RADAR REMOTE SENSING FOR RETRIEVAL OF SOIL PARAMETERS

Abstract

Soil moisture, surface roughness and soil texture are important soil parameters in various applications, such as, agriculture, weather forecasting, soil erosion, hydrological studies and flood and draught prediction. Studies in the field of active microwave remote sensing have shown the feasibility to monitor these soil parameters with ground based, airborne and spaceborne sensors. Therefore, this thesis has two main objectives. First main aim is to analyze soil parameters, specifically soil texture which is less reported in bistatic domain because it is well known that bistatic domain provides several flexibilities over monostatic domain and very lesser experimental studies have been carried out to characterize the soil parameters in this domain. Second aim is to retrieve the soil moisture in vegetated area with SAR and optical satellite data for minimizing the use ofapriori information. Thesis has been composed of seven chapters. First chapter deals with the introduction, experiment performed, data used and study area. The second chapter presents the brief literature review, discussing the studies carried out in the field of microwave remote sensing for the soil parameters retrieval and its limitations along with the need of research in the present scenario. Chapter 3 analyzes the effect of soil texture on specular scattering coefficient at 10 GHz (X-band) in both like polarizations, i.e., HH- and VV-polarization for various incidence angles (i.e., 30° to 70° in step of 10°). If ((9, <p) (where, 9 and (p are the incidence angle and azimuthal angle respectively) define the incident direction of the transmitted power Pp(6, (p) at polarization p, and (<9,, <ps) is the direction of the received power Pq{6s, (ps) at polarization q then specular scattering is defined as: 6 = 9S, cp = 0 and <ps = 0. To study the soil texture effect on the specular scattering coefficient, four different soil texture fields were prepared on the basis of variation in soil constituent, i.e., the percentage of sand (i.e., 78.14% to 28.72%), silt (i.e., 33.16% to 9.12%) and clay (i.e., 18.75% to 4.32%) was changed for different fields. The major changes were observed for sand constituent. It is observed from the study of dry and smooth soil that by changing soil texture there is a considerable amount of change in specular scattering coefficient for both like polarizations. The effect was more prominent at higher incidence angles (i.e., > 50°), i.e., better differentiation in specular scattering coefficient for different soil texture field was observed at higher incidence angles. The dynamic range of specular scattering coefficient with incidence angle changes with the change is sand constituent in soil. The dynamic range of 8.7 dB and 10.5 dB was observed for higher sand content (i.e., 78.14%) and lower sand content (i.e., 28.72%), respectively for dry and smooth soil in HH-polarization whereas in case of VV-polarization dynamic range was 18.2 dB and 19.4 dB respectively. Further, the effect of soil texture on specular scattering coefficient was examined in the presence of four different soil moisture contents (i.e., dry to 0.21 cm cm"3) and three different periodic surface roughness conditions (i.e., smooth to 1.4 cm). Thus in order to check the soil texture effect, 48 different field conditions were considered for the observations of specular scattering coefficient in both like polarizations at various incidence angles. The effect of soil texture on specular scattering has been clearly observed at different soil moisture condition. A better differentiation in specular scattering coefficient was observed for change in soil texture with the increase in moisture content though the effect is more prominent at higher incidence angles. Further, the effect of soil texture on specular scattering coefficient has been observed in the presence of different periodic surface roughness conditions. The observation has shown the difficulty in analyzing the effect of soil texture on specular scattering at higher roughness values (i.e., 0.9cmand 1.4 cm). Chapter 4 analyzes the effect of soil texture on microwave specular scattering at 6 GHz (C-band) in both like polarizations, i.e., HH- and VV-polarization. A more regress analysis was carried out because it is observed that soil constituents with roughness and moisture are giving more effect on specular scattering coefficient at 6 GHz than 10 GHz. So, ten different soil texture fields were prepared and to check the angular behavior, the incidence angle was changed from 25° to 70° degree in step of 5°. The percentage of sand (i.e., 85.3% to 2.3%), silt (i.e., 70.6% to 7.5%) and clay (i.e., 81.6% to 2.5%) was changed for different soil texture fields. Six soil moisture and five periodic roughness conditions were studied to get an understanding of the soil texture effect on specular scattering in presence of various moisture contents and periodic surface roughness conditions. Therefore, total 300 different field conditions were analyzed. The study at 6 GHz is important as the longer wavelength is relatively less sensitive to the roughness variation in comparison to the shorter wavelength. Although the study of this