Cloud Screening Algorithm Research Articles

An Algorithm to Screen Cloud-A?ected Data for Sky Radiometer Data Analysis

Aerosol optical parameters obtained from sky radiometer instrument are important not only for studying aerosol effects on climate change, but also for validating several results obtained from satellite retrievals and numerical simulations. However, the greatest challenge is to separate cloud-affected and cloud free data from data measured by sky radiometer. In this study, we present an algorithm to separate such cloud-affected and cloud free data. The proposed algorithm is comprehensively tested with observational data. The algorithm consists of three tests: (i) test with global irradiance data, (ii) spectral variability test, and (iii) statistical analyses test. Though the test with the global irradiance data is the most powerful test, our study shows that it has some limitations, which can sometimes cause some clear sky data to be detected as cloud-affected data. In order to cope with this problem, a modified version of spectral variability algorithm is proposed. As the second test, the modified spectral variability algorithm is applied to filter clear sky data from data detected as cloud-affected by the first test. Finally, statistical analyses tests are performed to remove any outlier, if exists, from clear sky data detected by the first and second tests. It is shown that our proposed algorithm can screen cloud-affected data more effectively in comparison to other cloud screening algorithms. An application of this algorithm to screen observation data of one year collected in Chiba, Japan produces the seasonal means of optical thickness at 500 nm (Angstrom exponent) as ∼0.17(∼1.42), ∼0.38(∼0.98), ∼0.53(∼1.21), and ∼0.21(∼1.28) for winter, spring, summer and autumn seasons, respectively. Depending on the season, the initial seasonal mean optical thicknesses at 500 nm decrease by ∼0.07 to ∼0.16 and mean Angstrom exponents increase by ∼0.087 to ∼0.162 due to cloud screening. An application of this algorithm to dust-loaded atmospheres is also discussed. The proposed algorithm can be applied to any sky radiometer observation site as long as global irradiance data are available.

LAI, fAPAR and fCover CYCLOPES global products derived from VEGETATION: Part 1: Principles of the algorithm

This article describes the algorithmic principles used to generate LAI, fAPAR and fCover estimates from VEGETATION observations. These biophysical variables are produced globally at 10 days temporal sampling interval under lat–lon projection at 1/112° spatial resolution. After a brief description of the VEGETATION sensors, radiometric calibration process, based on vicarious desertic targets is first presented. The cloud screening algorithm was then fine tuned using a global network of cloudiness observations. Atmospheric correction is then achieved using the SMAC code with inputs coming from meteorological values of pressure, ozone and water vapour. Aerosol optical thickness is derived from MODIS climatology assuming continental aerosol type. The Roujean BRDF model is then adjusted for red, near infrared and short wave infrared bands used to the remaining cloud free observations collected over a time window of ± 15 days. Outliers due to possible cloud contamination or residual atmospheric correction are iteratively eliminated and prior information is used to get more robust estimates of the three BRDF kernel coefficients. Nadir viewing top of canopy reflectance in the three bands is input to the biophysical algorithm to compute the products at 10 days sampling interval. This algorithm is based on training neural networks over SAIL + PROPSPECT radiative transfer model simulations for each biophysical variable. Details on the way the training data base was generated and the neural network designed and calibrated are presented. Finally, theoretical performances are discussed. Validation over ground measurement data sets and inter-comparison with other similar biophysical products are presented and discussed in a companion paper. The CYCLOPES products and associated detailed documentation are available at http://postel.mediasfrance.org.

Consistency of global MODIS aerosol optical depths over ocean on Terra and Aqua CERES SSF data sets

