Search Results
You are looking at 1 - 10 of 29 items for
- Author or Editor: Louis Garand x
- Refine by Access: All Content x
Abstract
Infrared heating rate calculations with the emissivity formulation are improved by developing simple methods to minimize the effects of the two main sources of error 1) the use of the strong-line approximation; 2) the neglect of temperature variations along the optical path. When compared to band-by-hand calculations, the resulting new scheme maintains a 5% accuracy on cooling rates, and errors generally less than 2 W m−2 on fluxes. The most recent band parameters of MeClatchey et al. (1973) are used in the band-by-band calculations, including measured data for the vibration-rotation bands.
Another novel aspect of the new scheme is the treatment of “e-type” absorption, which is a very important effect in hot humid air near the ground. An implicit parameterization is developed which allows for “e-type” absorption in the pre-tabulated transmission functions, and so requires no computational overhead. This is based on the derivation of an approximate formula which expresses the optical depth due to “e-type” absorption in terms of the optical depth due to water vapor.
A 10-level version requires 1.7 ms on a CDC 7600 computer. Carbon dioxide and clouds are included, and provision is made for the inclusion of ozone.
Abstract
Infrared heating rate calculations with the emissivity formulation are improved by developing simple methods to minimize the effects of the two main sources of error 1) the use of the strong-line approximation; 2) the neglect of temperature variations along the optical path. When compared to band-by-hand calculations, the resulting new scheme maintains a 5% accuracy on cooling rates, and errors generally less than 2 W m−2 on fluxes. The most recent band parameters of MeClatchey et al. (1973) are used in the band-by-band calculations, including measured data for the vibration-rotation bands.
Another novel aspect of the new scheme is the treatment of “e-type” absorption, which is a very important effect in hot humid air near the ground. An implicit parameterization is developed which allows for “e-type” absorption in the pre-tabulated transmission functions, and so requires no computational overhead. This is based on the derivation of an approximate formula which expresses the optical depth due to “e-type” absorption in terms of the optical depth due to water vapor.
A 10-level version requires 1.7 ms on a CDC 7600 computer. Carbon dioxide and clouds are included, and provision is made for the inclusion of ozone.
Abstract
A retrieval technique based on cloud classification is designed to derive humidity profiles from Meteosat visible (VIS), infrared window (IR), and water vapor (WV) channels, or equivalent sensors available on other satellites. Dewpoint depression (DPD) is the variable retrieved at six standard levels: 1000, 850, 700, 500, 400, and 300 mb. Collocation of soundings and Meteosal-2 imagery was obtained over Europe for March, June, and July 1988. Results are derived from over 2000 dependent and 1000 independent samples.
It is found that a classification in seven (IR only) or nine (VIS-IR) classes contains the essential information on cloud type for the application sought. Measures were extracted from approximately 8-km pixel resolution images on 80-km × 80-km and 160-km × 160-km arm, little dependency on horizontal scale was found for the mean humidity profiles associated with each cloud class. The WV channel proved very useful in improving DPDs at higher levels while the VIS channel improved inferences of low-level humidity in classes associated with precipitation. Overall DPD errors range from 3 to 5 K rms depending on level; this corresponds to 13%–20% rms in terms of relative humidity and to approximately 4.4 mm rms in terms of total precipitable water. The three GOES-7 channels closest to Meteosal-2 VIS, IR, and WV channels were used to extend the study to the tropics and to the winter season from data collected in 1991 and 1992. The main advantages of the technique are its applicability to cloudy atmospheres, its robustness and the fact that it can efficiently provide retrievals from 60°S to 60;deg;N every half-hour.
Abstract
A retrieval technique based on cloud classification is designed to derive humidity profiles from Meteosat visible (VIS), infrared window (IR), and water vapor (WV) channels, or equivalent sensors available on other satellites. Dewpoint depression (DPD) is the variable retrieved at six standard levels: 1000, 850, 700, 500, 400, and 300 mb. Collocation of soundings and Meteosal-2 imagery was obtained over Europe for March, June, and July 1988. Results are derived from over 2000 dependent and 1000 independent samples.
