Realistic assessment of the vertical distribution of clouds, particularly the occurrence of multi-layered systems, is critical for accurate calculations of radiative transfer in general circulation models. Such information is also useful in the design and improvement of satellite retrieval techniques. Current methods for retrieving cloud properties from satellite data assume that the clouds reside in single-layered systems. These methods are not expected to be successful for multi-layered systems. Attempts to specifically address the question of cloud layering have thus far been limited, due in part to the difficulties of inferring vertical cloud structure from either surface or satellite data. In situ observations, such as those provided by aircraft, are available only for localized regions and are limited in time. This study uses data from a lidar instrument flown onboard the space shuttle and satellite imagery data to identify the frequencies of occurrence of layered cloud systems at different spatial scales over various regions of the globe. The Lidar In-Space Technology Experiment (L1TE) was flown on Space Shuttle Discovery in September 1994 and provided global-scale, high vertical resolution profiles of the earth's troposphere and lower stratosphere. Analysis of the LITE observations requires distinguishing clouds residing in organized, well-defined layers from clouds that are distributed in altitude throughout the troposphere. The analysis employs a histogram technique in which peaks having some critical number of observations are considered to correspond to observations belonging to well-defined cloud layers. Advanced Very High Resolution Radiometer (AVHRR) data for the 11-day duration of the LITE mission are analyzed using the spatial coherence method. This method identifies regions of locally uniform emission which are associated either with cloud-free pixels or with overcast pixels corresponding to clouds in a single layer at a well-defined altitude. The number of layers present is determined by the number of overcast radiances associated with pixel arrays exhibiting locally uniform emission within the region. Layer statistics are compiled for the Pacific, Atlantic, and Indian Oceans and the North and South American, African, European, Asian, and Australian continents using horizontal scales of 60 and 250 km. The results indicate a strong dependence on the spatial scale chosen for the analysis, with two- and three-layered systems more prevalent at the 250-km scale. Analysis of cloud-top altitudes from LITE and AVHRR show that low-level cloud systems comprise the majority of the observations over both ocean and land. / Graduation date: 1998

A radiation model was developed for retrieving cloud visible optical depth, droplet effective radius, and cloud top emission temperature using AVHRR satellite observations at 0.63, 3.7, and 11 ��m. The model was used to determine the sensitivity of the retrieved properties to various approximations often employed in such retrievals. Droplet effective radius appears to be the most sensitive to the commonly used approximations. Cloud properties retrieved using a 16-stream scheme were within ��5% of those retrieved using a 148-stream scheme. Cloud properties retrieved using double Henyey-Greenstein phase functions were within ��10% of those retrieved using Mie scattering. The retrieved cloud properties were used to investigate biases that arise when partly cloudy pixels were assumed to be overcast and biases that arise due to oblique satellite view angles. On average, cloud visible optical depths retrieved for partly cloudy pixels were 40-60% of those retrieved for overcast pixels. Likewise, cloud liquid water paths were 30-50%, droplet effective radii were 1-3 ��m smaller, and cloud top emission temperatures were 2-4K larger. Cloud visible optical depths retrieved at 60�� satellite zenith angles were 60-70% of those retrieved at nadir. The retrieved droplet effective radii and cloud top emission temperatures varied little with changing satellite zenith angle. For March 1989, cloud optical depths and cloud emission temperatures retrieved for pixels overcast by single-layer, low-level clouds were negatively correlated. Cloud optical depth, liquid water path, and droplet effective radius were positively correlated with the sea surface-cloud top temperature difference. The retrieved cloud properties were also compared for the spatial coherence, CLAVR (Clouds from AVHRR), and a threshold method based on International Satellite Cloud Climatology Project procedures. For regions containing single-layered cloud systems, fractional cloud cover and cloud brightness temperatures derived by the ISCCP-like threshold method were systematically larger than those derived by the spatial coherence method, whereas cloud reflectivities were systematically smaller. Cloud reflectivities and brightness temperatures derived by CLAVR and the spatial coherence method were in better agreement. / Graduation date: 1998

An improved physically-based rainfall algorithm was developed using AMSR-E data based on a radiative transfer model. In addition, error models were designed and embedded in the algorithm to assess retrieval errors quantitatively and to reduce net retrieval uncertainties. The algorithm uses six channels (dual polarizations at 36.5, 18.7 and 10.65GHz) and retrieves rain rates on a pixel-by-pixel basis. Monthly rain totals are estimated by summing average rain rates computed by merging six rain rates based on proper weights that are estimated from error models. Error models were constructed based upon the principal error sources of rainfall retrieval such as beam filling error, drop size distribution uncertainty and instrument calibration errors. Several improved schemes that minimize uncertainties of the rainfall retrieval were developed in this study. In particular, improved offset correction that corrects the biases near zero rain plays a very important role for reducing uncertainties which are mainly driven by calibration uncertainty including the modeling errors. AMSR-E's larger calibration uncertainty was substantially absorbed by this offset correction as well as by the weighted average scheme to combine all six channels optimally. As a framework for inter-comparison with the experimental algorithm, the current operational algorithm (NASA, level 3 algorithm) was also updated with respect to AMSR-E data. The experimental algorithm was compared with the operational algorithm for both AMSR-E and TMI data and rainfall retrieval uncertainties were analyzed using error models. When the experimental algorithm was used, many limitations of the operational algorithm were overcome and uncertainties of rainfall retrieval were considerably eliminated.