Aerosol retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard Terra and Aqua platforms are available from the Clouds and the Earth's Radiant Energy System (CERES) Single Scanner Footprint (SSF) data sets. Over ocean, two aerosol products are reported side by side. The primary M product is generated by subsetting and remapping the multispectral (from 0.47 to 2.1 μm) MOD04/MYD04 oceanic aerosol data onto CERES footprints. M*D04 processing uses cloud screening and aerosol algorithms developed by the MODIS science team. The secondary AVHRR‐like A product is generated in only two MODIS bands. The A processing uses the CERES cloud‐screening algorithm and NOAA/NESDIS glint identification and single‐channel aerosol retrieval algorithms. The M and A products have been documented elsewhere and preliminarily compared using 2 weeks of global Terra CERES SSF edition 1A data, in which the M product was based on MOD04 collection 3. In this study, the comparisons between the M and A aerosol optical depths (AOD) in MODIS band 1 (0.64 μm), τ1M and τ1A, are reexamined using 9 days of global CERES SSF Terra edition 2A and Aqua edition 1B data from 13 to 21 October 2002 and extended to include cross‐platform comparisons. The M and A products on the new CERES SSF release are generated using the same aerosol algorithms as before but with different preprocessing and sampling procedures, thus lending themselves to a simple sensitivity check to nonaerosol factors. Both τ1M and τ1A generally compare well across platforms. However, the M product shows larger differences, which increase with ambient cloud amount and toward the solar side of the orbit. The cross‐platform, cross‐product, and cross‐release comparisons conducted in this study confirm an earlier observation that the major area for improvement in the current aerosol processing lies in a more formalized and standardized sampling (most importantly, cloud screening), whereas optimization of the aerosol algorithm is deemed to be an important yet less critical element.

Evaluating the quality of AVHRR Pathfinder NDVI data for the African continent using SPOT-4 Vegetation data

Geografisk Tidsskrift, Danish Journal of Geography 106(1):87–102, 2006 The AVHRR (Advanced Very High Resolution Radiometer) 8 km resolution Pathfinder NDVI (Normalized Difference vegetation Index) 10-day composite data set has been used for numerous local and global scale vegetation time series studies during recent decades. The dataset is a result of considerable processing and resampling from the original 1 km resolution AVHRR LAC (Local Area Coverage) data and combined with inappropriate sensor band design for vegetation monitoring these factors potentially introduce noise in the Pathfinder NDVI. The quality of the Pathfinder data has been difficult to assess but the recent release of SPOT-4 Vegetation (VGT) 10-day composite (S10) NDVI data is considered to be an improvement on the AVHRR data and offers the first possibility for a correct analysis/evaluation of the Pathfinder data.—In this study, three years of AVHRR Pathfinder and resampled SPOT-4 VGT (1998–2000) data has been evaluated/analysed using an ortho-regression analysis on 10-day, monthly and annual integrated values. Annual integrated values of AVHRR Pathfinder and SPOT-4 VGT data can be characterized by a linear relationship and correlate well on a continental scale. However the dynamic range of the SPOT-4 VGT NDVI is higher than the AVHRR Pathfinder NDVI. Analysis of individual IGBP biomes shows biome specific differences; evergreen broadleaf forest characterized by lower r values and highest RMSE values and normalized RMSE values indicate largest divergence for the shrubland biome. When performing the analysis on monthly maximum composites and 10-day composite data, intra-annual variations appear. Analyses of 10-day data also show a linear relationship between the NDVI of the two composite products; the relationship being characterized by lower r values and higher RMSE's. Monthly and 10-day maximum composites reveal intra-annual variations in the correlation between the SPOT-4 VGT and AVHRR Pathfinder data. This is attributed to different cloud masking algorithms; the SPOT-4 VGT cloud screening algorithm being insufficient thereby suppressing the rainy season ND VI. However, the correspondence between the SPOT-4 VGT and A VHRR PAL data is considerably better for a transect analysis covering an area not severely influenced by clouds when compared to the continental scale analysis. This suggests that the major part of the noise found in the continental scale SPOT-4 VGT NDVI and AVHRR PAL NDVI analyses can be attributed the different cloud masking algorithms and is to a lesser degree a consequence of the AVHRR PAL NDVI resampling. It is concluded that the quality of the AVHRR NDVI PAL is adequate for time series continental scale vegetation analysis despite the resampling scheme.