It is found that a classification in seven (IR only) or nine (VIS-IR) classes contains the essential information on cloud type for the application sought. Measures were extracted from approximately 8-km pixel resolution images on 80-km × 80-km and 160-km × 160-km arm, little dependency on horizontal scale was found for the mean humidity profiles associated with each cloud class. The WV channel proved very useful in improving DPDs at higher levels while the VIS channel improved inferences of low-level humidity in classes associated with precipitation. Overall DPD errors range from 3 to 5 K rms depending on level; this corresponds to 13%–20% rms in terms of relative humidity and to approximately 4.4 mm rms in terms of total precipitable water. The three GOES-7 channels closest to Meteosal-2 VIS, IR, and WV channels were used to extend the study to the tropics and to the winter season from data collected in 1991 and 1992. The main advantages of the technique are its applicability to cloudy atmospheres, its robustness and the fact that it can efficiently provide retrievals from 60°S to 60;deg;N every half-hour.
Abstract
Two methods of deriving probability of precipitation fields (PP) over oceanic areas are presented and compared. The cloud fields are analyzed at the scale of ∼130–150 km from satellite visible and infrared imagery and collocated with ship observations of present weather. Method 1 is based on a detailed cloud classification scheme in 20 classes: a mean PP is determined for each cloud class. Method 2 assigns a PP based on cloud top temperature and mean cloud albedo only. For both methods, a normalization with respect to cloud fraction is applied. Method 1 involves more cloud field descriptors than Method 2, but the latter is simpler to implement and much faster. The PPs assigned to individual cloud fields vary between 0% and 65%. The importance of the visible sensor is clearly demonstrated, i.e., infrared-only techniques will be much less accurate.
For real-time applications, the two methods provide similar results except for some specific cloud classes where maximum differences reach 13%, due to the lower level of classification used in Method 2. On a monthly time scale, the absolute accuracy of both methods is about 1.2% rms, based on independent data taken during the winters of 1984 (1064 samples) and 1986 (673 samples) over the northwestern Atlantic. From the 3 months for which PP maps are produced, the average local variability between months is 4.8% rms. It follows that the PP variance between months is typically 16 times larger than the error variance of the satellite estimates. Thus both methods provide a reliable precipitation indicator.
Abstract
Two methods of deriving probability of precipitation fields (PP) over oceanic areas are presented and compared. The cloud fields are analyzed at the scale of ∼130–150 km from satellite visible and infrared imagery and collocated with ship observations of present weather. Method 1 is based on a detailed cloud classification scheme in 20 classes: a mean PP is determined for each cloud class. Method 2 assigns a PP based on cloud top temperature and mean cloud albedo only. For both methods, a normalization with respect to cloud fraction is applied. Method 1 involves more cloud field descriptors than Method 2, but the latter is simpler to implement and much faster. The PPs assigned to individual cloud fields vary between 0% and 65%. The importance of the visible sensor is clearly demonstrated, i.e., infrared-only techniques will be much less accurate.
For real-time applications, the two methods provide similar results except for some specific cloud classes where maximum differences reach 13%, due to the lower level of classification used in Method 2. On a monthly time scale, the absolute accuracy of both methods is about 1.2% rms, based on independent data taken during the winters of 1984 (1064 samples) and 1986 (673 samples) over the northwestern Atlantic. From the 3 months for which PP maps are produced, the average local variability between months is 4.8% rms. It follows that the PP variance between months is typically 16 times larger than the error variance of the satellite estimates. Thus both methods provide a reliable precipitation indicator.