Cirrus clouds are known to play a key role in the earth's radiation budget, yet are one of the most uncertain components of the earth-atmosphere system. With the development of instruments such as the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and the Moderate-resolution Infrared Spectroradiometer (MODIS), scientists now have an unprecedented ability to study cirrus clouds. To aid in the understanding of such clouds, a significant study of cirrus radiative properties has been undertaken. This research is composed of three parts: 1) the retrieval of tropical cirrus optical thickness using MODIS level-1b calibrated radiance data, 2) a survey of tropical cirrus cloud cover, including seasonal variations, using MODIS level-3 global daily gridded data, and 3) the simultaneous retrieval of cirrus optical thickness and ice crystal effective diameter using AVIRIS reflectance measurements.

Land surface evaporation has considerable spatial variability that is not reflected in meteorological station data alone. Knowing the spatial variability of evaporation is important for describing drought, managing agricultural land, and is valuable for improving the parameterization of hydrological models and land surface schemes over large areas. General difficulties arise for obtaining reliable, spatially distributed evaporation estimates as a result of uncertainty in estimation techniques, scale issues and complexities regarding land surface and atmospheric interactions, and the spatial and temporal variability of key factors governing the evaporation process. Estimating evaporation is further complicated when soil moisture becomes a critical limitation, particularly during drought. An examination of the spatial variability of evaporation and its association with governing factors was conducted in Prairie landscapes using three modelling techniques. First, eddy covariance measurements and reference meteorological data were obtained at two Prairie locations to assess the accuracy of physically-based models for calculating point estimates of actual evaporation under non-limited soil moisture conditions and during drought. Second, estimates of actual evaporation were distributed at the field scale in order to examine the impacts of driving factors and their spatial associations on upscaled evaporation estimates. This required the assimilation of high resolution visible and thermal images which were used to derive estimates of surface albedo and surface emitted longwave radiation. These were combined along with surface reference observations to develop an index of the mid-day radiation in order to distribute a known value of mean daily net radiation over the field. Third, archived historical climate data were used as input for a continuous hydrological simulation to examine spatial and temporal variations in evaporation across the Prairie region of Western Canada during a drought and non-drought period. Results of this research showed that the spatial variability of evaporation could be derived at the field scale by integrating remote sensing and surface reference climate data with a physically-based evaporation model. Surface temperature and soil moisture, and net radiation were found to be highly variable spatially at field scales whilst meteorological conditions tended to be less variable spatially but showed strong temporal variability. At the field scale it was found that the variability in albedo and surface temperature were both important for characterizing differences in surface state conditions. Their combined influence was reflected in the resulting pattern of net radiation that governed the distribution of actual evaporation estimates obtained with the Granger and Gray evaporation model. It was found that an areal estimate of evaporation obtained from the means of driving factors was similar to the areal average obtained from the distributed estimates. This was attributed to the offsetting interactions among the driving factors which effectively reduced the variability of the model estimates. In general, the physically-based models examined were found to provide reasonable estimates of actual evaporation when driven by observations at point-scales over multi-day and seasonal periods. This included periods when soil moisture was not a strong limitation and also under drought conditions. Variations in the spatial pattern of actual evaporation provided a useful indicator of drought across the Prairie region of Western Canada. The results contribute to a better understanding of the effects of spatial associations of key factors on evaporation estimates in a Prairie landscape. The methodology developed for distributing net radiation from assimilated visible and thermal images could potentially be used in regional scale modelling applications for improving evaporation estimates using point scale estimation techniques. The modelling algorithms applied to derive point estimates of evaporation from surface reference data may be useful for operational purposes that require estimates of actual evaporation and for characterizing drought.

