kriebel 1988; Puri and Davidson 1992). Low-resolution (i.e., 5° × 5°) International Satellite Cloud Climatology Project (ISCCP)–based cloud analyses have already been used to reveal deficiencies in global climate models (Weare et al. 1995).
Cloud-type information is also important, because cloud-top height and optical thickness affect outgoing longwave radiation and albedo as much as cloud cover does (Ockert-Bell and Hartmann 1992; Weare 1992; Oreopoulos and Davies 1993). Similarly, to estimate the surface spatial distribution of such environmentally important quantities as UV-B (290–320 nm) irradiance there is a need to have knowledge of both cloud cover and type (McKenzie et al. 1998).
To investigate any of these questions requires the use of satellite data, but little research has been reported on the problem of determining high-resolution cloud-amount and -type climatological distributions over topographically complex regions. Reinke et al. (1992) generated monthly climatological distributions as a function of the time of day by compositing Geostationary Operational Environmental Satellite (GOES) data at 2.5-km spatial resolution. That analysis clearly shows the effects of topography on the diurnal cycle of cloudiness and identifies sharp gradients in cloudiness at coastlines and along mountain ranges. Mean monthly cloudiness in the Australian region during 1979 is reported in McGuffie (1993); who used U.S. Air Force 3D nephanalysis products (Hughes and Henderson-Sellers 1985) to determine a medium-resolution (i.e., 46 km) climatological distribution. The most striking spatial pattern in the results is a strong land/sea contrast in cloudiness, with higher cloud amounts over ocean areas. Advanced Very High Resolution Radiometer (AVHRR) data have been used to investigate the occurrence of cloud over the Nordic countries (Karlsson 1997). Although the results presented in that study are used primarily to validate the cloud classification algorithm used, differences in surface characteristics are also shown to modify inferred cloud frequencies. Land–sea contrasts are particularly pronounced during the summer of the year analyzed (1993), with cloud frequencies being lowest over sea areas, the reverse of the previous studies noted. Karlsson also detected a minimum in cloud occurrence adjacent to the Norwegian coast. This minimum is attributed to a minimum in sea surface temperature (SST) adjacent to the coast (perhaps caused by snowmelt runoff) leading to reduced convection, as compared with regions farther offshore that are affected by warmer Gulf Stream waters. The study also showed that 2.5° × 2.5° resolution ISCCP (Rossow and Schiffer 1991) data have inadequate spatial resolution for regional cloud-climate studies over complex topography.
In this paper, the high–spatial resolution “SRTex” cloud classification algorithm described in Uddstrom and Gray (1996) is used to analyze cloud cover and cloud-type characteristics over a topographically complex region. The analysis domain (Fig. 1) includes the South Island of New Zealand and its 800-km-long mountain range (the Southern Alps) in which many peaks exceed 2500 m in height. In the central section of the island, this range lies within 40 km of the western coast; east of it there is a substantial plain that is subject to strong föhn effects (see Fig. 2). Because of the strong relief of the Southern Alps and their barrier location in the southern midlatitude westerly belt, they have a substantial effect upon the weather of New Zealand (Wratt et al. 1996). The spatial characteristics of SST features (i.e., water masses) surrounding the South Island are also complex (Uddstrom and Oien 1999). At the large scale, west of the South Island, the Tasman Sea has the highest meridional gradient in SST of any ocean area in the New Zealand–Australia region and is one of the centers of action for the development of cyclonic storms in the Southern Hemisphere (Sinclair 1995, 1997). Close to the west coast (of the South Island), warm subtropical waters are found offshore, but the surface location of the Subtropical Front (STF) is not distinct. In the mean, it probably lies close to 42°S, curving southward adjacent to the coast (Uddstrom and Oien 1999), thereby allowing subtropical waters to flow around the southern tip of the island. A narrow band of colder upwelled and/or river-runoff water lies adjacent to the west coast and is associated with a northward-moving nearshore current. East of the South Island, the STF is bathymetrically locked to the shelf edge and lies over the 500-m isobath. To the north, it is constrained by the Chatham Rise, a bathymetric platform that extends 1000 km east of the island (see Fig. 1). In Fig. 2, the mean position of the STF east of the South Island is represented (approximately) by the 11.5°C isotherm. South of the STF, much colder sub-Antarctic waters (SAW) are found, leading to high spatial gradients in SST close to the east coast of the South Island.
In this paper, cloud type and amount are estimated from NOAA-14 daytime passes over the region during 1995, and high–spatial resolution annual and seasonal statistics are estimated. This year was chosen because it includes a crossover from El Niño to La Niña conditions, both of which have strong influences on New Zealand’s weather and climate (Gordon 1986; Mullan 1995). Although it would be preferable to analyze a longer data series to specify a realistic climatological distribution, the current analysis is still expected to yield insights into the issue of cloud-type spatial patterns in the presence of strong topographic forcing. The temporal compositing method used here also allows coanalysis of contemporaneous high-resolution AVHRR radiance data and SST observations. These ancillary data provide opportunities to analyze cloud–SST interactions, to specify the characteristic radiance temperatures (and albedos) of the classified cloud types, and to estimate cloud-top height.
Section 2 outlines the data used and the analysis methods employed, both those relating to the SRTex algorithm and those relating to the design of the cloud database from which the cloud climatological analyses are drawn. The results from the annual and seasonal analyses and their validation are reported in section 3, and the conclusions are reported in section 4.
Data and analysis methods
The SRTex analyses reported here utilize measurements from the AVHRR on the NOAA-14 spacecraft (Planet 1988) during 1995. More particularly, these include data from the 0.55–0.68-μm (R1), 0.725–1.1-μm (R2), 3.55–3.93-μm (T3), 10.5–11.5-μm (T4), and 11.5–12.5-μm (T5) channels. Cloud-cover validation data are derived primarily from two sources: daily 0900 (local time) cloud-cover observations, and estimates of percentage of possible bright sunshine that are derived from daily Campbell–Stokes sunshine recorder measurements (New Zealand Meteorological Service 1987; Penney 1997) adjusted for local horizon effects. Cloud statistics from Warren et al. (1986) surface-observed climatological distributions (henceforth abbreviated SOCS) are also considered.
SRTex algorithm
The SRTex cloud classification algorithm (Uddstrom and Gray 1996) uses multispectral AVHRR radiative observations together with first- and second-order measures of spatial texture within local tiles of AVHRR data to classify cloud into multilevel meteorological classes. It is similar to the method developed by Ebert (1987) and used by Lubin and Morrow (1998) to classify cloud type over the Arctic but utilizes a wider set of textural and multispectral features and a very extensive “training” sample to derive the Bayesian discriminant functions.
