Hosni Ghedira

Effect of Land Cover Heterogeneity on Soil Moisture Retrieval Using Active Microwave Remote Sensing Data

This study addresses the issue of the variability and heterogeneity problems that are expected from a sensor with a larger footprint having homogenous and heterogeneous sub-pixels. Improved understanding of spatial variability of soil surface characteristics such as land cover and vegetation in larger footprint are critical in remote sensing based soil moisture retrieval. This study analyzes the sub-pixel variability (standard deviation of sub- grid pixels) of Normalized Difference Vegetation Index and SAR backscatter. Back- propagation neural network was used for soil moisture retrieval from active microwave remote sensing data from Southern Great Plains of Oklahoma. The effect of land cover heterogeneity (number of different vegetation species within pixels) on soil moisture retrieval using active microwave remote sensing data was investigated. The presence of heterogeneous vegetation cover reduced the accuracy of the derived soil moisture using microwave remote sensing data. The results from this study can be used to characterize the uncertainty in soil moisture retrieval in the context of Soil Moisture Active and Passive (SMAP) mission which will have larger footprint.

Non-parametric Methods for Soil Moisture Retrieval from Satellite Remote Sensing Data

Satellite remote sensing observations have the potential for efficient and reliable mapping of spatial soil moisture distributions. However, soil moisture retrievals from active microwave remote sensing data are typically complex due to inherent difficulty in characterizing interactions among land surface parameters that contribute to the retrieval process. Therefore, adequate physical mathematical descriptions of microwave backscatter interaction with parameters such as land cover, vegetation density, and soil characteristics are not readily available. In such condition, non-parametric models could be used as possible alternative for better understanding the impact of variables in the retrieval process and relating it in the absence of exact formulation. In this study, non-parametric methods such as neural networks, fuzzy logic are used to retrieve soil moisture from active microwave remote sensing data. The inclusion of soil characteristics and Normalized Difference Vegetation Index (NDVI) derived from infrared and visible measurement, have significantly improved soil moisture retrievals and reduced root mean square error (RMSE) by around 30% in the retrievals. Soil moisture derived from these methods was compared with ESTAR soil moisture (RMSE ~4.0%) and field soil moisture measurements (RMSE ~6.5%). Additionally, the study showed that soil moisture retrievals from highly vegetated areas are less accurate than bare soil areas.

Sea-ice monitoring over the Caspian Sea using geostationary satellite data

A new technique is proposed for sea-ice mapping using observations from geostationary satellite over the Caspian Sea. A two end-member linear-mixture approach has been employed. A neural-network-based approach was used to simulate water and ice reflectances for all possible sun–satellite geometries. The ice-mapping technique incorporates an advanced cloud-detection algorithm with adaptive threshold values. The average percentage of cloud reduction because of the daily compositing ranged from 22% to 25%. Daily maps of ice distribution and concentration with minimal cloud coverage were produced for the winter seasons of 2007 and 2008. The retrieved ice distribution demonstrated a good agreement with Moderate Resolution Imaging Spectroradiometer (MODIS) images and National Oceanic Atmospheric Administration (NOAA) Interactive Multisensor Snow and Ice Mapping System (IMS) snow and ice charts. The obtained correlation coefficients with IMS charts for 2007 and 2008 were 0.92 and 0.83, respectively. The technique has been proposed as one of the candidate ice-mapping techniques for the future Geostationary Operational Environmental Satellite-R Series (GOES-R) Advance Baseline Imager (ABI) instrument.

Soil Moisture Retrieval from Radarsat Data: A Neuro-Fuzzy Approach

The spatial dynamic of soil moisture is generally affected by the variation in soil surface characteristics such as: land cover, vegetation density, soil texture, and soil material. The mapping of soil moisture by remote sensing tools has several advantages over the conventional field measurement techniques especially in the case of heterogeneous landscapes. The use of microwave remote sensing offers fast and reliable ways in mapping the spatial distribution of soil moisture. Microwave remote sensing systems is used to measure soil moisture on the basis of a large contrast that exists between the dielectric constant values for dry and wet soils. This study describes the use of non-parametric classifiers for the soil moisture retrieval from active microwave remote sensing data. The difficulty facing the soil moisture retrieval process is due, in large part, to the lack of a precise mathematical description of the observed land cover parameters and to the extent of their variability. Two non-parametric techniques have been used: Neural Networks and Fuzzy Logic. Different configurations of these two classifiers have been tested and compared by assessing the accuracy of soil moisture.