Abstract
Geostationary Operational Environmental Satellite (GOES)-East and -West window channel radiances are directly assimilated using a 1D variational technique, providing surface skin temperature (T s ) estimates over all surface types (land, water, or ice) from a unique system. This is an important advantage over commonly used regression methods, such as split window. The physical nature of the method allows any combination of channels to be used, and adaptation to new sensors is straightforward. A full month (May 2001) of GOES-8 and -10 data is processed every 6 h; T s estimates are obtained using radiances from imager channels 4 (11 μm) and 5 (12 μm). Imager channel 2 (3.9 μm) can also be used at night. Surface emissivity maps were constructed from available information based on surface type. The diurnal cycle is studied; its range is on the order of 0.7 K over the ocean. Over land, the diurnal range reaches 30 K for mountainous regions, such as the Rockies or Andes. A full GOES disk image can be processed in 4 min. The resulting retrieval error can be estimated locally. It is usually in the range of 0.5–1.0 K over ocean and 0.9–2.4 K over land. Differences between collocated GOES-8 and -10 retrievals are examined. These are as low as 0.23-K rms over ocean; over land they are in the range of 1.4–2.2-K rms with higher values in midafternoon because of higher emission anisotropy and local variability. The model background (6-h forecast) is found to underestimate the diurnal cycle over land by nearly 50%. A validation is done over land using surface data of upward broadband longwave radiation converted into equivalent skin temperature. These data confirm the range of the diurnal cycle and the model biases inferred from satellite data. Over oceans, the agreement between retrievals and ship, fixed-buoy, and drifting-buoy observations is 1.05, 0.90, and 0.65 K, respectively.
Abstract
Geostationary Operational Environmental Satellite (GOES)-East and -West window channel radiances are directly assimilated using a 1D variational technique, providing surface skin temperature (T s ) estimates over all surface types (land, water, or ice) from a unique system. This is an important advantage over commonly used regression methods, such as split window. The physical nature of the method allows any combination of channels to be used, and adaptation to new sensors is straightforward. A full month (May 2001) of GOES-8 and -10 data is processed every 6 h; T s estimates are obtained using radiances from imager channels 4 (11 μm) and 5 (12 μm). Imager channel 2 (3.9 μm) can also be used at night. Surface emissivity maps were constructed from available information based on surface type. The diurnal cycle is studied; its range is on the order of 0.7 K over the ocean. Over land, the diurnal range reaches 30 K for mountainous regions, such as the Rockies or Andes. A full GOES disk image can be processed in 4 min. The resulting retrieval error can be estimated locally. It is usually in the range of 0.5–1.0 K over ocean and 0.9–2.4 K over land. Differences between collocated GOES-8 and -10 retrievals are examined. These are as low as 0.23-K rms over ocean; over land they are in the range of 1.4–2.2-K rms with higher values in midafternoon because of higher emission anisotropy and local variability. The model background (6-h forecast) is found to underestimate the diurnal cycle over land by nearly 50%. A validation is done over land using surface data of upward broadband longwave radiation converted into equivalent skin temperature. These data confirm the range of the diurnal cycle and the model biases inferred from satellite data. Over oceans, the agreement between retrievals and ship, fixed-buoy, and drifting-buoy observations is 1.05, 0.90, and 0.65 K, respectively.
Abstract
A scheme is presented for the automated classification of oceanic cloud patterns. The 20 cloud classes reflect the rich variety of morphologies that are detectable from space. A training set is defined by 2000 samples of size 128 × 128 km taken from GOES visible and infrared images over the western Atlantic in February 1984. Class discrimination is obtained from 13 features representing height, albedo, shape and multilayering characteristics of the cloud fields. Two features derived from the two-dimensional power spectrum of the visible images proved essential for the detection of directional patterns (cloud “streets or rolls) and open cells. Based on the assumption of multinormal distributions of the features, a simple classification algorithm is developed. The generation of artificial samples yields a theoretical separability of 97% while the actual separability obtained on the training set is 95%. From 1020 independent samples, the separate verification of three expert nephanalysts indicates strict accuracy in 79% of the cases while there is agreement with their first or second choice in 89% of the cases.
The cloud climatology is compared in 20 classes for January and February 1984. In agreement with available climatology, multilayered cloud fields are observed 42% of the time. The cloud fraction maps are also compared with the observed fields from ships.
