Cloud Detection Algorithm Research Articles

Validation of automated cloud top phase algorithms: distinguishing between cirrus clouds and snow in <italic>a priori</italic> analyses of AVHRR imagery

Quantitative assessments on the performance of automated cloud analysis algorithms require the creation of highly accurate, manual cloud, no cloud (CNC) images from multispectral meteorological satellite data. In general, the methodology to create these a priori analyses for the evaluation of cloud detection algorithms is relatively straightforward, although the task becomes more complicated when little spectral signature is evident between a cloud and its background, as appears to be the case in advanced very high resolution radiometer (AVHRR) imagery when thin cirrus is present over snow-covered surfaces. In addition, complex procedures are needed to help the analyst distinguish between water and ice cloud tops to construct the manual cloud top phase analyses and to ensure that inaccuracies in automated cloud detection are not propagated into the results of the cloud classification algorithm. Procedures are described that enhance the researcher’s ability to (1) distinguish between thin cirrus clouds and snow-covered surfaces in daytime AVHRR imagery, (2) construct accurate a priori cloud top phase manual analyses, and (3) quantitatively validate the performance of both automated cloud detection and cloud top phase classification algorithms. The methodology uses all AVHRR spectral bands, including a band derived from the daytime 3.7-?m channel, which has proven most valuable for discriminating between thin cirrus clouds and snow. It is concluded that while the 1.6-?m band is needed to distinguish between snow and water clouds in daytime data, the 3.7-?m channel remains essential during the daytime to differentiate between thin ice clouds and snow. Unfortunately this capability that may be lost if the 3.7-?m data switches to a nighttimeonly transmission with the launch of future National Oceanographic and Atmospheric Administration (NOAA) satellites.

Cloud detection from thermal infrared images using a segmentation technique

In this paper, a fast and accurate method is proposed for two cloud detection tests on thermal infrared (IR) data obtained from satellite images over sea; an IR gross cloud test and a spatial coherence test. The proposed method is based on a regional segmentation technique. After the segmentation of an IR image, small regions were regarded as cloudy due to their high spatial variability in temperature. This technique preserved the spatial resolution of the detected cloud image which would be degraded by the conventional spatial coherence test. It also reduced the computation dramatically, compared to the conventional spatial coherence test. An accurate temperature threshold between clear sea and clouds was determined directly from the segmented image. This post-determined threshold was found to be more accurate than pre-determined temperature thresholds. Since this algorithm does not require any human interaction, it can be combined with other tests in an automatic cloud detection algorithm.

Highers—The AVHRR-based higher spatial resolution sea surface temperature data set intended for studying the ocean south of Japan

For the study of the ocean south of Japan, the HIGHER spatial resolution Sea surface temperature (HIGHERS) data have been produced from the AVHRR/NOAA data. The grid size is 0.075 degree and is much smaller than that of the usual sea surface temperature (SST) analysis field, which is gridded at most in 1 degree. In order to calculate accurate SST, the split-window Multi-Channel SST method is applied to cloudless pixels, which are detected with a cloud detection algorithm. The accuracy of the HIGHERS data is checked in comparison with the in-situ observation data.

Cloud detection from a sequence of SST images

A cloud detection algorithm was designed as an adjunct to a companion edge-detection algorithm. The cloud detection integrates two distinct algorithms: one based on multiimage processing, the other on single-image analysis. The multiimage portion of the cloud detection algorithm operates on a time sequence of sea surface temperature (SST) images. It is designed to detect clouds associated with regions of apparently lower temperatures than the underlying SST field. A pixel in the current image is initially considered to be corrupted by clouds if it is significantly cooler than the corresponding pixel in a neighbor image. To refine the initial classification, the algorithm checks the current image and the neighbor image for the presence of water masses, which through displacement could explain the change in temperature. The single-image cloud detection algorithm is designed to detect clouds associated with regions of the SST image where gradient vectors have a large magnitude. These regions are flagged in the map of potential clouds. multiimage processing is integrated with the single-image algorithm by adding pixels classified as cloudy at the multiimage level to the map of potential clouds. Further analysis of the gradient vector field and of the shapes of potentially cloudy areas allows one to determine whether these regions correspond to clouds or SST fronts. A previous study has shown that the clouds identified by the single-image algorithm were in close agreement with those detected by a human expert. To validate the additional multiimage processing, the effect of the integrated cloud detection on the performance of a companion edge detection algorithm is examined. These results and a direct comparison with the cloud masks produced by a human expert indicate that, compared to the single-image algorithm, the multiimage algorithm successfully identify additional cloud-corrupted regions while keeping a low rate for the detection of false clouds.