To summarize, the SRTex algorithm involves two steps. In the first, a very large sample of AVHRR measurements from all times of the year and a range of midlatitude locations is labeled by cloud type. This procedure is carried out using a supervised classification approach, with clouds being labeled according to the meteorological context and their radiative and textural signatures. This process is not restricted to classes of cloud that are separable in one- or two-dimensional radiance space. Indeed some labeled samples may not be separable from the AVHRR radiance domain regardless of the radiative features employed. Presently, the labeling analysis yields 13 classes (with abbreviations in parentheses): no cloud over sea (NC), no cloud over land (NCL), stratus (St), stratocumulus (Sc), cumulus (Cu), altostratus (As), altocumulus (Ac), nimbostratus (Ns), cumulonimbus (Cb), thick cirrus (Cs), transmissive cirrus (Ci), embedded cumulus (eCu), and open-cell cumulus (oCu). Each sample is characterized on an 8 × 8 pixel (1.1-km resolution) AVHRR data tile by its multispectral, mean radiative, gray-level difference (GLD) textural (Weszka et al. 1976), standard deviation and range statistics. The scan angle of the instrument, together with the solar geometry, is also included.
In the second step, the radiative and spatial data are used to specify Bayesian discriminant functions that may then be used to classify cloud, given the satellite observations alone. These discriminant functions can be derived using all prior (meteorological) cloud classes or a lesser number of classes, either by aggregating prior classes, or by selecting a subset of the classes.
The feature vector for the discriminant function employed here includes: R2, the GLD entropy of R2, [(R2/R1) − 1], T4, (T4 − T5), the GLD entropy of T4, and the scan angle. It is applied to the orbital data after they have been corrected for residual navigation errors through coastal landmark matching. The [(R2/R1) − 1] vegetation-index feature is used for this purpose, because, under transmissive cloud conditions, the differential reflectance of chlorophyll in R2 (as compared with R1) often reveals land–sea boundary features when they are not evident in either channel individually. The navigation correction step is necessary because it reduces the impact of navigation errors (invariably attributable to undetected satellite ephemeris errors) on the composited (i.e., mean climate) data. Here, based on the posterior probability (using equal priors), any particular 8 × 8 tile of AVHRR data is classified into one of 12 classes: NC, NCL, St, Sc, Cu, As, Ac, Ns, eCu, Ci, Cs, and unclassifiable. The data are assigned to the “unclassifiable” bin if the class with the highest posterior probability is less than 0.25. In this case, they are likely to represent coastlines, nonhomogeneous cloud fields (at the 8-km scale) such as cloud edges, and other situations not well represented in the labeled (a priori) dataset. Data are only classified if they lie within ±50° of nadir, the local solar zenith angle is less than 80°, and they lie outside the specular sunglint region.
In practice, the discriminant function is applied using an overlapping-data-tile approach. After a classification is made (and the result assigned to the “center” pixel position), the column index within the analysis domain is increased by one, and the discriminant function is applied again. After the last column has been processed, the row index is increased by one, and the column processing is repeated. This approach yields a 1-km-resolution cloud classification over the analysis domain (apart from the outer four rows and columns).
Clearly, the interpretation of ensuing climatological analyses derived through temporal compositing of these data is dependent upon the fidelity of the discriminant function employed. This fidelity can be summarized using standard skill-score measures for categorical contingency tables. At a more detailed level, the probability of detection (POD) and the false-alarm rate (FAR) (Murphy and Katz 1985) of each class provide additional information about the performance of the discriminant function, and still more information is provided by the contingency table itself, which indicates where “misclassified” samples were placed. An understanding of which classes may be accurately classified and which may not can be gained from examination of these statistics.
The contingency table resulting from applying the specified discriminant function (above) to the a priori labeled data is shown in Table 1, which also includes the resulting POD and FAR statistics. The overall hit rate (i.e., the trace of the contingency table divided by the total sample size) is 0.78, and the Kuipers performance index is 0.76, indicating high overall skill. In all cases, the POD for each cloud class is better than 0.5 (the prior probability being 0.09). The PODs for no cloud, both land and sea, are essentially unity, with near-zero FARs. Likewise, the PODs for transmissive and thick cirrus are high and the FARs are low, with most misclassifications lying within these two classes. The POD for stratus is high at 0.98, but so too is its FAR at 0.45. This result suggests that distributions derived from the current discriminant function should not be used to examine differences between stratus and stratocumulus climatological distributions. The nimbostratus and cumulus classes also have high FARs, but again, the true classes of samples classified into these two classes are unlikely to cause difficulties in interpreting resultant analyses (e.g., a small number of stratocumulus samples classified as cumulus is acceptable in the current context). The altostratus classification may be contaminated by low-level cumuliform clouds (i.e., stratocumulus), which may make it less useful for further analysis, although the majority of samples classified into this class are midlevel clouds (i.e., altocumulus and altostratus). The POD for the “embedded cumulus” class is the lowest of all, but this result reflects the situation that this class of cloud is by definition composed of a mixture of cumuliform cloud types through a range of altitudes.
An example classification of the meteorological situation indicated in Fig. 3a, an R2 (i.e., 0.9 μm) image, is shown in Fig. 3b, in which the cloud class is represented by a color, and the posterior probability is represented by the intensity of that color (less intense colors correspond to lower posterior probabilities—see the figure legend). Evidently, the classification algorithm has detected the no-cloud-over-land and no-cloud-over-sea areas, as well as the areas of thick and transmissive cirrus. Likewise, the cumulus (magenta) and stratocumulus (teal, i.e., blue–green) fields over the ocean are well identified as are the areas of embedded cumulus (red–brown, or scale values 130–139). The gray areas around the edges of many clouds indicate that these regions are, as desired, unclassifiable, because they contain mixtures of cloud types.
Coincident SST data are also available. These data are derived by mean-value time-compositing SSTs derived from cloud-cleared AVHRR data (Uddstrom and Oien 1999). The cloud mask used in that analysis employs a two-state (cloudy/not cloudy) Bayesian classifier on 3 × 3 AVHRR instantaneous-field-of-view data tiles (Uddstrom et al. 1999).
Cloud database design
Spatial and temporal distributions of cloud amount, cloud type, and cloud radiative characteristics have been estimated by constructing cloud databases (CDBs) that collate the relevant satellite information (AVHRR, SRTex) and associated data (e.g., SST). The contents of any particular CDB are defined by a set of selection rules that specify the data fields to be included, the observing system (e.g., a specific satellite, ascending and/or descending passes), as well as time-of-year selection details. These rules, together with audit-trail information such as the SRTex discriminant function used and the orbital details of all data analyzed, are stored within each CDB so that they may be automatically updated as new data become available—without risk of corruption.
At each location within the analysis area (Fig. 1), the CDB accumulates sums of quantities (e.g., SRTex cloud class) and maintains sample counts for each such quantity. This approach avoids round-off errors and permits the CDB to be updated with new information at a later date. Also, in addition to the frequency of occurrence of a cloud class, that is, the ratio of class count to the sum of all class counts, the conditional frequency of occurrence may be estimated. This metric is defined as its frequency of occurrence as a fraction of the total number of times that it could have been observed; that is, it would not have been obscured by higher-level cloud. This method is identical in concept to the “frequency-of-occurrence” statistic specified in the Warren et al. (1986, 1988) surface-based cloud climate descriptions, but results will differ because one is a sky view and the other is an earth view.