The use of remote sensing for estimating ET of irrigated wheat and cotton in Northwest Mexico

Components of a satellite-based system for estimating the crop water requirements of irrigated vegetation have been combined, applied, and tested against field data in the Yaqui Valley, northwest Mexico. Frequent satellite observations have the potential to provide snap shots of cloud variability at the high spatial and temporal resolutions that are needed for making simple, near real-time estimates of incoming solar radiation and, thus, daytime evaporation required for irrigation scheduling. Less frequent polar orbiting satellites offer the capacity of following the vegetation development at higher spatial resolution. The operational framework for obtaining cloud cover has been developed and applied using hourly sampled, 1 km resolution, GOES-10 data received in real-time. The high-resolution, cloud-screening algorithm has proved to be efficient and reliable and has been used to provide high-resolution (4 km) estimates of solar radiation. Relationships between vegetation indices (NDVI and SAVI) and crop coefficients (the ratio of measured to reference evapotranspiration) have been derived with four different models (Shuttleworth, Penman, Priestley–Taylor and Makkink), using ground-based surface reflectance measured over the crop. Continuous measurements of surface fluxes and other meteorological variables were made following almost the entire vegetative cycle of the plant using a station equipped with standard meteorological instruments and an eddy-correlation system. Actual evapotranspiration was computed as the product of the estimated crop coefficients, derived from field radiometer measurements, and reference evapotranspiration. In comparison with ground data, RMSE values are on the order of 1 mm per day. Finally the opportunity to use high-resolution satellite data to make near real-time estimates of crop evaporation is discussed.

Analysis of along track scanning radiometer-2 (ATSR-2) data for clouds, glint and sea surface temperature using neural networks

Improved oceanic cloud climatologies and more accurate sea surface temperature (SST) from satellite data (e.g., the Along Track Scanning Radiometer-2 (ATSR-2)) depend upon the identification, and in the case of SST, masking of cloud from the data. Few cloud-screening algorithms for ATSR-2 data, however, have been published. This is unfortunate because, unlike the original ATSR (hereafter referred to as ATSR-1), ATSR-2 has three bands of visible data in addition to its traditional suite of near-to-far thermal infrared bands. A new neural network-based cloud detection algorithm, which accommodates both the nadir and forward views of the ATSR-2, is presented. It evaluates every pixel in the scene, is statistically reproducible, computationally efficient, and requires no knowledge of cloud type. Moreover, the algorithm accurately detects glint radiance, which is often seen in at least one of the 1.6 μm (and/or visible) views, subpixel cloud near cloud boundaries, low-lying marine stratiform cloud, and does not misinterpret actual ocean thermal gradients as cloud. These latter issues have interfered with many SST retrievals in the past. Dual view/dual channel SST retrievals were derived and validated against buoy data (1 m depth). RMS error between retrieved SST and the buoy data is typically about 0.3 K in regions of minimum SST gradient. Pre-processing of the data is done prior to analysis because some artifacts not seen in ATSR-1 data can seriously compromise both cloud detection and SST retrieval. Detailed comparison between the ATSR operational cloud detection scheme and that developed herein, based upon 1077 scenes, shows that the operational product consistently misinterprets regions of cold but cloud-free oceanic thermal gradient as cloud, leading to a sampling-related warm bias in many of the operational SST products. The basis for this misinterpretation has been identified and quantified. Thus, the operational cloud detection scheme may have to be revised if a fully representative global ocean SST dataset is to be obtained from ATSR data.

Mapping annual mean ground‐level PM2.5 concentrations using Multiangle Imaging Spectroradiometer aerosol optical thickness over the contiguous United States

We present a simple approach to estimating ground‐level fine particulate matter (PM2.5, particles smaller than 2.5 μm in diameter) concentrations by applying local scaling factors from a global atmospheric chemistry model (GEOS‐CHEM with GOCART dust and sea salt data) to aerosol optical thickness (AOT) retrieved by the Multiangle Imaging Spectroradiometer (MISR). The resulting MISR PM2.5 concentrations are compared with measurements from the U.S. Environmental Protection Agency's (EPA) PM2.5 compliance network for the year 2001. Regression analyses show that the annual mean MISR PM2.5 concentration is strongly correlated with EPA PM2.5 concentration (correlation coefficient r = 0.81), with an estimated slope of 1.00 and an insignificant intercept, when three potential outliers from Southern California are excluded. The MISR PM2.5 concentrations have a root mean square error (RMSE) of 2.20 μg/m3, which corresponds to a relative error (RMSE over mean EPA PM2.5 concentration) of approximately 20%. Using simulated aerosol vertical profiles generated by the global models helps to reduce the uncertainty in estimated PM2.5 concentrations due to the changing correlation between lower and upper tropospheric aerosols and therefore to improve the capability of MISR AOT in estimating surface‐level PM2.5 concentrations. The estimated seasonal mean PM2.5 concentrations exhibited substantial uncertainty, particularly in the west. With improved MISR cloud screening algorithms and the dust simulation of global models, as well as a higher model spatial resolution, we expect that this approach will be able to make reliable estimation of seasonal average surface‐level PM2.5 concentration at higher temporal and spatial resolution.