Improved cloud detection for daytime AVHRR scenes over land

Accurate cloud detection in Advanced Very High Resolution Radiometer (AVHRR) data over land is a difficult task complicated by spatially and temporally varying land surface reflectances and emissivities. The AVHRR Split-and-Merge Clustering (ASMC) algorithm for cloud detection in A VHRR scenes over land provides a computationally efficient, scene-specific, objective way to circumvent these difficulties. The algorithm consists of two steps: 1) a split-and-merge clustering of the input data (calibrated channel 2 albedo, calibrated channel 4 temperature, and a channel 3- channel 4 temperature difference), which segments the scene into its natural groupings; and 2) a cluster-labelling procedure that uses scene-specific, joint three-dimensional adaptive labelling thresholds (as opposed to constant static thresholds) to label the clusters as either cloud, cloud free land, or uncertain. The uncertain class is used for those pixels whose signature is not clearly cloud free land or cloud (e.g., pixels at cloud boundaries that often contain subpixel cloud and land information that has been averaged together by the integrating aperture function of the AVHRR instrument). Results show that the ASMC algorithm is neither regionally nor temporally specific and can be used over a large range of solar altitudes. Sensitivity of the segmentation and labelling steps to the choice of input variables also was studied. Results obtained with the ASMC algorithm also compare favorably with those obtained from a wide range of currently used algorithms to detect cloud over land in A VHRR data. Moreover, the ASMC algorithm can be adapted for use with data to be taken by the Moderate Resolution Imaging Spectrometer-Nadir (MODIS-N).

A correlation based cloud-detection algorithm combined with an examination of some split-window assumptions

Abstract Twenty-six daytime NOAA-11 AVHRR images covering the Danish waters are analysed together with 99 cloud-free bulk temperature measurements. The images are cloud screened with a cloud-detection algorithm, which in excess of the usual threshold and variation tests consists of: a min_max-routine which among suspicious pixels points out the cloud contaminated pixels, and a separation of cloud pixels from cold water pixels by means of the correlation coefficient between the channel 4 brightness temperature and the channel 4-channel 5 brightness temperature. In the cloud screened images, the channel 4-channel 5 brightness temperatures do not increase with the satellite zenith angle and do not seem to carry any significant information about the true sea surface temperature (SST). Local meteorological and oceanographic effects are believed to be more important than the channel 4-channel 5 brightness temperatures.

A simple cloud masking approach using NOAA AVHRR daytime data for tropical areas

Abstract A simple method for cloud detection for AVHRR daytime data is presented and checked for equatorial/tropical areas based on a study area in northeastern Brazil. Five different cloud masking techniques based on visible and infrared spectral information for cloud detection are calibrated. The significant differences between the equatorial threshold obtained in this work and the midlatitude thresholds given by Saunders and Kriebel and by Thiermann and Ruprecht are compared and discussed. Results from the cloud masking algorithm are presented and comments are made about the problems related to the automatic cloud detection algorithm presented in this study.

Improved cloud detection in GOES scenes over the oceans

Accurate cloud detection in GOES data over the ocean is a difficult task complicated by poor spatial resolution (4 km) in the GOES IR data, relatively coarse quantization (6 bits) for GOES VIS data, a visible sensing region of the spectrum not ideally suited for cloud versus ocean segmentation, and relative small oceanic signal dynamic range compared to that of either cloud or land structures found in a typical GOES scene. The GOES Adapted LDTNLR Ocean Cloud Mask (GALOCM) algorithm for cloud detection in GOES scenes over the oceans provides a computationally efficient, scene-specific way to circumvent these difficulties. The algorithm consists of four steps: 1) generate a cloud mark using the Local Dynamic Threshold Non-Linear Rayleigh (LDTNLR) algorithm of Simpson and Humphrey (1990); 2) generate a second cloud mask using an adaptive threshold: 3) divide the pixels in the scene into three groups (both methods agree that pixel is ocean, pixel is cloud, or the pixel is in contention); and 4) iteratively apply an adaptive threshold to the contested pixels. Convergence occurs when pixels are no longer in contention based on statistical criteria. Results show that the GALOCM method produces accurate cloud masks over the oceans which are neither regionally dependent nor temporally specific. GOES scenes containing ocean, cloud, and land are best cloud screened using a combination of the GOES Split-and-Merge Clustering (Simpson and Gobat, 1995) and the GALOCM algorithms.

Improved cloud detection in GOES scenes over land

Accurate cloud detection in satellite data over land is a difficult task complicated by spatially and temporally varying land surface reflectivities and emissivities. The GOES split-and-merge clustering (GSMC) algorithm for cloud detection in GOES scenes over land provides a computationally efficient, scene specific way to circumvent these difficulties. The algorithm consists of three steps: 1) a split-and-merge clustering of the input data which segments the scene into its natural grouping; 2) a cluster labeling procedure which uses scene specific adaptive thresholds (as opposed to constant static thresholds) to label the clusters as either cloud or cloud-free land; and 3) a post-processing step which imposes a degree of spatial uniformity on the labeled land and cloud pixels. An “a priori” mask feature also enhances cloud detection in traditionally difficult scenes (e.g., clouds over bright desert). Results show that the GSMC algorithm is neither regionally nor temporally specific and can be used over a large range of solar altitudes.