In addition to SRTex data (i.e., cloud class and posterior probability), other associated data may be accumulated in the CDB for later analysis. These data can be included simply by location, referred to as ancillary data (AD), and/or partitioned according to SRTex cloud class, and hence referred to as cloud-class-specific ancillary data (CCSAD). CCSAD data allow within–cloud class variations in cloud characteristics to be analyzed. For example, the CDB could include the T4 radiance temperature and SST data averaged over all occurrences of cumulus at a particular location, allowing coanalysis of these quantities, or their dependence on location within the area under analysis. The AD data provide background fields of relevant quantities, independent of cloud class but averaged using the time-selection rules utilized by the CDB. A given quantity may be included as both CCSAD and AD data types. Using SST as an example, differences in SST associated with some particular cloud class as compared with the mean SST for the CDB period may be determined. CCSAD and AD data are not limited to a particular satellite system; data from other platforms (e.g., radar rain rate or geostationary satellite radiances) may also be included in a CDB. When AD or CCSAD data are not exactly cotemporal with the SRTex data (e.g., SST) they are selected so as to minimize the time difference, subject to some maximum allowable time difference. Here, the collocated SST data are derived from 14-day mean-value composites (Uddstrom and Oien 1999) computed at 7-day intervals. Accordingly, general associations between cloud frequencies and SST should be detectable for monthly and longer timescales.
Results
Annual cloud cover
The no-cloud-cover (complement of the cloud amount) frequency (i.e., NC and NCL), averaged over the year (from 339 NOAA-14 orbits), and the mean SST from the same period (i.e., an AD analysis) are shown in Fig. 4. Because the current implementation of the SRTex algorithm does not discriminate between snow and low cloud over highly variable terrain, areas with elevations above 750 m have been masked in this figure. This limitation does not affect the detection of mid- and high-level clouds over high terrain. Also, because of the size of the discriminant function tile (i.e., 8 km × 8 km), the climate dataset is unlikely to yield accurate estimates in regions of strong coastal curvature, as is evident in the Tasman Bay region (geographical locations noted in the text are identified on Fig. 1).
A number of features are evident in the figure. Apart from areas along the west coast, in mountain valleys, and over Tasman Bay, the cloud amount over land is much lower than that over adjacent ocean areas. Reinke et al. (1992) noted similar land–sea contrasts in an analysis of GOES data from the western United States, as did an analysis of 3D Nephanalysis data from the Australian region (McGuffie 1993). The least-cloudy locations are the Canterbury Plains in the lee of the Southern Alps, the Wairau and Waitaki Valleys, and the Nelson region and adjacent Tasman Bay, both of which are sheltered from prevailing westerly quarter winds by mountains. Clearly the upstream orography has a strong effect on cloud-cover frequency. Over SAW to the east of the South Island (see Fig. 1 for subanalysis areas), the spatial average of the cloud amount is 95% (±3%); that over SAW west of the island is 90% (±2%). Adjacent to the west coast, the cloud-cover frequency is much lower, being less than 75% on average. This result was unexpected, because predominant upslope winds from the westerly quarter were expected to lead to high cloud-cover frequencies in this region, in line with shore-based observations. Coincidentally, Karlsson (1997) observed a similar situation in the annual mean for 1993 west of Norway, which he attributed to cold nearshore waters suppressing convection. However, the SST structure west of the South Island is different from that noted by Karlsson (1997). Although cold river and upwelled waters are found within about 30 km of the west coast of the South Island, the dominant SST feature is the presence of warm subtropical waters (STW) beyond the nearshore waters and colder SAW waters farther offshore (see Fig. 4).
On the east coast, there is some evidence of both SST and orographic effects on the cloud-amount statistics. The spatially averaged cloud-cover frequency over the STW north of the STF is 81% (±4%), or 14% lower than over the adjacent SAW water mass. However, this reduction may also be associated with the exit region from Cook Strait, because the cloud-cover frequency is higher for STW in the eastern Tasman Sea. On either side of Cook Strait, mountains rise to approximately 1000 m, leading to convergence and significant wind speedup in the strait. Accordingly, flows are channeled in a northwest/southeast plain, leading to ascending air at the entry region and descending (drying) air in the exit region. The predominant flow is northwesterly.
Verification
Five representative coastal sites (Nelson, Blenheim, Christchurch, Hokitika, and Invercargill) were chosen to validate the SRTex cloud-amount results. These sites encompass a location west of the mountains (Hokitika) and a coastal site to the east of the mountains (Christchurch), a southern site (Invercargill), and two orographically sheltered sites (Nelson and Blenheim) at the northern end of the island.
Statistics of mean monthly cloud cover at 0900 local time (LT) were available from Blenheim, Invercargill, and Hokitika, but not Christchurch or Nelson; percentage-of-possible-bright-sunshine statistics were available from all five sites. To allow intercomparison of cloud cover at all sites, the sunshine data were regressed against the 0900 LT cloud-cover data. The adjusted R2 of the fit was 0.96 and the standard error of the estimate 1.5%. Applying the resulting equation to the Christchurch and Nelson sunshine data led to estimates of the 0900 LT cloud cover that could then be compared with the SRTex data.
The results are shown in Fig. 5. The (annual) mean surface observed cloud amount from the five sites is 67.5%; that derived from the SRTex analysis is 67.2%, and the explained variance is 0.79. This value is similar to that given in the SOCS analysis, for which the mean annual cloud amount is 64.4% over the four 5° × 5° grid squares bracketing the South Island. However, the differences in the analysis scales and the spatial representativeness of the data used in the SOCS analysis preclude further comparison of the results. The essential point here is that these results confirm the deductions drawn from analysis of the contingency table (Table 1), which indicated that the SRTex algorithm accurately detects cloud over land regions. Given that the performance of the SRTex cloud/no-cloud algorithm is essentially equal for both land and sea areas, this result suggests that the spatial patterns of cloud cover identified in Fig. 4 are real.
Seasonal cloud cover
Cloud-cover statistics for the seasonal periods (see Trenberth 1983): January, February, March (JFM); April, May, June (AMJ); July, August, September (JAS); and October, November, December (OND) in 1995 are plotted in Fig. 6. To reduce sampling noise, the resolution of the analysis has been reduced from 1 km × 1 km (as in Fig. 4) to 16 km × 16 km through spatial averaging. Even with this averaging, there is an area of missing data over the ocean in the JAS plot. This area corresponds to a very cloudy region in which the SRTex posterior probability for the NC class was always too low to pass the posterior probability threshold test used in the orbital processing. As above, land over 750-m elevation has been masked. The coincident SST analyses shown in Fig. 6 are derived from the noncloud-class specific (i.e., AD) SST data in the CDBs and, hence, correspond exactly with the selected (3 month) data periods.