Abstract
A scheme is presented for the automated classification of oceanic cloud patterns. The 20 cloud classes reflect the rich variety of morphologies that are detectable from space. A training set is defined by 2000 samples of size 128 × 128 km taken from GOES visible and infrared images over the western Atlantic in February 1984. Class discrimination is obtained from 13 features representing height, albedo, shape and multilayering characteristics of the cloud fields. Two features derived from the two-dimensional power spectrum of the visible images proved essential for the detection of directional patterns (cloud “streets or rolls) and open cells. Based on the assumption of multinormal distributions of the features, a simple classification algorithm is developed. The generation of artificial samples yields a theoretical separability of 97% while the actual separability obtained on the training set is 95%. From 1020 independent samples, the separate verification of three expert nephanalysts indicates strict accuracy in 79% of the cases while there is agreement with their first or second choice in 89% of the cases.
The cloud climatology is compared in 20 classes for January and February 1984. In agreement with available climatology, multilayered cloud fields are observed 42% of the time. The cloud fraction maps are also compared with the observed fields from ships.
Abstract
A strong linearity exists between the 6.7-μm clear-sky outgoing brightness temperature (BT) and dewpoint depression (DPD) at upper-tropospheric levels. A similar relationship, using the logarithm of relative humidity instead of DPD, was developed by Soden and Bretherton. Here, however, the humidity at specific levels is derived as opposed to the humidity integrated over upper-tropospheric levels. Linear relationships are obtained between a 6-h model forecast of DPD and calculated BTs at different viewing angles. The data are further stratified in terms of 400-mb temperature as an indicator of airmass type. Applying these relationships using observed 6.7-μm BTs and a 6-h forecast of 400-mb temperature yields vertically correlated estimates of DPD between 200 and 500 mb, with DPD typically decreasing with height, and corresponding rms error estimates in the range 3–6 K. The retrieval technique is applied to GOES-8 and GOES-9 data, which cover about 40% of the globe. In cloudy regions, proxy humidity estimates based on cloud classification are used. These clear- and cloudy-sky DPD estimates are assimilated every 6 h in a global forecast model, taking into consideration the horizontal correlation of the error. The system is supplemented by quality-control procedures.
In parallel runs at the Canadian Meteorological Centre, the analyses and forecasts with satellite data (SAT) were found significantly improved with respect to those without satellite data (NOSAT). The system was therefore implemented. The superiority of the SAT forecasts in terms of 6.7-μm BT, 2-K versus 4-K rms at initial time, gradually decreases to the level of the NOSAT forecasts in 48 h. A slight improvement on geopotential, DPD, and temperature is observed in 48-h forecasts with respect to radiosondes over North America. The new upper-tropospheric DPD retrieval technique is robust and could easily be applied to other geostationary or polar-orbiting platforms providing 6.7-μm imagery.
Abstract
A strong linearity exists between the 6.7-μm clear-sky outgoing brightness temperature (BT) and dewpoint depression (DPD) at upper-tropospheric levels. A similar relationship, using the logarithm of relative humidity instead of DPD, was developed by Soden and Bretherton. Here, however, the humidity at specific levels is derived as opposed to the humidity integrated over upper-tropospheric levels. Linear relationships are obtained between a 6-h model forecast of DPD and calculated BTs at different viewing angles. The data are further stratified in terms of 400-mb temperature as an indicator of airmass type. Applying these relationships using observed 6.7-μm BTs and a 6-h forecast of 400-mb temperature yields vertically correlated estimates of DPD between 200 and 500 mb, with DPD typically decreasing with height, and corresponding rms error estimates in the range 3–6 K. The retrieval technique is applied to GOES-8 and GOES-9 data, which cover about 40% of the globe. In cloudy regions, proxy humidity estimates based on cloud classification are used. These clear- and cloudy-sky DPD estimates are assimilated every 6 h in a global forecast model, taking into consideration the horizontal correlation of the error. The system is supplemented by quality-control procedures.