Analysis of cloud top height and cloud coverage from satellites using the O2A and B bands

Cloud height and cloud coverage detection are important for total ozone retrieval using ultraviolet and visible scattered light. Use of the O2 A and B bands, around 761 and 687 nm, by a satellite‐borne instrument of moderately high spectral resolution viewing in the nadir makes it possible to detect cloud top height and related parameters, including fractional coverage. The measured values of a satellite‐borne spectrometer are convolutions of the instrument slit function and the atmospheric transmittance between cloud top and satellite. Studies here determine the optical depth between a satellite orbit and the Earth or cloud top height to high accuracy using FASCODE 3. Cloud top height and a cloud coverage parameter are determined by least squares fitting to calculated radiance ratios in the oxygen bands. A grid search method is used to search the parameter space of cloud top height and the coverage parameter to minimize an appropriate sum of squares of deviations. For this search, nonlinearity of the atmospheric transmittance (i.e., leverage based on varying amounts of saturation in the absorption spectrum) is important for distinguishing between cloud top height and fractional coverage. Using the above‐mentioned method, an operational cloud detection algorithm which uses minimal computation time can be implemented.

A new cloud detection algorithm for nighttime AVHRR/HRPT data

A new cloud detection algorithm for nighttime Advanced Very High Resolution Radiometer (AVHRR) data has been developed and applied to a large number of images from various locations around Japan. The algorithm is characterized by a recovery function and the use of a two-dimensional histogram. Results obtained after applying the algorithm are presented and compared with those of previous algorithms. The comparison reveals that the new algorithm appears to be more successful than the previous algorithms.

Cloud screening in AVHRR digital data over Arctic regions

A cloud screening method initially generated to mask cloud contaminated pixels over the ocean in visible/infrared imagery, has been revised and adapted to detect clouds over Arctic regions with encouraging results. Although the method is quite successful in eliminating very cold clouds, it underestimates low level clouds. However, this does not appear to interfere with monitoring of ice related features such as leads or the ice edge in Advanced Very High Resolution Radiometer (AVHRR) scenes. The method uses: a multiple-band approach to produce signatures not readily available in single channel data, an edge detection/dilation technique to locate features in the clouds and to join isolated edges, and a polygon identification technique to remove noise in the form of isolated pixels and separate clear regions from cloud contaminated areas. The method has been tested over a limited set of data with consistent results. Initial evaluation of the usefulness of this cloud-detection algorithm in data-fusion experiments indicate a potential in locating areas in AVHRR data which are cloud contaminated and which could yield a far superior representation of the ice features if replaced with data from a different sensor such as the Special Sensor Microwave/lmager (SSM/I).

Cloud screening in AVHRR digital data over Arctic regions

A cloud screening method initially generated to mask cloud contaminated pixels over the ocean in visible/infrared imagery, has been revised and adapted to detect clouds over Arctic regions with encouraging results. Although the method is quite successful in eliminating very cold clouds, it underestimates low level clouds. However, this does not appear to interfere with monitoring of ice related features such as leads or the ice edge in Advanced Very High Resolution Radiometer (AVHRR) scenes. The method uses: a multiple-band approach to produce signatures not readily available in single channel data, an edge detection/dilation technique to locate features in the clouds and to join isolated edges, and a polygon identification technique to remove noise in the form of isolated pixels and separate clear regions from cloud contaminated areas. The method has been tested over a limited set of data with consistent results. Initial evaluation of the usefulness of this cloud-detection algorithm in data-fusion experiments indicate a potential in locating areas in AVHRR data which are cloud contaminated and which could yield a far superior representation of the ice features if replaced with data from a different sensor such as the Special Sensor Microwave/lmager (SSM/I).

Cloud cover analysis with Arctic AVHRR data: 1. Cloud detection

Automated analyses of satellite radiance data have concentrated heavily on low and middle latitude situations. Some of the design objectives for the International Satellite Cloud Climatology Project (ISCCP) cloud detection procedure such as space and time contrasts are used in a basic algorithm from which a polar cloud detection algorithm is developed. This algorithm is applied to Arctic data for January and July conditions. Both advanced very high resolution radiometer (AVHRR) and scanning multichannel microwave radiometer (SMMR) data are utilized. Synthetic AVHRR and SMMR data for a 7‐day analysis period are also generated to provide a data set with known characteristics on which to test and validate algorithms. Modifications to the basic algorithm for polar conditions include the use of SMMR and SMMR‐derived data sets for the estimation of surface parameters, elimination of the spatial test for the warmest pixel, the use of AVHRR channels 1 (0.7 μm), 3 (3.7 μm), and 4 (11 μm) in the temporal tests and the final multispectral thresholding, and the use of surface class characteristic values when clear‐sky values cannot be obtained. Additionally, the difference between channels 3 and 4 is included in the temporal test for the detection of optically thin cloud. Greatest improvement in computed cloud fraction is realized over snow and ice surfaces; over open water or snow‐free land, all versions perform similarly. Since the inclusion of SMMR for surface analysis and additional spectral channels increases the computational burden, its use may be justified only over snow and ice‐covered regions.