Figure 6 reveals a clear annual cycle in cloud-cover frequency over both land and ocean surfaces. The amplitude is larger over land. East of the Southern Alps and in Tasman Bay, January, February, and March (i.e., summer) are the least cloudy months. Also, apart from around Banks Peninsula, the area of low cloud-cover frequency evident in the lee of the Southern Alps extends a significant distance offshore over neritic and STW water masses shoreward of the Southland Front. Coincidentally, Uddstrom and Oien (1999) show that, in this same region, the amplitude of the annual cycle in SST is among the highest (>4°C) in the New Zealand region (one other being Tasman Bay). Also, the phase of the SST annual cycle, between 2 and 6 weeks, is more typical of low-latitude locations. It seems likely that the SST is responding to changes in insolation as a result of changes in cloud cover, because this lag is much less than that expected in the oceanic midlatitudes (e.g., Prescott and Collins 1951; Trenberth 1983).
Figure 6 also indicates that throughout the year there is a cloud-cover minimum or clear slot near to the western coast over the region where warm STW push against the cold waters adjacent to the coast. This area extends between 60 (in JAS, winter) and 150 km (in JFM, summer) offshore, with a stronger minimum closer to the coast extending about 50–70 km west from the region of steepest orography (see Fig. 2). This feature is present throughout the year, leading to cloud-cover frequencies as low as 65%, as compared with 90% farther offshore and 80% onshore. It is well known that the west coast area is more likely to be cloud free when an anticylone is centered southeast of New Zealand, leading to a flow over the South Island from the southeasterly quarter or when a northeasterly flow anomaly is present. Although the orography of the Southern Alps is probably the most important factor in understanding the presence of this clear slot, it also coincides with a region where there is southwestward advection of warm STW, sandwiched between SAW farther offshore and cold northeastward-flowing waters adjacent to the coast.
A comparison of the SRTex seasonal cloud cover and percentage of bright sun is shown in Fig. 7, which indicates that much of the variance (R2 = 0.69) in the observed variations in bright sun can be explained by the SRTex cloud-cover estimates. The 0900 LT cloud-cover statistics (not shown) yield an R2 of 0.64, with the mean SRTex cloud cover being 65.9%, and mean observed cloud cover being 65.6%. These values again indicate that, at least in case of total cloud cover and for the chosen sites, the SRTex estimates are essentially unbiased with respect to a ground-based observer.
Annual cloud-type distribution
The spatial distributions of transmissive cirrus (Ci), thick cirrus (Cs), cumulus (Cu), and stratocumulus (Sc) during 1995 are shown in Fig. 8. The cloud-cover frequencies for the Sc and Cu classes are expressed using the conditional frequency approach noted above, that is, as a percentage of the possible occurrence, rather than of total occurrence.
Over ocean areas to the west of the South Island, the distribution of transmissive cirrus (Fig. 8a) is essentially spatially homogeneous and on the order of 10%. However, east of the Southern Alps the mean occurrence is close to 20%, rising to 25% over the southern region of the island. This region of orographically enhanced cirrus extends offshore along much of the South Island’s east coast. The distribution of thick cirrus (Fig. 8b) is more spatially uniform, with values of about 6% over the ocean to the west of the South Island, and peak values of 11%–14% in the lee of the Southern Alps. These values are comparable to High-Resolution Infrared Sounder results reported in Wylie and Menzel (1999), which, in turn, are somewhat larger than equivalent ISCCP values. However, all three satellite estimates of high-cloud occurrence (i.e., ISCCP, Wylie and Menzel, and SRTex) are much lower than the frequency of occurrence values given in the SOCS climatological distribution.
The cumulus (Fig. 8c) and stratocumulus (Fig. 8d) spatial distributions show much larger variances over the analysis domain and may be dominated by orographic effects. Of those occasions on which cumulus clouds could have been observed, they were present 27% of the time on average (as compared with 15% of all cloud occurrences). Maximum occurrences reach 35% along the south coast and over STW and SAW west of the South Island. The maximum along the south coast suggests that enhanced convection occurs as air flows (from the southerly quarter) off cold SAW over warm STW in this region (Uddstrom and Oien 1999). Cumulus occurrence is suppressed over the cold waters adjacent to the west coast (19%), over the Tasman Bay (9%), and Cook Strait (11%) regions, and over eastern areas of the South Island (7%). The SOCS analyzed annual frequency of occurrence of cumulus over the land area of the South Island (i.e., the four bracketing 5° × 5° grid boxes) is 17%, close to the SRTex annual mean conditional frequency (over the land and coastal areas) of 18%.
Considering the whole area, stratocumulus is present on 45% of all occasions when it could have been observed from space, as compared with 25% of all cloud occurrences. There is some indication that stratocumulus occurrence is suppressed immediately west of the Southern Alps, in Tasman Bay, and in the exit region from Cook Strait. There is also some indication of a relationship between stratocumulus occurrence and SST features. Over the cold, bathymetrically locked SAW water mass to the east of the South Island, the conditional frequency of stratocumulus occurrence reaches values above 60%; over the warmer STW waters shoreward of the Southland Front, stratocumulus occurrence is less than 45%. Likewise, north of the STF, the frequency of occurrence of stratocumulus falls to about 40%. An equivalent pattern is less obvious west of the South Island, although it is clear that stratocumulus occurrence over the SAW water mass in the southern sector is greater than that over the STW water mass to the north. These patterns are expected, because northerly quarter flows will be cooled as they move over SAW, leading to an increase in boundary layer cloud occurrence.
More detailed measures of cloud-type conditional frequency, associated 11-μm radiance temperature (i.e., T4), and mean (SST − T4) difference statistics over the domains identified in Fig. 1 and the four cloud types discussed above are given in Table 2. The conditional frequency statistics suggest that there are significant differences in stratocumulus occurrence over STW and SAW east of the South Island—where the regional SST differences are large, distinct, and bathymetrically locked. Cumulus occurrence is more frequent west of the island (as also reported in the SOCS analysis), but, as already mentioned, the reverse is true for cirrus and thick cirrus clouds. Perhaps of more interest in Table 2 are the cloud-class-specific (CCSAD) statistics for the mean 11-μm radiance temperature (T4) and SST over the areas in question. The quantity (SST − T4) can be regarded as a proxy for cloud-top height, if it is assumed that spatial and temporal averaging yield a similar mean lapse rate over all sea areas in the analysis domain. This fact suggests that stratocumulus and cumulus cloud tops extend to higher altitudes over STW (in comparison with SAW), and to the west of the island (as compared with the east). If a standard lapse rate of −6.5°C km−1 is assumed, then the mean cloud-top height of stratocumulus over SAW to the east of the island is around 970 ± 70 m; west of the island it is 1260 ± 110 m. The SOCS climatological distribution also reports an east–west difference in cloud-base height for the “stratus and stratocumulus” class (the closest class to the SRTex Sc class reported here). The SOCS western grid box has a mean cloud-base height of 900 m; to the east the mean base height is 700 m, consistent with the SRTex cloud-top height results. For cumulus clouds there are no significant east–west or north–south differences in the SRTex mean cloud-top heights (being near 1830 ± 170 m).