In parallel runs at the Canadian Meteorological Centre, the analyses and forecasts with satellite data (SAT) were found significantly improved with respect to those without satellite data (NOSAT). The system was therefore implemented. The superiority of the SAT forecasts in terms of 6.7-μm BT, 2-K versus 4-K rms at initial time, gradually decreases to the level of the NOSAT forecasts in 48 h. A slight improvement on geopotential, DPD, and temperature is observed in 48-h forecasts with respect to radiosondes over North America. The new upper-tropospheric DPD retrieval technique is robust and could easily be applied to other geostationary or polar-orbiting platforms providing 6.7-μm imagery.
Abstract
This study explores the feasibility of performing an objective analysis of instantaneous rain rate combining satellite estimates (and eventually other types of observations) with those from a numerical prediction model using the method of statistical interpolation. Results demonstrate that the quality of the short-term precipitation forecasts serving as background field has reached a level that makes such an objective analysis possible.
The two main requirements to obtain an accurate analysis from available information are a realistic estimate of background field and observation errors and knowledge of the horizontal correlation of these errors with distance. The importance of specifying the errors for joint model-observation situations is emphasized; it is especially important in situations where model and observations are in conflict. These aspects of the problem are studied using collocated 6-h forecast with satellite estimates derived from visible and infrared imagery, and ground-truth rainfall data available over Japanese territory from the Global Precipitation Climatology Project. Over 90 000 truth-model-satellite collocations are available at the common scale of 130 km × 130 km. An alternative means of establishing the model error correlation with distance and azimuth direction from 6- and 18-h forecast differences valid at the same time yield results that are similar to those derived from collocations with truth rainfall over large domains, but not locally, this result suggests a means of relaxing the assumption of homogeneity and isotropy of model errors. The sensitivity of the rain rate analysis to different specifications of the satellite to model error ratios is shown with an example.
Abstract
This study explores the feasibility of performing an objective analysis of instantaneous rain rate combining satellite estimates (and eventually other types of observations) with those from a numerical prediction model using the method of statistical interpolation. Results demonstrate that the quality of the short-term precipitation forecasts serving as background field has reached a level that makes such an objective analysis possible.
The two main requirements to obtain an accurate analysis from available information are a realistic estimate of background field and observation errors and knowledge of the horizontal correlation of these errors with distance. The importance of specifying the errors for joint model-observation situations is emphasized; it is especially important in situations where model and observations are in conflict. These aspects of the problem are studied using collocated 6-h forecast with satellite estimates derived from visible and infrared imagery, and ground-truth rainfall data available over Japanese territory from the Global Precipitation Climatology Project. Over 90 000 truth-model-satellite collocations are available at the common scale of 130 km × 130 km. An alternative means of establishing the model error correlation with distance and azimuth direction from 6- and 18-h forecast differences valid at the same time yield results that are similar to those derived from collocations with truth rainfall over large domains, but not locally, this result suggests a means of relaxing the assumption of homogeneity and isotropy of model errors. The sensitivity of the rain rate analysis to different specifications of the satellite to model error ratios is shown with an example.
Abstract
Both the issues of high-resolution satellite analysis and model evaluation for a region centered on the Arctic Circle (60°–75°N) are addressed. Model cloud fraction, cloud height, and outgoing radiation are compared to corresponding satellite observations using a model-to-satellite approach (calculated radiances from model state). The dataset consists of forecasts run at 15-km resolution up to 30 h and nearly coincident Advanced Very High Resolution Radiometer (AVHRR) imagery during the Beaufort and Arctic Storm Experiment over the Mackenzie Basin for a monthly period in the fall of 1994. A cloud detection algorithm is designed for day and night application using the 11-μ channel of AVHRR along with available information on atmospheric and ground temperatures. The satellite and model estimates of cloud fraction are also compared to observations at 20 ground stations.