Automated cloud screening algorithm for MFRSR data

A new automated cloud screening algorithm for ground‐based sun‐photometric measurements is described and illustrated on examples of real (MFRSR) and simulated data. The algorithm uses single channel direct beam measurements and is based on variability analysis of retrieved optical thickness. To quantify this variability the inhomogeneity parameter ε is used. This parameter is commonly used for cloud remote sensing and modeling, but not for cloud screening. In addition to this an adjustable enveloping technique is applied to control strictness of the selection method. The key advantages of this technique are its objectivity, computational efficiency and the ability to detect short clear‐sky intervals under broken cloud cover conditions. Moreover, it does not require any knowledge of the instrument calibration. The performance of the method has been compared with that of AERONET cloud screening algorithm.

Short-Term Performance of MM5 with Cloud-Cover Assimilation from Satellite Observations

Abstract This study investigates the extent to which assimilating high-resolution remotely sensed cloud cover into the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5) provides an improved regional diagnosis of downward shortwave surface radiation fluxes and precipitation and enhances the model's ability to make short-range prediction. The high-resolution (4 km × 4 km) clear- and cloudy-sky radiances derived using a cloud-screening algorithm from visible band Geostationary Operational Environmental Satellite (GOES) data were used in the University of Maryland Global Energy and Water Cycle Experiment's Surface Radiation Budget (UMD GEWEX/SRB) model to infer the vertically integrated cloud mass via cloud optical thickness. Three-dimensional cloud fields were created that took their horizontal distribution from the satellite image but derived their vertical distribution, in part, from the fields simulated by MM5 during the time step immediately prior to assimilation and, in part, from the observed cloud-top height derived from the infrared band of GOES. Linear interpolation was used to derive 1-min cloud images between 15-min GOES samples, and the resulting images were ingested every minute. Comparisons were made between modeled and observed data taken from the Arizona Meteorological Network (AZMET) in southern Arizona for model runs with and without cloud ingestion. Cloud ingestion substantially improved the ability of the MM5 model to capture temporal and spatial variations in surface fields associated with cloud cover. Experiments in which the model was operated in forecast mode suggest that cloud ingestion gave some limited enhancement in MM5 short-term prediction ability for up to 3 h. However, an analysis suggests that, in order to get additional forecasting capability, it will be necessary to modify the atmospheric dynamics and thermodynamics in the model to be consistent with the ingested cloud fields.

Regional aerosol retrieval results from MISR

Examples of aerosol retrieval results, derived from the Multi-angle Imaging SpectroRadiometer (MISR) on the Earth Observation Science (EOS) Terra platform, are shown and the performance of the retrieval algorithms are discussed, following the first 18 months of operational data processing. A number of algorithm modifications were implemented, based on an analysis of aerosol retrieval results during this period, and these changes are described. Two cloud-screening algorithms, the angle-to-angle smoothness and angle-to-angle correlation tests, which were used in the preprocessing phase of the analyses are also described. The aerosol retrieval examples cover a wide variety of conditions, both over land and water. Particular aerosol types include dust clouds, forest fire and volcanic plumes, and localized dense haze. Finally, some ideas are discussed for additional improvement of the MISR aerosol data product, based on the experience gained in analyzing multiangle data and the associated geophysical products.