Assuming no atmospheric contribution, the measured radiance of the transmissive cirrus, R4, is εB(ν, T) where ε is the effective cloud emissivity, B is the blackbody function, ν is the (observed) wavenumber, and T is the cloud-top temperature. Wylie and Menzel (1999) report that, for semitransparent clouds, the infrared effective emissivity lies between 0.5 and 0.6. The mean radiance temperature for transparent cirrus identified by SRTex (CCSAD data), ranges from −10°C over western SAW to −17.2°C over eastern STW downstream from the Kaikoura Mountains. Assuming ε = 0.55, these values indicate cloud-top temperatures in the range −30° to −43°C. Although there are no significant differences in the value of (SST − T4) between these areas, the cloud-top temperatures are significantly different, indicating higher cloud tops downstream from the mountains. For the thick cirrus class for which the emissivity is expected to be closer to unity, the mean value of T4 is near −40°C and the value of (SST − T4) varies from 50°C over eastern SAW to 57°C over eastern STW.
Consequently, Table 2 suggests that, over the analysis area used in this study, the simple quantity (SST − T4) leads to separability of these four cloud classes (Sc, Cu, Ci, and Cs). This result is significant, because SST was not used by the SRTex classifier and accordingly is an independent variable here.
When the data from all four marine areas are considered together (sample size 2005), the relationships among cloud frequency, cloud-top height (for the nontransmissive classes), and SST can be examined. For the stratocumulus class, there is no relationship between the cloud-top height (i.e., SST − T4) and cloud frequency (R2 = 0.004). However, as noted above, there is a significant relationship between cloud-top height and SST (R2 = 0.57), with increasing cloud-top height associated with increasing SST, as also noted by Weare (1994). The conditional cloud frequency and SST show a weak negative relationship (R2 = 0.20), indicating that higher cloud occurrences are associated with lower SSTs, as observed above and as noted by other authors such as Hanson (1991), Weare (1992), and Oreopoulos and Davies (1993). However, if the same analysis is carried out for the simple cloud-frequency data, the R2 value is similar (at 0.18), but the correlation is of the opposite sign. Evidently, as noted elsewhere (e.g., Ockert-Bell and Hartmann 1992), analyses of satellite estimates of cloud frequency should consider the effect of overlying clouds. Accordingly, although the relationship between cloud amount and SST is, in general, weak, that for cloud-top height and SST is strong for stratocumulus clouds, indicating that surface radiation and/or sensible heat fluxes have some effect on the characteristics of low-level clouds.
For the cumulus class, as for stratocumulus, there is no strong relationship between cloud-top height and cloud frequency. Neither is there a relationship between cloud-top height and SST or between the cloud conditional frequency and SST. As for the stratocumulus and cumulus classes, the thick cirrus class shows no relationship between cloud frequency and cloud-top height. There is, however, a positive relationship between cloud-top height and SST (R2 = 0.48) but not between cloud frequency and SST.
Seasonal cloud type
An analysis of data from a single year means that no general statements can be made about “seasonal” cloud cover by type. However, examination of the resulting spatial distributions is instructive. Figures 9 and 10 show the SRTex-derived (conditional frequency) spatial distributions for the stratocumulus and cirrus classes and for the seasons defined in section 3b. The stratocumulus figure also includes an analysis of the coincident and cotemporal SST at the time stratocumulus was observed (i.e., CCSAD SST).
Although the mean occurrence of stratocumulus over the whole area and year is 50%, with maxima in autumn (AMJ) and spring (OND), Fig. 9 shows significant seasonality in its spatial distribution. There is some evidence that the land–sea contrast is stronger in summer and winter, and, except in summer, that the region of maximum occurrence is associated with SAW east of the South Island. West of the island there appears to be a minimum in stratocumulus occurrence adjacent to the coast, regardless of season. Unlike the situation east of the island, Fig. 9 indicates this minimum is not associated with SST or SST spatial gradients but rather with the orography. This observation supports the discussion above, relating to the presence of minima in the annual and seasonal spatial distributions of cloud amount adjacent to the western coast.
The mean incidence of transmissive cirrus over the region during 1995 was 13.3% ± 5.3%. The spring (OND) season has the highest mean occurrence at 16.8%; the means for the other periods are close to 12%. The (spatial) standard deviations for all four periods lie between 4.0% and 5.4%. Although these differences are not statistically significant, Fig. 10 shows that the seasonal variations in the spatial distribution of the cirrus almost certainly are significant. The predominant influence of the Southern Alps on the preferred location of cirrus formation is clearly evident. In all seasons there is a maximum, on the order of 20%–25%, in cirrus occurrence east of the mountains (both the Southern Alps and the Kaikoura Mountains). This maximum extends over the largest area in the spring period (OND), when westerly quarter winds dominate.
Summary
In this paper, the high-resolution (1 km), SRTex multispectral textural, Bayesian cloud classification algorithm was applied to one year of NOAA-14 daylight passes over a region of complex topography, and exploratory cloud climatological distributions were developed. This is the first application of an optimal high-resolution cloud classification algorithm to the general problem of specifying regional cloud climatological distributions. Previously reported high-resolution regional cloud distributions have been developed through the application of decision tree algorithms (e.g., Karlsson 1997), or compositing techniques (e.g., Reinke et al. 1992).
The cloud database methodology developed here also allows coanalysis of physically related collocated quantities, such as albedo, radiative temperature, and SST, either as a function of the cloud type or not, and at the same spatial resolution as the cloud climatological distribution. Further, it allows conditional cloud frequencies to be estimated, that is, the frequency of occurrence as a fraction of the total number of times that the cloud could have been observed. From the satellite point of view, this quantity is equivalent to the frequency-of-occurrence statistic reported by Warren et al. (1986, 1988) in their surface-based analyses. This result appears to differ from the quantity often used in satellite-based cloud distributions, in which the cloud frequency is often expressed as a fraction of all possible observations [e.g., Rossow et al. (1993), Ockert-Bell and Hartmann (1992), Doutriaux-Boucher and Sèze (1998)]. Differences between cloud frequencies expressed in terms of conditional frequency versus simple overall occurrence can be large, especially for low- and midlevel clouds, for which the simple method always underestimates the cloud frequency.
The SRTex classifications have been validated both by contingency table measures and, for the no-cloud class, by collocation with ground-based measures of cloud amount (both observer and sunshine data) from a representative set of stations. At the seasonal and yearly timescales analyzed, the SRTex and ground-based cloud-cover statistics show no bias and are found to explain 90% of the annual spatial variance and 70% of the seasonal spatial variance. Comparisons between elements of the Warren et al. (1986) climatological distributions and the SRTex results are also encouraging—especially for the low-level clouds.