A significant result of the validation is that the model has a higher frequency of low cloud tops and a lower frequency of midlevel cloud tops than the observations. On a monthly basis, the model 11-μ outgoing brightness temperature (TB) is consequently higher than observed by about 4.4 K at all forecast times, which corresponds to a deficit of 760 m in mean cloud-top height and about 10 W m−2 in outgoing flux at the top of the atmosphere. Possible errors in the parameterization of ice or water cloud emissivity are evaluated but ruled out as the dominant cause for the warm TB bias in the model. Rather, the problem is attributed to low clouds being trapped in the boundary layer, whereas high clouds appear to be reasonably well modeled.
The role of thin ice clouds is further evaluated by comparing distributions of observed and modeled 11-μ minus 12-μ TB differences, DIF45 (channel 4 minus channel 5). The relationship between the true height of the clouds and the effective height observed by satellite is modeled from forecast outputs as a function of DIF45. The quality of daily estimates is evaluated from time series at various locations. The time series shows that there was a marked drop in DIF45 during the month, which is attributed to a decrease in the occurrence of cirrus clouds. Finally, the diurnal cycle of TB and cloud fraction is found to be relatively large with average monthly 0600–1800 UTC TB differences of both signs of the order of 4–8 K in broad sectors and cloud fraction differences of 10%–30%. Where low clouds prevail, the cloud fraction tends to decrease at night and TB increases. Overall, model–observation differences are dominated by differences in the vertical distribution of clouds. A reduction of this effect implies a modification of the “preferred” model climatology in terms of its vertical distribution of humidity and cloud water.
Abstract
Both the issues of high-resolution satellite analysis and model evaluation for a region centered on the Arctic Circle (60°–75°N) are addressed. Model cloud fraction, cloud height, and outgoing radiation are compared to corresponding satellite observations using a model-to-satellite approach (calculated radiances from model state). The dataset consists of forecasts run at 15-km resolution up to 30 h and nearly coincident Advanced Very High Resolution Radiometer (AVHRR) imagery during the Beaufort and Arctic Storm Experiment over the Mackenzie Basin for a monthly period in the fall of 1994. A cloud detection algorithm is designed for day and night application using the 11-μ channel of AVHRR along with available information on atmospheric and ground temperatures. The satellite and model estimates of cloud fraction are also compared to observations at 20 ground stations.
A significant result of the validation is that the model has a higher frequency of low cloud tops and a lower frequency of midlevel cloud tops than the observations. On a monthly basis, the model 11-μ outgoing brightness temperature (TB) is consequently higher than observed by about 4.4 K at all forecast times, which corresponds to a deficit of 760 m in mean cloud-top height and about 10 W m−2 in outgoing flux at the top of the atmosphere. Possible errors in the parameterization of ice or water cloud emissivity are evaluated but ruled out as the dominant cause for the warm TB bias in the model. Rather, the problem is attributed to low clouds being trapped in the boundary layer, whereas high clouds appear to be reasonably well modeled.
The role of thin ice clouds is further evaluated by comparing distributions of observed and modeled 11-μ minus 12-μ TB differences, DIF45 (channel 4 minus channel 5). The relationship between the true height of the clouds and the effective height observed by satellite is modeled from forecast outputs as a function of DIF45. The quality of daily estimates is evaluated from time series at various locations. The time series shows that there was a marked drop in DIF45 during the month, which is attributed to a decrease in the occurrence of cirrus clouds. Finally, the diurnal cycle of TB and cloud fraction is found to be relatively large with average monthly 0600–1800 UTC TB differences of both signs of the order of 4–8 K in broad sectors and cloud fraction differences of 10%–30%. Where low clouds prevail, the cloud fraction tends to decrease at night and TB increases. Overall, model–observation differences are dominated by differences in the vertical distribution of clouds. A reduction of this effect implies a modification of the “preferred” model climatology in terms of its vertical distribution of humidity and cloud water.