The Effect of Vegetation Cover on Snow Cover Mapping from Passive Microwave Data

Snow-cover parameters are being increasingly used as inputs to hydrological models. Having an accurate estimation of the snow cover characteristics during the snowmelt season is indispensable for an efficient hydrological modeling and for an improved snowmelt runoff forecasts. In this paper, we used an adaptive neural network system to generate the spatial distribution of snow accumulation from multi-channel SSM/I data in the Northern Midwest of the United States. Five SSM/I channels were used in this experiment (19H, 19V, 22V, 37V, and 85V). Three snow days with high snow accumulation and no precipitation have been selected during the 2001/2002 winter season to train and test the neural network system. Snow depth measurements have been collected from the National Climatic Data Center (NCDC) through the Cooperative Observer Network for the U.S. snow Monitoring. The snow depths have been compiled and gridded into 25 km times 25 km grid to match the final SSM/I resolution. Different vegetation-related parameters (NDVI, optical depth, homogeneity) have been collected and gridded over the study area. The current results have shown a significant effect of vegetation cover properties on the mapping accuracy. Furthermore, the addition of vegetation related information to the mapping process has shown to have a positive impact on mapping performance, especially for areas with shallow snow cover (less than 5 cm)

Robust neural network system design for detecting and estimating snowfall from the Advanced Microwave Sounding Unit

The principal intent of this research is to: (a) investigate the potential of passive microwave data from Advanced Microwave Sounding Unit (AMSU) in detecting snowfall events and in measuring their intensity, and (b) evaluate the effect of both land cover and atmospheric conditions on the retrieval accuracy. A neural-network-based model has been developed and has shown a great potential in detecting snowfall events and classifying their intensity into light, moderate or heavy. This algorithm has been applied for different snow storms which occurred in four winter seasons in the Northeastern United States. Additional information such as cloud cover and air temperature were added to the process to reduce misidentified snowfall pixels. Only pixels with cloud cover and falling within a specific range of temperature are presented to the snowfall detection model. Surface temperature collected from ground station-based observations and archived by the National Climatic Data Center (NCDC) were used for this test. Different heavy storm events and non-snowfall observations that occurred at the same time as AMSU acquisition were selected. Hourly snow accumulation data collected by the NCDC were used as truth data to train and validate the model. The preliminary results indicate that the neural-network-based model provides a significant improvement in snowfall detection accuracy over existing satellite-based methods.

Improvement in Estimating Snowpack Properties with SSM/I Data and Land Cover Using Artificial Neural Networks

The goal of this study is to develop an algorithm to estimate snow water equivalent (SWE) in Great Lakes area based on a three year time series of SSM/I data along with corresponding ground truth data. An assortment of SSM/I EASE-GRID pixels was selected for time series analysis. The pixels were selected based on the amount of snow, latitude, and land cover. Two types of ground truth data were used: 1 - point-based snow depth observations from NCDC; 2 - grid based SNODAS-SWE dataset, produced by NOHRSC. To account for land cover variation in a quantitative way a NDVI was used. To do the time series analysis, three scattering signatures of GTVN (19V-37V), GTH (19H-37H), and SSI (22V-85V) were derived. The analysis shows at lower latitudes of the study area there is no correlation between GTH and GTVN vs snow depth. On the other hand SSI shows an average correlation of 75 percent with snow depth in lower latitudes makes it suitable for shallow snow. In model development a multi-linear algorithm was defined to estimate SWE using NDVI values along with the location of the pixels as classification criteria. The results show up to 60 percent correlation between estimated and ground truth SWE