The domain over which the climatological distributions are developed is both orographically and radiatively (i.e., SST) complex—an ideal environment to investigate the effect of surface topography on the formation and distribution of clouds. Although derived from a limited sample, the SRTex distributions reveal a number of features that probably result from the effects of orographic and SST variations. The no-cloud distributions reveal a strong land–sea contrast, as also was observed by Reinke et al. (1992) for the western United States and McGuffie (1993) for the Australian region. Lowest cloud frequencies are found in the lee of the Southern Alps and in mountain-sheltered valleys. Over the oceans, cloud frequencies are highest over the colder (SAW) water masses, ranging from 90% to 95%. However, over the sea adjacent to the western coast (on the upwind side of the mountains) there is a distinct minimum in cloud-cover frequency. This minimum extends up to 150 km offshore and was present throughout the year analyzed, but it is weakest in winter (July, August, and September). It is probably related to the orography of the Southern Alps, with the lowest cloud frequencies extending between 50 and 70 km west from the region of steepest orography. Karlsson (1997) found a minimum in cloud occurrence over the ocean west of Norway, which he attributed to nearshore cold SSTs suppressing convection; however, that process appears not to be the one at work in the New Zealand situation. Orography also probably plays a role in the spatial distribution of cloud cover east of the exit from Cook Strait—a region of significant local convergence. Last, there is some evidence that the large amplitude (and early phase) of the annual cycle in SST immediately east of the central region of the South Island [reported in Uddstrom and Oien (1999)] results from increased local insolation arising from low cloud cover in this area during summer (January, February, and March).
The cloud-type climatological distributions reveal the presence of preferred locations for different cloud types. For example, the annual spatial average for cirrus is near 13%, but there is a substantial area of higher occurrence, on the order of 25%, in the lee of the mountains. Further, when analyzed by season, cirrus is most extensive during spring (i.e., October, November, and December) when westerly quarter winds prevail, suggesting a link with uplift over the orography, as was noted also by Jin et al. (1996). Stratocumulus spatial distributions display characteristics that are related to both orographic and SST features. Over the ocean area, stratocumulus observations make up 25% of all cloud occurrences but 45% of all possible observed occurrences (i.e., the cloud conditional frequency). Stratocumulus (and cumulus) occurrence may be suppressed adjacent to the western coast (upwind of the mountains). When considered on a seasonal basis, there is evidence that this effect is unrelated to the presence of cold upwelled water adjacent to the coast, because the pattern is evident both when the cold water is present and when it is not, suggesting (as noted above) an orographic source. East of the island, SST effects dominate, with higher cloud frequencies over sub-Antarctic waters than over subtropical waters, as demonstrated by a negative correlation between SST and the conditional frequency of stratocumulus occurrence. Also, there is a positive correlation between mean cloud-top height and SST for stratocumulus but not for other cloud types (e.g., cumulus).
These results suggest that multispectral textural, Bayesian cloud classification algorithms such as SRTex may be used to derive high-resolution regional cloud climatological distributions. It is hoped that such datasets eventually will provide useful input to GCSS goals that relate to the description of cloud characteristics resulting from interactions with orography and surface radiative fluxes. Future work is expected to extend the analysis through the inclusion of land variability information in the SRTex algorithm (allowing snow detection), application to a much larger (regional) area and to a longer time period, and coanalysis of cloud-albedo (i.e., optical depth) data. Future analyses will also incorporate nighttime data and, by developing distributions as a function of synoptic class using the synoptic classification of Kidson (1999), will examine the question of whether the cloud minimum west of the Southern Alps is merely due to wind-direction effects.
Acknowledgments
This research has been carried out under the support of the New Zealand Foundation for Research Science and Technology (Contracts CO1521 and C01829). The authors thank Stuart Burgess for supplying the cloud and radiation data. Comments from three anonymous reviewers are gratefully acknowledged and led to improvements in the interpretation of the results.
REFERENCES
Browning, K. A., 1993: The GEWEX Cloud System Study (GCSS). Bull. Amer. Meteor. Soc.,74, 387–399.
Doutriaux-Boucher, M., and G. Sèze, 1998: Significant changes between ISCCP C and D cloud climatologies. Geophys. Res. Lett.,25, 4193–4196.
Ebert, E., 1987: A pattern recognition technique for distinguishing surface and cloud types in polar regions. J. Climate Appl. Meteor.,26, 1412–1427.
Gordon, N. D., 1986: The Southern Oscillation and New Zealand weather. Mon. Wea. Rev.,114, 371–387.
Hanson, H. P., 1991: Marine stratocumulus climatologies. Int. J. Climatol.,11, 147–164.
Hughes, N. A., and A. Henderson-Sellers, 1985: Global 3D Nephanalysis of total cloud amount for 1979. J. Climate Appl. Meteor.,24, 669–686.
Jin, Y., W. B. Rossow, and D. P. Wylie, 1996: Comparison of the climatologies of high-level clouds from HIRS and ISCCP. J. Climate,9, 2850–2879.
Karlsson, K.-G., 1996: Validation of modelled cloudiness using satellite-estimated cloud climatologies. Tellus,48A, 767–785.
Karlsson, K.-G., 1997: Cloud climate investigations in the Nordic region using NOAA AVHRR data. Theor. Appl. Climatol.,57, 181–195.
Kidson, J. W., 1999: An analysis of New Zealand synoptic types and their use in defining weather regimes. Int. J. Climatol.,20, 299–316.
Lubin, D., and E. Morrow, 1998: Evaluation of an AVHRR cloud detection and classification method over the central Arctic Ocean. J. Appl. Meteor.,37, 166–183.
McGuffie, K., 1993: Australian cloudiness from a high resolution satellite archive. Aust. Meteor. Mag.,42, 7–15.
McKenzie, R. L., K. J. Paulin, G. E. Bodeker, and J. B. Liley, 1998:Cloud cover measured by satellite and from the ground: Relationship to UV radiation at the surface. Int. J. Remote Sens.,19, 2969–2985.
Mullan, A. B., 1995: On the linearity and stability of Southern Oscillation–climate relationships for New Zealand. Int. J. Climatol.,15, 1365–1386.
Murphy, A. H., and R. W. Katz, 1985: Probability, Statistics, and Decision-Making in the Atmospheric Sciences. Westview Press, 545 pp.
New Zealand Meteorological Service, 1987: Instructions for meteorological observations at reporting stations. New Zealand Meteorological Service Misc. Publ. 101, 107 pp. [Available from Meteorological Service of New Zealand Ltd., P.O. Box 722, Wellington, New Zealand.].
Ockert-Bell, M. E., and D. L. Hartmann, 1992: The effect of cloud type on the earth’s energy balance: Results for selected regions. J. Climate,5, 1157–1171.
Oreopoulos, L., and R. Davies, 1993: Statistical dependence of albedo and cloud cover on sea surface temperature for two marine stratocumulus regions. J. Climate,6, 2434–2447.
Penney, A. C., 1997: Climate database (CLIDB) user’s manual. 4th ed. NIWA Tech. Rep. 4, 95 pp. [Available from National Institute of Water and Atmospheric Research, P.O. Box 14 901, Wellington, New Zealand.].