Abstract
Using Global Precipitation Climatology Project data gathered during June, July, and August 1989 over Japan, rainfall estimates are examined from both geostationary satellite imagery using a multifeature classification approach, and from short-term weather prediction model fields. Additionally, the utility of combining model forecast information within such a classifier to improve the final estimate is investigated. During both months satellite estimates are superior to model forecasts in detecting heavy rain events associated with extremely cold cloud tops, and in identifying cloud-free regions. Model estimates are superior to satellite retrievals in terms of dynamic range and regional bias. Addition of visible data to an infrared-only scheme improved monthly rainfall estimates during June, and hourly estimates during both months. In June it is shown that a combined satellite-model method clearly yields improved retrievals of rainfall relative to those obtained by using either satellite data or model forecasts alone at both monthly and hourly time scales. However, in July and August, satellite retrievals largely underestimated monthly rainfall, and the model produced poor hourly forecasts. Generally, a good model forecast of rainfall can enhance the satellite estimate, while a poor forecast will degrade it. If results obtained during June are found to be valid for other regions of the globe, such a method could be used to develop rainfall climatologies. It could also be used in a real-time operational numerical weather prediction environment since it is computationally rapid, with only geostationary satellite observations and model-predicted fields needed to derive the estimates.
Abstract
Using Global Precipitation Climatology Project data gathered during June, July, and August 1989 over Japan, rainfall estimates are examined from both geostationary satellite imagery using a multifeature classification approach, and from short-term weather prediction model fields. Additionally, the utility of combining model forecast information within such a classifier to improve the final estimate is investigated. During both months satellite estimates are superior to model forecasts in detecting heavy rain events associated with extremely cold cloud tops, and in identifying cloud-free regions. Model estimates are superior to satellite retrievals in terms of dynamic range and regional bias. Addition of visible data to an infrared-only scheme improved monthly rainfall estimates during June, and hourly estimates during both months. In June it is shown that a combined satellite-model method clearly yields improved retrievals of rainfall relative to those obtained by using either satellite data or model forecasts alone at both monthly and hourly time scales. However, in July and August, satellite retrievals largely underestimated monthly rainfall, and the model produced poor hourly forecasts. Generally, a good model forecast of rainfall can enhance the satellite estimate, while a poor forecast will degrade it. If results obtained during June are found to be valid for other regions of the globe, such a method could be used to develop rainfall climatologies. It could also be used in a real-time operational numerical weather prediction environment since it is computationally rapid, with only geostationary satellite observations and model-predicted fields needed to derive the estimates.
Abstract
A structural-stochastic image model is developed for the analysis and synthesis of cloud images. The ability of the model to characterize the visual appearance of cloud fields observed by satellite with a limited number of parameters is demonstrated. The model merges structural and stochastic information, the stochastic model acting as a local statistical operator applied to the output of the structural model. The structural or large-scale organization of the scene is retrieved from the two-dimensional Fourier representation of the digital image. The pattern generated by the major Fourier components provides a first guess of the scene. The stochastic aspect is described by a Markov model of texture that assumes a binomial probability distribution for the local grey-level variability. This Markov model provides four parameters that represent the clustering strength in the horizontal, vertical and diagonal directions. These parameters are estimated by a standard maximum-likelihood technique. The image can be reproduced with a fair degree of verisimilitude from these parameters. The data compression factor is of the order of one hundred to several hundreds.
Abstract
A structural-stochastic image model is developed for the analysis and synthesis of cloud images. The ability of the model to characterize the visual appearance of cloud fields observed by satellite with a limited number of parameters is demonstrated. The model merges structural and stochastic information, the stochastic model acting as a local statistical operator applied to the output of the structural model. The structural or large-scale organization of the scene is retrieved from the two-dimensional Fourier representation of the digital image. The pattern generated by the major Fourier components provides a first guess of the scene. The stochastic aspect is described by a Markov model of texture that assumes a binomial probability distribution for the local grey-level variability. This Markov model provides four parameters that represent the clustering strength in the horizontal, vertical and diagonal directions. These parameters are estimated by a standard maximum-likelihood technique. The image can be reproduced with a fair degree of verisimilitude from these parameters. The data compression factor is of the order of one hundred to several hundreds.