Characterizing land surface anisotropy from AVHRR data at a global scale using high performance computing

We used the multi-temporal ten-day composite data from the Advanced Very High Resolution Radiometer (AVHRR) for the years 1983 to 1986 to retrieve the Bidirectional Ree ectance Distribution Function (BRDF) using high performance computing techniques. Three di Verent models are used: a simple linear model, a semi-empirical iterative model and a temporal model. The objectives of this study were to compare the performance of di Verent BRDF models at a global scale, assess the computational requirements and optimize the algorithm implementation using high performance computational techniques, and to determine ifthere is any coherent spatial structure in the coe Ycients ofdiVerent BRDF models corresponding to di Verent land cover types. The standard error between model computed ree ectances and the input data was used to quantify the performance ofthe models.Even though the iterative model is computationally more expensive (158 minutes) than either the simple linear model (15 minutes) or the temporal model (16 minutes), the results from all the three models were very similar when the BRDF was estimated at discrete time periods. If the BRDF models were applied without dividing the input data into discrete time intervals, then the temporal model gave better results than the other two. All the models were run on an IBM SP2 parallel machine with 16 CPUs. Most of the mountain- ous and snow covered areas in high latitudes had null values since the cloud screening algorithm used in the Pathe nder processing performed poorly in distin- guishing between snow and clouds. The BRDF coe Ycients of the iterative model and the Fourier coe Ycients of the temporal model showed a strong spatial structure corresponding to known variations in land cover.

Characterizing land surface anisotropy from AVHRR data at a global scale using high performance computing

We used the multi-temporal ten-day composite data from the Advanced Very High Resolution Radiometer (AVHRR) for the years 1983 to 1986 to retrieve the Bidirectional Reflectance Distribution Function (BRDF) using high performance computing techniques. Three different models are used: a simple linear model, a semi-empirical iterative model and a temporal model. The objectives of this study were to compare the performance of different BRDF models at a global scale, assess the computational requirements and optimize the algorithm implementation using high performance computational techniques, and to determine if there is any coherent spatial structure in the coefficients of different BRDF models corresponding to different land cover types. The standard error between model computed reflectances and the input data was used to quantify the performance of the models. Even though the iterative model is computationally more expensive (158 minutes) than either the simple linear model (15 minutes) or the temporal model (16 minutes), the results from all the three models were very similar when the BRDF was estimated at discrete time periods. If the BRDF models were applied without dividing the input data into discrete time intervals, then the temporal model gave better results than the other two. All the models were run on an IBM SP2 parallel machine with 16 CPUs. Most of the mountainous and snow covered areas in high latitudes had null values since the cloud screening algorithm used in the Pathfinder processing performed poorly in distinguishing between snow and clouds. The BRDF coefficients of the iterative model and the Fourier coefficients of the temporal model showed a strong spatial structure corresponding to known variations in land cover.

High-Resolution Daytime Cloud Observations for Northwestern Mexico fromGOES-7Satellite Observations

The first stage in a program of research to develop a regional model capable of describing the hydrology of semiarid areas of northwest Mexico and southwest United States, using remotely sensed data, is described in this paper. Finescale information on cloud cover is required to provide the radiation forcing for making simple, near-real-time estimates of daytime evaporation in hydrologic models, and frequent satellite observations have the potential to document cloud variability at high spatial and temporal resolutions. In this study, the operational framework for obtaining information on cloud cover was developed and applied, using hourly sampled, 1-km resolution GOES-7 data as received in real time in Obregon, Mexico. These satellite data were collected and analyzed from 1 July 1993 to 31 July 1994 for an approximately 106 km2 rectangular area in northwest Mexico. An efficient method was devised to provide clear-sky radiance images for the study area, at 4 km 3 4k m resolution, and updated at monthly intervals, by applying thresholds indexed to the locally appropriate clearsky radiance, thereby allowing for spatial and temporal changes in surface conditions. Manual image inspection and comparison with ground-based measurements of cloud cover and surface solar radiation provided reassurance that the high-resolution cloud-screening algorithm gave satisfactory results. This algorithm was applied to investigate the effects of temporal sampling frequency on estimates of daytimeaverage cloud cover and to document aspects of the cloud characteristics for the study area. The high-resolution algorithm proved to be efficient and reliable and bodes well for its future use in providing high-resolution estimates of surface solar radiation for use in a hydrologic model. Monthly clear-sky composite images were consistently generated, showing little evidence of contamination by persistent clouds, and tracked the seasonal evolution in surface radiance. Comparison with ground-based measurements gave confidence in the credibility of the satellite estimates and revealed weaknesses in the Campbell‐Stokes solarimeter. The seasonal evolution of spatial patterns of cloud and its diurnal cycle were investigated. The average cloudiness for the study area is 0.25, with a substantial annual variation from 0.19 in April to 0.40 in December. Persistent cloudy conditions throughout the year were detected over the Pacific Ocean west of Baja California. The derived high-resolution cloud estimates, when compared with similar estimates from the International Satellite Cloud Climatology Project (ISCCP D1), were about half those obtained with the low-resolution data, indicating that, in this complex study area where land and water boundaries are in close proximity, low-resolution satellite observations of clouds may not be able to depict the true cloud cover.

