The thesis research was devoted to the development of the radar and optical data fusion technique for land surface physical parameters restoration.
The research object was the land surface physical parameters and their display on Earth observation data.
The research subject was multi-sensor Earth observation data fusion algorithms.
The aim of the study was to improve the accuracy of the land surface physical parameters estimation, especially soil moisture, by developing the new technique for the multi-sensor radar and optical satellite data fusion.
In the framework of the thesis research, Earth observation (EO) data retrieved in several spectral ranges were used to estimate the land surface parameters. In particular, the C-SAR Sentinel-1 A/B GRDH products were used to calculate land surface permittivity and roughness based on Integral Equation Model (IEM). The land surface temperature was estimated based on the inverse Planck’s equation for «grey» bodies, by using the multispectral EO data in Visible/Near Infrared (PlanetSope PS2.SD) and Thermal Infrared (Landsat-7 ETM+, Landsat-8 OLI та EOS MODIS) spectral ranges. Geometric parameters were calculated based on the ALOS AW3D digital terrain elevation model.
The physical and geometric parameters obtained were used as a components of the model developed for the multi-sensor EO data fusion based on the multidimensional regression analysis. The accuracy of the model was verified by an example of the soil moisture estimation. For this purpose, a soil samples were selected within test sites and the reference soil moisture were calculated by the thermostat-weight method. The multi-dimensional regression analysis of the reference and model soil moisture dependence allowed to conclude that the developed model provides the high accuracy of computation, as indicated by the determination coefficient of 0.84 and an root mean square error of 4.37 % (N = 96).
As a result, the new technique was developed for multispectral optical and double polarization radar satellite data fusion for the land surface physical parameters, soil moisture especially, estimation. Unlike existing approaches, the developed technique uses the complex linearized multidimensional regression model with absolute deviations minimization, the original approach for the land surface temperature estimating for non-synchronous radar and optical observations, and additionally relies on the radar signal local fluctuations and relief heterogeneity.
The linearized multidimensional regression model with absolute deviations minimization was proposed. Unlike existing models, the developed model relies on the number of physical parameters, e.g. backscattering coefficients, permittivity, roughness, land cover temperature, vegetation cover parameters, and additional geometric parameters, e.g. the relief elevation, slope, aspect and orthogonal concave, radar signal local fluctuations and mutual orientation of the relief element to a sensor.
A well-known method for the land surface permittivity estimation was improved by using the dual polarization radar sensing based on the IEM calibration. Additionally, the model was improved by the computer assisted soil roughness estimation algorithm implementation. In addition, the theoretical criteria for the function rupture detection were proposed based on the soil permittivity and roughness tolerance ranges. The theoretical criteria for the double polarization radar images filtering were also proposed based on the permittivity and roughness tolerance ranges, which can identify the model disruption cases before estimations.
The well-known method for the land surface temperature estimation, by using visible, near and thermal infrared multispectral optical data, based on Planck’s law was improved. In particular, the temperature obtained remotely was recalculated to a temperature at the time of radar sensing.