At present one of the challenges of soil erosion research in South Africa is the limited information on the location of gullies. This is because traditional techniques for mapping erosion which consists of the manual digitization of gullies from air photos or satellite imagery, is limited to expert knowledge and is very time consuming and costly at a regional scale (50-10000km²). Developing a robust, reliable and accurate means of mapping gullies is a current focus for the Institute for Soil, Climate and Water Conservation (ISCW) of the Agricultural Research Council (ARC) of South Africa. The following thesis attempted to answer the question whether “medium resolution multi-spectral satellite observations, such as Landsat TM, combined with information extraction techniques, such as Vegetation Indices and multispectral classification algorithms, can provide a semi-automatic method of mapping gullies and to what level of accuracy?”. More specifically, this thesis investigated the utility of three Landsat TM-derived Vegetation Index (VI) techniques and three classification techniques based on their level of accuracy compared to traditional gully mapping methods applied to SPOT 5 panchromatic imagery at selected scales. The chosen study area was located in the province of KwaZulu-Natal (KZN) South Africa, which is considered to be the province most vulnerable to considerable levels of water erosion, mainly gully erosion. Analysis of the vegetation indices found that Normalized Difference Vegetation Index (NDVI) produced the highest accuracy for mapping gullies at the sub-catchment level while Transformed Soil Adjusted Vegetation Index (TSAVI) was successful at mapping gullies at the continuous gully level. Mapping of gullies using classification algorithms highlighted the spectral complexity of gullies and the challenges faced when trying to identify them from the surrounding areas. The Support Vector Machine (SVM) classification algorithm produced the highest accuracy for mapping gullies in all the tested scales and was the recommended approach to gully mapping using remote sensing

The advent of satellite limb scatter measurements has allowed the stratosphere to be studied at a scope unparalleled by previous observational techniques, affording the opportunity to study structures on both small spacial and temporal scales. Utilizing these measurements to their fullest has fueled the development of radiative transfer models to simulate the measurements, but also inversion techniques to retrieve atmospheric parameters. The limb scatter instrument OSIRIS, onboard the Odin satellite, is currently used in conjunction with the SASKTRAN radiative transfer model and multiplicative algebraic reconstruction technique to retrieve stratospheric aerosol extinction. In this work, the aerosol information content of limb scatter measurements is explored and an improved version of the aerosol retrieval is developed through the simultaneous retrieval of a second aerosol parameter, the Angstrom coefficient, which is related to particle size. The sensitivity of limb scatter measurements to aerosol is investigated through forward modelling of OSIRIS measurements as a function of wavelength, satellite geometry and particle size. Information content of the measurements is investigated to determine the feasibility of retrieving various aerosol size parameters and a simple linear inversion technique is tested. Results from this study are used to develop a non-linear inversion technique with minimal sensitivity to the required assumptions. Incorporation of longer wavelength data into the retrieval allows for the determination of the wavelength dependence of the scattered signal, which when combined with a lognormal particle size distribution of constant mode width allows for the retrieval of aerosol number density and mode radius. Conversion of these parameters to extinction and the Angstrom coefficient provides retrieved quantities with minimal dependence on the assumed size distribution. Application of this technique to the OSIRIS data set shows improved extinction results through both internal comparisons of the data and when compared with other results from SAGE II, III and CALIPSO satellite measurements. Although the retrieved Angstrom coefficient shows some bias due to the required assumptions, comparisons with the SAGE II data set show considerable improvement over the apriori estimate.

At present one of the challenges of soil erosion research in South Africa is the limited information on the location of gullies. This is because traditional techniques for mapping erosion which consists of the manual digitization of gullies from air photos or satellite imagery, is limited to expert knowledge and is very time consuming and costly at a regional scale (50-10000km²). Developing a robust, reliable and accurate means of mapping gullies is a current focus for the Institute for Soil, Climate and Water Conservation (ISCW) of the Agricultural Research Council (ARC) of South Africa. The following thesis attempted to answer the question whether “medium resolution multi-spectral satellite observations, such as Landsat TM, combined with information extraction techniques, such as Vegetation Indices and multispectral classification algorithms, can provide a semi-automatic method of mapping gullies and to what level of accuracy?”. More specifically, this thesis investigated the utility of three Landsat TM-derived Vegetation Index (VI) techniques and three classification techniques based on their level of accuracy compared to traditional gully mapping methods applied to SPOT 5 panchromatic imagery at selected scales. The chosen study area was located in the province of KwaZulu-Natal (KZN) South Africa, which is considered to be the province most vulnerable to considerable levels of water erosion, mainly gully erosion. Analysis of the vegetation indices found that Normalized Difference Vegetation Index (NDVI) produced the highest accuracy for mapping gullies at the sub-catchment level while Transformed Soil Adjusted Vegetation Index (TSAVI) was successful at mapping gullies at the continuous gully level. Mapping of gullies using classification algorithms highlighted the spectral complexity of gullies and the challenges faced when trying to identify them from the surrounding areas. The Support Vector Machine (SVM) classification algorithm produced the highest accuracy for mapping gullies in all the tested scales and was the recommended approach to gully mapping using remote sensing