Planet, W. G., Ed., 1988: Data extraction and calibration of TIROS-N/NOAA radiometers. NOAA Tech. Memo. NESS 107-Rev 1, 58 pp. [Available from NOAA/NESDIS, Washington, DC 20233.].
Prescott, J. A., and J. A. Collins, 1951: The lag of temperature behind solar radiation. Quart. J. Roy. Meteor. Soc.,77, 121–126.
Puri, K., and N. E. Davidson, 1992: The use of infrared satellite cloud imagery data as proxy data for moisture and diabatic heating in data assimilation. Mon. Wea. Rev.,120, 2329–2341.
Reinke, D. L., C. L. Combs, S. Q. Kidder, and T. H. Vonder Haar, 1992: Satellite cloud composite climatologies: A new high-resolution tool in atmospheric research and forecasting. Bull. Amer. Meteor. Soc.,73, 278–285.
Rossow, W. B., and R. A. Schiffer, 1991: ISCCP cloud data products. Bull. Amer. Meteor. Soc.,72, 2–20.
Rossow, W. B., A. W. Walker, and L. C. Gardner, 1993: Comparison of ISCCP and other cloud amounts. J. Climate,6, 2394–2418.
Saunders, R. W., and K. T. Kriebel, 1988: An improved method for detecting clear sky radiances from AVHRR data. Int. J. Remote Sens.,9, 123–150.
Sinclair, M. R., 1995: A climatology of cyclogenesis for the Southern Hemisphere. Mon. Wea. Rev.,123, 1601–1619.
Sinclair, M. R., 1997: Objective identification of cyclones and their circulation intensity, and climatology. Wea. Forecasting,12, 595–612.
Trenberth, K. E., 1983: What are the seasons? Bull. Amer. Meteor. Soc.,64, 1276–1282.
Uddstrom, M. J., and W. R. Gray, 1996: Satellite cloud classification and rain-rate estimation using multispectral radiances and measures of spatial texture. J. Appl. Meteor.,35, 839–858.
Uddstrom, M. J., and N. A. Oien, 1999: On the use of high resolution satellite data to describe the spatial and temporal variability of sea surface temperatures in the New Zealand region. J. Geophys. Res.,104, 20 729–20 751.
Uddstrom, M. J., W. R. Gray, R. Murphy, N. A. Oien, and T. Murray, 1999: A Bayesian cloud mask for sea surface temperature retrieval. J. Atmos. Oceanic Technol.,16, 117–132.
Warren, S. G., C. J. Hahn, J. London, R. M. Chervin, and R. L. Jenne, 1986: Global distribution of total cloud and cloud type amounts over land. NCAR Tech. Note TN-273+STR/DOE Tech. Rep. ER/60085-HI, 29 pp and 200 maps. [NTIS DE87-00-6903.].
Warren, S. G., C. J. Hahn, J. London, R. M. Chervin, and R. L. Jenne, 1988: Global distribution of total cloud and cloud type amounts over the ocean. NCAR Tech. Note TN-317+STR/DOE Tech. Rep. ER-0406, 42 pp and 170 maps. [NTIS DE90-00-3187.].
Weare, B. C., 1992: A comparison of ISCCP C1 cloud amounts with those derived from high resolution AVHRR images. Int. J. Remote Sens.,13, 1965–1980.
Weare, B. C., 1994: Interrelationships between cloud properties and sea surface temperatures on seasonal and interannual time scales. J. Climate,7, 248–260.
Weare, B. C., I. I. Mokhov, and Project Members, 1995: Evaluation of total cloudiness and its variability in the Atmospheric Model Intercomparison Project. J. Climate,8, 2224–2238.
Weszka, J. S., C. R. Dyer, and A. Rosenfeld, 1976: A comparative study of texture measures for terrain classification. IEEE Trans. Syst. Man Cybernetics,SMC-6, 269–285.
Wratt, D. S., and Coauthors, 1996: The New Zealand Southern Alps Experiment. Bull. Amer. Meteor. Soc.,77, 683–692.
Wylie, D. P., and W. P. Menzel, 1999: Eight years of high cloud statistics using HIRS. J. Climate,12, 170–184.
Map showing the 500-m and 1000-m isobaths, cloud analysis domain (solid line), and subanalysis areas (smaller polygons), i.e., STW (west), STW (east), SAW (west), and SAW (east), together with the geographical locations referred to in the text.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Map showing the 500-m and 1000-m isobaths, cloud analysis domain (solid line), and subanalysis areas (smaller polygons), i.e., STW (west), STW (east), SAW (west), and SAW (east), together with the geographical locations referred to in the text.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Map showing the 500-m and 1000-m isobaths, cloud analysis domain (solid line), and subanalysis areas (smaller polygons), i.e., STW (west), STW (east), SAW (west), and SAW (east), together with the geographical locations referred to in the text.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Orography of the analysis area. The topography is dark gray for areas between 0 and 500 m in elevation, gray for elevations between 500 and 1000 m, light gray for elevations between 1000 and 1500 m, and white for all areas above 1500 m. Over the adjacent sea areas, the topography is indicated by the annual mean SST and resultant isotherm analysis.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Orography of the analysis area. The topography is dark gray for areas between 0 and 500 m in elevation, gray for elevations between 500 and 1000 m, light gray for elevations between 1000 and 1500 m, and white for all areas above 1500 m. Over the adjacent sea areas, the topography is indicated by the annual mean SST and resultant isotherm analysis.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Orography of the analysis area. The topography is dark gray for areas between 0 and 500 m in elevation, gray for elevations between 500 and 1000 m, light gray for elevations between 1000 and 1500 m, and white for all areas above 1500 m. Over the adjacent sea areas, the topography is indicated by the annual mean SST and resultant isotherm analysis.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Example SRTex classification. (a) Channel R2 (0.9 μm) image for 14 Jun 1995 at 0245 UTC, and (b) SRTex classification multiplexed with the posterior probability. The scale multiplexes a cloud index in the tens column and the posterior probability (range 0–9 corresponding to 0%–99.9%) in the units, where 0–9 are missing data (Ms), 10–19 are unclassifiable data (Unc), 20–29 are no-cloud sea (NC), etc. The analysis domain is a Lambert conformal grid of 960 × 912 1-km-resolution pixels.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Example SRTex classification. (a) Channel R2 (0.9 μm) image for 14 Jun 1995 at 0245 UTC, and (b) SRTex classification multiplexed with the posterior probability. The scale multiplexes a cloud index in the tens column and the posterior probability (range 0–9 corresponding to 0%–99.9%) in the units, where 0–9 are missing data (Ms), 10–19 are unclassifiable data (Unc), 20–29 are no-cloud sea (NC), etc. The analysis domain is a Lambert conformal grid of 960 × 912 1-km-resolution pixels.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Example SRTex classification. (a) Channel R2 (0.9 μm) image for 14 Jun 1995 at 0245 UTC, and (b) SRTex classification multiplexed with the posterior probability. The scale multiplexes a cloud index in the tens column and the posterior probability (range 0–9 corresponding to 0%–99.9%) in the units, where 0–9 are missing data (Ms), 10–19 are unclassifiable data (Unc), 20–29 are no-cloud sea (NC), etc. The analysis domain is a Lambert conformal grid of 960 × 912 1-km-resolution pixels.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Fig. 3.(Continued)