chapter is concentrated for the bare soil only, it may be tested for the vegetated area as a future work. The analysis part is subdivide, firstly analyzing the effect of soil texture on specular scattering coefficient in the presence of various soil moisture contents (i.e., 0.027 cm3 cm"3 to 0.425 cm3 cm"3) by keeping the periodic surface roughness condition constant. Secondly, the effect of soil texture on specular scattering coefficient while varying the periodic surface roughness (i.e., 0.43 cm to 2.46 cm) was studied. Furthermore, the combined effect of soil moisture and periodic surface roughness on specular scattering coefficient for change in soil texture has been considered. The observation in HH-polarization with dry and smooth field has shown that the dynamic range of specular scattering coefficient with high sand content (i.e., 85.3% sand) was 9.8dB whereas in case of high clay content (i.e., 81.6% clay) the dynamic range was 13.6 dB. The differentiation among different soil texture field based on specular scattering coefficient can be made when the volumetric soil moisture content is less 0.368 cm3 cm"3 whereas, it is difficult to observe the soil texture effect for higher soil moisture contents (i.e., > 0.368 cm3 cm"3) in both like polarizations. Further, the surface roughness does not exhibit such kind of effect and even at higher periodic roughness value (i.e., < 2.46 cm) the change in specular scattering coefficient can be observed with the change in soil texture at 6 GHz. The experimental observations carried out at 6 GHz points out that higher incidence angles (i.e., > 45°) are more appropriate than the lower incidence angles for the study of soil texture in presence of soil moisture and periodic surface roughness in both like polarizations. Chapter 5 deals with the retrieval of soil parameters at 6 GHz in association to the observation made in specular direction. Specular scattering data at 6 GHz has been considered for soil parameters retrieval studies instead of specular scattering data at 10 GHz due to its response for different soil texture field in the presence of periodic surface roughness whereas, in case of 10 GHz it has been very difficult to distinguish among different soil texture fields based on specular scattering coefficient in the ix presence of periodic surface roughness. Soil parameters retrieval in domain of microwaveremote sensing is a challenging task owing to the dependence of many soil parameters on one scattering coefficient value only. By the virtue of this, more sensor parameters were utilized to minimize the effect of one or more soil parameters on specular scattering coefficient, in consequence of which the desired soil parameters can be retrieved easily and more accurately. In the first approach, copolarization ratio has been utilized for minimizing the soil texture effect in order to retrieve the volumetric soil moisture content. The periodic surface roughness condition during the observation was kept constant whereas, the changes in soil texture and soil moisture were considered. There was negligible changes in copolarization ratio with the change in soil texture whereas, copolarization ratio changes with the change in soil moisture. In accordance to the observations obtained with the copolarization ratio an empirical relation between the copolarization ratio and volumetric soil moisture was developed. The volumetric soil moisture content was retrieved with developed empirical relationship along with the Kirchhoff scalar approximation (SA) to draw the comparison. The RMSE for soil moisture retrieval was 0.021, 0.079 and 0.095 for developed empirical relationship, SA in HH-polarization and VV-polarization, respectively. The obtained results clearly signify that the developed empirical relationship performed better than SA for retrieving the bare soil moisture. In next approach, multi-incidence data have been used to retrieve the soil surface roughness. These surface roughness values were subsequently used to retrieve the soil moisture and soil texture. Ratio of specular scattering coefficient at two different incidence angles provides the normalized specular scattering coefficient which depends on surface roughness and shows negligible dependency on soil moisture and soil texture. This property led to the development of a relationship between normalized specular scattering coefficient and roughness parameters, rms surface height, s and correlation length, /. The developed empirical relationship in conjunction with SA and empirical relationship developed by Hallikainen et al. (1985) has been utilized for the retrieval of roughness parameters, soil moisture and soil texture. The retrieved results were in good agreement with the ground truth data. Root mean square error (RMSE) for the retrieval of rms surface height, correlation length, volumetric soil moisture, percentage of sand, and percentage of clay were 0.027, 0.051, 0.036, 5.94, and 8.15, respectively. The developed approach reduces the need of apriori information which is required in form of surface roughness characterization when the objective is to retrieve the soil moisture and soil texture. Till now we have concentrated our work for analyzing soil parameters in bistatic domain