EFFECT OF SUB-PIXEL VARIABILITY AND LAND-COVER ON SOIL MOISTURE RETRIEVAL FROM RADARSAT-1 DATA

The dynamic of soil moisture is generally affected by the spatial variation in soil surface characteristics such as land cover, vegetation density, soil texture, and soil material. The main purpose of this project is to develop neural network algorithm for soil moisture retrieval from active microwave data. A back-propagation neural network has been used to estimate the soil moisture from Synthetic Aperture Radar data. Soil moisture data with a spatial resolution of 800 m acquired during the SGP97 campaign, were used as truth data in the training and the validation processes. In addition to backscatter values retrieved from RADARSAT-1 image, normalized difference vegetation index (NDVI), land cover and soil texture have been added as an input to neural network algorithm. The effects of sub-pixels variability of the NDVI and land cover type on the retrieval of soil moisture have been investigated by comparing the measured and the predicted soil moisture. Further, all training and validation pixels (800 m resolution) have been labeled as either homogeneous or heterogeneous based on the occurrence of the same land cover type. The results showed that, homogeneous pixels are more likely to have better accuracy than heterogeneous pixels in soil moisture classification. A better correlation between soil moisture and SAR backscattering was found in areas with high soil moisture content, where the surface wetness dominated the vegetation contribution to the radar backscatter

An adaptive neural network system for improving the filtration of non-snow pixels from SSM/I images

Snow-cover parameters are being increasingly used as input to hydrological models. Having an accurate estimation of snow cover characteristics during the snowmelt season is indispensable for an efficient hydrological modeling and snowmelt runoff forecasting. Direct measurements of snow depth at a single station are generally not very useful in making estimates of distribution over large areas, since the measured depth may be highly unrepresentative of the neighboring area because of drifting or blowing. Additionally, the traditional field sampling methods and the ground-based data collection are often very sparse, time consuming, and expensive compared to the coverage provided by remote sensing techniques. At present, most of hydrological models that require snowpack information are using maps obtained by gridding standard point gauge measurements or derived from physically/statistically-based models. In this paper, we used an adaptive neural network system to generate the spatial distribution of snow accumulation from multi-channel SSM/I data in the Northern Midwest of the United States. Five SSM/I channels were used in this experiment (19H, 19V, 22V, 37V, and 85V). Three snow days with high snow accumulation and no precipitation have been selected during the 2001/2002 winter season. Snow depth measurements have been collected from the National Climatic Data Center (NCDC) through the Cooperative Observer Network for the U.S. snow Monitoring. The snow depths have been compiled and gridded into 25 km x 25 km grid to match the final SSM/I resolution.

A neural netwotk based approach for multi-spectral snowfall detection and estimation

The principal intent of this research is to: (a) investigate the potential of passive microwave data from AMSU in detecting snowfall events and in measuring their intensity, and (b) evaluate the effect of both land cover and atmospheric conditions on the retrieval accuracy. Additional information such as cloud cover and air temperature were added to the process to reduce misidentified snowfall pixels. Two products retrieved from AMSU brightness temperatures were used to estimate surface temperature and to map the snow cover extent. In this project, a neural-network-based model has been developed and has shown a great potential in detecting and estimating the intensity of snowfall events. This algorithm has been applied for different snow storms occurred between 2002 and 2003 in three different locations in the North-East of United States. These locations were selected because of the high amount of snowfall every year. Only pixels with cloud cover and falling within a specific range of temperature are presented to the snowfall detection model. Surface temperature collected from ground station-based observations archived by the National Climatic Data Center (NCDC) were used for this test. Different heavy storm events and non-snowfall observations that occurred at the same time as AMSU acquisition were selected. Hourly snow accumulation data collected by the NCDC were used as truth data to train and validate the model. To reduce the risk of erroneous identification of snowfall pixels, only storms lasting more than three hours were selected. Such criteria will undoubtedly increase the level of confidence that snowfall coincides with AMSU acquisition time. The neural network based snowfall product was compared with the snowfall detection algorithm over land developed in 2003 by Kongoli et al [1]. The preliminary results indicate that the neural-network-based model provides a significant improvement in snowfall detection accuracy over existing satellite-based methods.