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Fig. 3.(Continued)
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Fig. 3.(Continued)
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
SRTex estimate of mean no-cloud frequency derived from NOAA-14 daylight passes in 1995. Data over land above 750 m have been masked (gray). An isotherm analysis of mean (cotemporal) SST is also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
SRTex estimate of mean no-cloud frequency derived from NOAA-14 daylight passes in 1995. Data over land above 750 m have been masked (gray). An isotherm analysis of mean (cotemporal) SST is also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
SRTex estimate of mean no-cloud frequency derived from NOAA-14 daylight passes in 1995. Data over land above 750 m have been masked (gray). An isotherm analysis of mean (cotemporal) SST is also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Scatterplot of ground and satellite observed annual (1995) cloud cover for Nelson (N), Blenheim (B), Christchurch (C), Hokitika (H), and Invercargill (I). The line of perfect fit is dashed, and the solid line is that of the best fit to the data. Tukey box plots of the SRTex and surface estimates are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Scatterplot of ground and satellite observed annual (1995) cloud cover for Nelson (N), Blenheim (B), Christchurch (C), Hokitika (H), and Invercargill (I). The line of perfect fit is dashed, and the solid line is that of the best fit to the data. Tukey box plots of the SRTex and surface estimates are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Scatterplot of ground and satellite observed annual (1995) cloud cover for Nelson (N), Blenheim (B), Christchurch (C), Hokitika (H), and Invercargill (I). The line of perfect fit is dashed, and the solid line is that of the best fit to the data. Tukey box plots of the SRTex and surface estimates are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
SRTex estimates of mean no-cloud frequency (on 16 km × 16 km tiles) derived from NOAA-14 daylight passes for (a) Jan, Feb, Mar (JFM); (b) Apr, May, Jun (AMJ); (c) Jul, Aug, Sep (JAS); and (d) Oct, Nov, Dec (OND). Land above 750 m has been blanked (white). Isotherm analyses of mean SSTs for the same periods are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
SRTex estimates of mean no-cloud frequency (on 16 km × 16 km tiles) derived from NOAA-14 daylight passes for (a) Jan, Feb, Mar (JFM); (b) Apr, May, Jun (AMJ); (c) Jul, Aug, Sep (JAS); and (d) Oct, Nov, Dec (OND). Land above 750 m has been blanked (white). Isotherm analyses of mean SSTs for the same periods are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
SRTex estimates of mean no-cloud frequency (on 16 km × 16 km tiles) derived from NOAA-14 daylight passes for (a) Jan, Feb, Mar (JFM); (b) Apr, May, Jun (AMJ); (c) Jul, Aug, Sep (JAS); and (d) Oct, Nov, Dec (OND). Land above 750 m has been blanked (white). Isotherm analyses of mean SSTs for the same periods are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Scatterplot of fraction of bright sun possible and satellite observed seasonal (i.e., JFM, AMJ, JAS, and OND 1995) cloud cover for Nelson (N), Blenheim (B), Christchurch (C), Hokitika (H), and Invercargill (I). Tukey box plots of the SRTex cloud cover and percentage of possible bright sunshine are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Scatterplot of fraction of bright sun possible and satellite observed seasonal (i.e., JFM, AMJ, JAS, and OND 1995) cloud cover for Nelson (N), Blenheim (B), Christchurch (C), Hokitika (H), and Invercargill (I). Tukey box plots of the SRTex cloud cover and percentage of possible bright sunshine are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Scatterplot of fraction of bright sun possible and satellite observed seasonal (i.e., JFM, AMJ, JAS, and OND 1995) cloud cover for Nelson (N), Blenheim (B), Christchurch (C), Hokitika (H), and Invercargill (I). Tukey box plots of the SRTex cloud cover and percentage of possible bright sunshine are also shown.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Cloud-type frequencies for 1995 using the conditional frequency approach (see text) for (a) transmissive cirrus (Ci), (b) thick cirrus (Cs), (c) cumulus (Cu), and (d) stratocumulus (Sc). For (c) and (d), the mean SST (for 1995) is displayed, and terrain above 500 m is masked. Areas where no analysis is possible are colored white.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Cloud-type frequencies for 1995 using the conditional frequency approach (see text) for (a) transmissive cirrus (Ci), (b) thick cirrus (Cs), (c) cumulus (Cu), and (d) stratocumulus (Sc). For (c) and (d), the mean SST (for 1995) is displayed, and terrain above 500 m is masked. Areas where no analysis is possible are colored white.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Cloud-type frequencies for 1995 using the conditional frequency approach (see text) for (a) transmissive cirrus (Ci), (b) thick cirrus (Cs), (c) cumulus (Cu), and (d) stratocumulus (Sc). For (c) and (d), the mean SST (for 1995) is displayed, and terrain above 500 m is masked. Areas where no analysis is possible are colored white.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Seasonal distribution of stratocumulus during 1995 for (a) JFM, (b) AMJ, (c) JAS, and (d) OND. The associated (CCSAD) SSTs are indicated via the isotherm analyses. Masked terrain is shown in white.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Seasonal distribution of stratocumulus during 1995 for (a) JFM, (b) AMJ, (c) JAS, and (d) OND. The associated (CCSAD) SSTs are indicated via the isotherm analyses. Masked terrain is shown in white.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Seasonal distribution of stratocumulus during 1995 for (a) JFM, (b) AMJ, (c) JAS, and (d) OND. The associated (CCSAD) SSTs are indicated via the isotherm analyses. Masked terrain is shown in white.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Seasonal distribution of cirrus during 1995 for (a) JFM, (b) AMJ, (c) JAS, and (d) OND.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Seasonal distribution of cirrus during 1995 for (a) JFM, (b) AMJ, (c) JAS, and (d) OND.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Seasonal distribution of cirrus during 1995 for (a) JFM, (b) AMJ, (c) JAS, and (d) OND.
Citation: Journal of Applied Meteorology 40, 1; 10.1175/1520-0450(2001)040<0016:AHRAOC>2.0.CO;2
Discriminant function contingency table for daytime measurements and feature vector {R2, entropy of R2, [(R2/R1) − 1], T4, (T4 − T5), entropy of T4, and scan angle}. Sample size is 1523.
Annual cloud statistics for 1995, four cloud types (stratocumulus, cumulus, transmissive cirrus, and thick cirrus), and four ocean areas [STW (west), SAW (west), STW (east), and SAW (east), as defined in Fig. 1]. The conditional frequency for each cloud type is given by the row labeled “Freq.,” the mean 11-μm temperature by T4, and the mean (SST − T4) difference by ΔT. The standard deviations follow each quantity.