Land surface evaporation has considerable spatial variability that is not reflected in meteorological station data alone. Knowing the spatial variability of evaporation is important for describing drought, managing agricultural land, and is valuable for improving the parameterization of hydrological models and land surface schemes over large areas. General difficulties arise for obtaining reliable, spatially distributed evaporation estimates as a result of uncertainty in estimation techniques, scale issues and complexities regarding land surface and atmospheric interactions, and the spatial and temporal variability of key factors governing the evaporation process. Estimating evaporation is further complicated when soil moisture becomes a critical limitation, particularly during drought. An examination of the spatial variability of evaporation and its association with governing factors was conducted in Prairie landscapes using three modelling techniques. First, eddy covariance measurements and reference meteorological data were obtained at two Prairie locations to assess the accuracy of physically-based models for calculating point estimates of actual evaporation under non-limited soil moisture conditions and during drought. Second, estimates of actual evaporation were distributed at the field scale in order to examine the impacts of driving factors and their spatial associations on upscaled evaporation estimates. This required the assimilation of high resolution visible and thermal images which were used to derive estimates of surface albedo and surface emitted longwave radiation. These were combined along with surface reference observations to develop an index of the mid-day radiation in order to distribute a known value of mean daily net radiation over the field. Third, archived historical climate data were used as input for a continuous hydrological simulation to examine spatial and temporal variations in evaporation across the Prairie region of Western Canada during a drought and non-drought period. Results of this research showed that the spatial variability of evaporation could be derived at the field scale by integrating remote sensing and surface reference climate data with a physically-based evaporation model. Surface temperature and soil moisture, and net radiation were found to be highly variable spatially at field scales whilst meteorological conditions tended to be less variable spatially but showed strong temporal variability. At the field scale it was found that the variability in albedo and surface temperature were both important for characterizing differences in surface state conditions. Their combined influence was reflected in the resulting pattern of net radiation that governed the distribution of actual evaporation estimates obtained with the Granger and Gray evaporation model. It was found that an areal estimate of evaporation obtained from the means of driving factors was similar to the areal average obtained from the distributed estimates. This was attributed to the offsetting interactions among the driving factors which effectively reduced the variability of the model estimates. In general, the physically-based models examined were found to provide reasonable estimates of actual evaporation when driven by observations at point-scales over multi-day and seasonal periods. This included periods when soil moisture was not a strong limitation and also under drought conditions. Variations in the spatial pattern of actual evaporation provided a useful indicator of drought across the Prairie region of Western Canada. The results contribute to a better understanding of the effects of spatial associations of key factors on evaporation estimates in a Prairie landscape. The methodology developed for distributing net radiation from assimilated visible and thermal images could potentially be used in regional scale modelling applications for improving evaporation estimates using point scale estimation techniques. The modelling algorithms applied to derive point estimates of evaporation from surface reference data may be useful for operational purposes that require estimates of actual evaporation and for characterizing drought.

This dissertation consists of three parts, each devoted to a particular issue of significant importance for satellite-based remote sensing of cirrus clouds. In the first part, we develop and present a fast infrared radiative transfer model on the basis of the adding-doubling principle. The model aims to facilitate the radiative transfer computations involved in hyperspectral remote sensing applications. The model is applicable to a variety of cloud conditions, including vertically inhomogeneous or multilayered clouds. It is shown that for hyperspectral applications the model is two order-of-magnitude faster than the well-known discrete ordinate transfer (DISORT) model, while maintains a similar accuracy. The second part is devoted to the investigation of uncertainties in the FSSP (Forward Scattering Spectrometer Probe) measurement of cloud extinction by small ice particles. First, the single-scattering properties of small ice particles in cirrus clouds are derived and compared to those of equivalent spheres according to various definitions. It is found that, although small ice particles in cirrus clouds are often “quasi-spherical”, their scattering phase functions and asymmetry factors are significant different from those of ice spheres. Such differences may lead to substantial underestimation of cloud extinction in FSSP measurement, if small ice particles are assumed to be spheres. In the third part, we present a comparison of cirrus cloud optical thickness retrievals from two important instruments, MODIS (Moderate Resolution Imaging Spectrometer) and POLDER (Polarization and Directionality of Earth’s Reflection), on board NASA’s A-train satellite constellation. The comparison reveals a large difference. Several possible reasons are discussed. It is found that much of the difference is attributable to the difference between the MODIS and POLDER retrieval algorithm in the assumption of cirrus cloud bulk scattering properties. Potential implications of the difference for climate studies are investigated. An important finding is that the use of an unrealistic cirrus bulk scattering model might introduce artificial seasonal variation of cirrus optical thickness and shortwave radiative forcing into the retrieval.