Cloud-Screening and Quality Control Algorithms for the AERONET Database

Automatic globally distributed networks for monitoring aerosol optical depth provide measurements of natural and anthropogenic aerosol loading, which is important in many local and regional studies as well as global change research investigations. The strength of such networks relies on imposing a standardization of measurement and processing, allowing multiyear and large-scale comparisons. The development of the Aerosol Robotic Network (AERONET) for systematic ground-based sunphotometer measurements of aerosol optical depth is an essential and evolving step in this process. The growing database requires the development of a consistent, reproducible, and system-wide cloud-screening procedure. This paper discusses the methodology and justification of the cloud-screening algorithm developed for the AERONET database. The procedure has been comprehensively tested on experimental data obtained in different geographical and optical conditions. These conditions include biomass burning events in Brazil and Zambia, hazy summer conditions in the Washington DC area, clean air advected from the Canadian Arctic, and variable cloudy conditions. For various sites our screening algorithm eliminates from ∼20% to 50% of the initial data depending on cloud conditions. Certain shortcomings of the proposed procedure are discussed.

Estimation of the aerosol perturbation to the Earth's Radiative Budget over oceans using POLDER satellite aerosol retrievals

POLDER satellite retrievals of aerosol properties over oceans are used to estimate a global‐mean clear‐sky aerosol shortwave flux perturbation of order −5 to −6 Wm−2. Uncertainties due to aerosol absorption and POLDER cloud screening algorithm are quantified. In order to bound the radiative forcing by anthropogenic aerosols, we attempt to remove the contribution of background aerosols from these estimates and present all‐sky aerosol radiative effects for three regions and two methods. The results are sensitive to the thresholds used to define the background conditions.

Cloud Detection from the Spaceborne POLDER Instrument and Validation against Surface Synoptic Observations

Abstract This paper describes the cloud screening algorithms that have been developed for the processing of Polarization and Directionality of the Earth Reflectances (POLDER) measurements over land surfaces. Four tests are applied to the measurements. The first one is a threshold on the 0.44-μm reflectance after atmospheric correction. The second one is similar but with a smaller threshold and is applied only over targets with significant spectral variation. The third one compares the surface pressure to an estimate derived from two POLDER channels centered on an oxygen absorption band. The fourth one makes use of POLDER polarization capabilities and seeks the presence of a rainbow generated by water clouds. The performance of the method is evaluated using a large dataset of nearly coincident POLDER measurements and surface observations of cloud cover. The validation dataset is fully independent of the spaceborne measurements and allows the sampling of a wide range of situations. The results demonstrate t...

Sea Surface Temperature Retrievals Optimized to the East Sea (Sea of Japan) using NOAA/AVHRR Data

The accuracy of sea surface temperatures derived by NOAA/NESDIS (National Oceanic and Atmospheric Administration/National Environmental Satellite, Data, and Information Service) equations was tested by comparison with temperatures measured by thirty-four satellite-tracked ARGOS drifters deployed in the East Sea (Sea of Japan) from 1993 to 1997. Using an improved cloud-screening algorithm for the East Sea, we obtained 362 matchup points between the NOAA satellite data (NOAA-ll, NOAA-12, and NOAA-J4) and the drifter buoy temperatures. The split window technique of linear MCSST, non-linear CPSST and NLSST showed relatively small rms (root mean square) errors in the range of O.9°C to 1.2°C compared with the other window methods. However, a predominant trend was found that satellite-derived SSTs are underestimated by as much as −2°C in dry atmospheric conditions during winter, and overestimated in very humid conditions in summer by approximately 2°C. The characteristic trend was removed using a regression method, and the rms errors of newly-derived equations for the split window MCSST and the non-linear SST optimized to the East Sea were improved to within O.3°C ∼ O.9°C. The locally-optimized SSTs may be more important than the SSTs based on the global database, particularly in the inaccessible regions off North Korea and sea ice regions that are important for the critical research issue of cold water formation in the East Sea.