but still satellite data for bistatic domain is awaited as TanDEM-X is inprogress. The satellite data which is available are for the case of backscattering. So, we have considered the satellite data such as PALSAR and MODIS for retrieval of soil moisture in vegetated area. The retrieval of vegetation covered soil moisture is a challenging task with the existing SAR sensors. Though, the backscattering coefficient contains the information of vegetation as well as underlying soil moisture, the complex scattering behavior of microwave form vegetated area give rise to the development of complex relationship among backscattering coefficient, vegetation parameter (e.g., leaf area index, biomass, plant height) and soil parameters (i.e., soil moisture, surface roughness). The optical data which can characterize the vegetation may be efficiently utilized in vegetation covered soil moisture retrieval algorithm with SAR data. The aim of the chapter 6 is to retrieve the soil moisture in vegetated area with minimum apriori information by using satellite data like the ALOS-PALSAR (Advanced Land Observation Satellite- Phased Array type L-band Synthetic Aperture Radar), a SAR data and MODIS (Moderate-resolution Imaging Spectroradiometer), an optical data. The PALSAR image is full polarimetric, L-band (1.27 GHz) image and was acquired on April 06, 2009. The MODIS data used is of band 1(620-670 nm) and band 2 (841-847 nm) and the date of image acquisition was April 03, 2009. The area of study was Roorkee city (Uttarakhand, India) and its surroundings. The first set of images (one PALASR and one MODIS) that lie between longitudes 77.803° Eand 77.980° Eand latitudes 30.000° Nand 29.823° Nwere used for the algorithm development and subsequently testing the developed algorithm for soil moisture retrieval. The second set of images that were used for validating the algorithm lie between longitudes 77.847° E and 78.024° E and latitudes 29.859° N and 29.682° N. The date of acquisition of PALSAR and MODIS images were April 6, 2009 and April 3, 2009, respectively. The land cover is fairly flat and mostly consists of the urban, water and agriculture classes. It is always the possibility to find the mixed land cover classes therefore, as a first step PALSAR data was used to classify the land cover in urban, water, vegetated land and bare soil by utilizing knowledge

based approach involving the various polarizations (linear, circular, linear 45°, coand cross-polarized ratios for both linear and circular polarization) and subsequently the urban and water region can be masked. The objective of this chapter is to analyze the feasibility of the relating the information available from SAR data and optical data to envisage such an approach that mostly rely on the information content of satellite image and require minimum apriori information. The concept of such an approach arises as the vegetation can be modeled through the SAR data as well as the optical data. In case of SAR data the backscattering is affected by the vegetation cover and contains the information regarding vegetation whereas, the normalized index vegetation index (NDVI) provides a good estimate of the crop cover. The utilization of the information content from the optical data reduces the requirement of apriori information which is required in the vegetation parameters characterization. Therefore, two different normalizing approaches have been used for scattering coefficient of PALSARdata and then the empirical relationships have been developed with NDVI to incorporate the vegetation effect in soil moisture retrieval. In first approach, scattering coefficient of image is normalized with the scattering coefficient of bare soil (calculated with in situ observation of moisture and roughness) making the normalized data a function of vegetation cover which is represented by the NDVI. A quadric relationship was observed between the normalized scattering coefficient data and NDVI with coefficient of determination (R2) 0.83. This quadric relationship sets the basis for the retrieval of soil moisture with the help of NDVI and normalized scattering coefficient from the PALSAR image. The root mean square error (RMSE) is 0.036 and 0.041 for retrieval of soil moisture when the algorithm is applied on test image and validating image, respectively. In the second approach, scattering coefficient of PALSAR data was normalized with the scattering coefficient of the dry soil providing the normalized data as a function of the soil moisture and vegetation cover defined by the NDVI. The relationship between normalized scattering coefficient and NDVI was explored which provide a set of lines defining the different range ofsoil moistures. The coefficient ofdetermination (R2) for all regressed lines was greater than 0.81. Hence, the soil moisture is retrieved with the developed empirical relationship between the normalized scattering coefficient of PALSAR data andNDVI. The root mean square error(RMSE) is 0.039 and 0.052 for retrieval of soil

moisture when the algorithm is applied on test image and validating image, respectively. Finally chapter 7 draws the conclusion and contribution made in the thesis as well as presents the future work of the study carried out in this thesis.