Improved Cloud Detection in Along Track Scanning Radiometer (ATSR) Data over the Ocean

Valid estimates of sea surface temperature (SST) from satellite data [e.g., the Along Track Scanning Radiometer (ATSR)] are critically dependent upon the identification and removal of cloud from the data, but few cloud-screening algorithms for ATSR data have appeared in the literature. A new algorithm, the ATSR Split-and-Merge Clustering (ATSR/SMC) algorithm, for cloud masking ATSR data is presented which evaluates every pixel in the image, is statistically reproducible, computationally efficient, and requires no knowledge of cloud type. Moreover, it is effective in detecting multilayer cloud structures in a scene, which is a difficult task because such systems generally have bimodal statistical distributions. It also accurately detects glint radiance, which is quite common in at least one of the 1.6 μm views, subpixel cloud contamination near cloud boundaries and low-lying marine stratiform cloud. Historically, these issues have interfered with ATSR-based SST retrieval [see the work of Jones et al., (1996a,b) and the references cited therein]. The SSTs derived from the cloud-free ocean pixels were validated with 96 buoy observations and the mean difference (buoy−SST) was +0.24°C±0.51°C. For the 103 pairs of images (forward/nadir views) tested, the mean 11 μm BTs that result from SADIST (standard ATSR processing) vs. ATSR/SMC cloud detection are 0.4°C (daytime) and 0.6°C (nighttime) colder for SADIST than for ATSR/SMC, even though the SADIST cloud masks generally overdetect clouds relative to ATSR/SMC cloud masks. These results, plus others discussed in the text, support the conclusion that the new procedure produces cloud masks which are superior to the standard ATSR operational cloud mask product and it retains substantially more valid pixels. The algorithm can be used in tropical and midlatitude regions. It is not designed to detect sea ice, and consequently should not be used in polar regions. Finally, the approach can easily be adapted to ATSR-2 data and to other data to be taken from soon to be launched sensors.

A cloud screening algorithm for daytime AVHRR data using dynamic thresholds

Cloud screening algorithms often use fixed thresholds for reflectances and temperatures and are not applicable to regions with a great varibility in surface reflectance characteristics. A dynamic threshold algorithm was developed for application to a broad spectrum of earth surfaces, such as desert areas with very high reflectances and areas with dense vegetation. No a priori information is needed to activate the unsupervised algorithm. Using the maximum Normalized Difference Vegetation Index (NDVI) composites of a ten-day period, the corresponding reflectance and temperature maps are used as dynamic thresholds. Empirically derived nadir correction factors reduce the influence of surface anisotropy and tighten the algorithm considerably. Thus the only use of reflectance data results in a reliable, surface independent, cloud screening. Additionally, tests with thermal channels can be applied for confirmation and classification of questionable mixed pixels. Since there is no need for input data, the method can be used on a near global scale. The algorithm was developed for NDVI evaluation. No attempts were made so far to distinguish between clouds and snow or ice.

Reduction of noise in AVHRR channel 3 data with minimum distortion

The channel 3 data of the Advanced Very High Resolution Radiometer (AVHRR) on the NOAA series of weather satellites (NOAA 6-12) are contaminated by instrumentation noise. The signal to noise ratio (S/N) varies considerably from image to image and the between sensor variation in S/N can be large. The characteristics of the channel noise in the image data are examined using Fourier techniques. A Wiener filtering technique is developed to reduce the noise in the channel 3 image data. The noise and signal power spectra for the Wiener filter are estimated from the channel 3 and channel 4 AVHRR data in a manner which makes the filter adaptive to observed variations in the noise power spectra. Thus, the degree of filtering is dependent upon the level of noise in the original data and the filter is adaptive to variations in noise characteristics. Use of the filtered data to improve image segmentation, labeling in cloud screening algorithms for AVHRR data, and multichannel sea surface temperature (MCSST) estimates is demonstrated. Examples also show that the method can be used with success in land applications. The Wiener filtering model is compared with alternate filtering methods and is shown to be superior in all applications tested. >