This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).
One of the main strengths of active microwave remote sensing, in relation to frequency, is its capacity to penetrate vegetation canopies and reach the ground surface, so that information can be drawn about the vegetation and hydrological properties of the soil surface. All this information is gathered in the so called backscattering coefficient (σ^{0}). The subject of this research have been olive groves canopies, where which types of canopy biophysical variables can be derived by a specific optical sensor and then integrated into microwave scattering models has been investigated. This has been undertaken by means of hemispherical photographs and gap fraction procedures. Then, variables such as effective and true Leaf Area Indices have been estimated. Then, in order to characterize this kind of vegetation canopy, two models based on Radiative Transfer theory have been applied and analyzed. First, a generalized two layer geometry model made up of homogeneous layers of soil and vegetation has been considered. Then, a modified version of the Xu and Steven Water Cloud Model has been assessed integrating the canopy biophysical variables derived by the suggested optical procedure. The backscattering coefficients at various polarized channels have been acquired from RADARSAT 2 (Cband), with 38.5° incidence angle at the scene center. For the soil simulation, the best results have been reached using a Dubois scattering model and the VV polarized channel (r^{2} = 0.88). In turn, when effective LAI (LAI_{eff}) has been taken into account, the parameters of the scattering canopy model are better estimated (r^{2} = 0.89). Additionally, an inversion procedure of the vegetation microwave model with the adjusted parameters has been undertaken, where the biophysical values of the canopy retrieved by this methodology fit properly with field measured values.
Microwave remote sensing has increasingly been used in the Earth Sciences, such as Environment, Hydrology, Agriculture, Forestry, Geology,
In the microwave domain, the study of the vegetation canopy requires one to evaluate separately the electromagnetic behavior of soil and vegetation layers, as well as the specific geometric properties of their constituents. Soil surface scattering models need to take into account statistical roughness parameters, such as profile height displacement standard deviation and correlation length, in addition to the dielectric properties that are responsible of the soil reflectivity properties [
Furthermore, for describing the vegetation layer, it is mandatory to define a geometrical model, as well as to define also the electromagnetic properties of its constituents. For this purpose, a certain number of models such as those based on Radiative Transfer (RT) theory have been developed [
A fundamental aim of this research was to relate vegetation biophysical variables acquired by an optical sensor to radar backscatter. In order to fulfill these prerequisites, it has been considered that CBand frequency of RADARSAT 2 (5.5 GHz − λ = 5.6 cm) was the most appropriate for studying these processes, due to the similarities of the leaf dimensions of the observed vegetation canopy compared with the Radar system wavelength. In addition, full polarimetric mode has been selected as reference imagery in order to investigate the sensitivities of the different polarized channels to the effect of the canopies geometries. In this respect, a key point of this research is to differentiate between ‘Effective LAI’ and ‘True LAI’, since they express different concepts of leaf distribution within the canopy. Methods for determining these two variables are described by the ‘Gap Fraction Theory’ [
This paper is organized as follows. In Section 2, the study area and the motivations that have led to this particular study are described. Additionally, details on the SAR data used for this work are also given, which are followed by a list of field materials and methods employed in this research. The scattering models used in this work for extracting and processing physical and biophysical variables of the canopy, are also presented there. Then, the results reached at each stage of this research are presented and discussed (Section 3). Finally, some conclusions are drawn from the suggested methodology and achieved results.
The research area is located in central Spain, near the city of Madrid, on the Southern part of a geomorphologic formation named ‘Páramo de la Alcarria’, demarcated by the Tajuña and Jarama rivers basins (
The most abundant landcover and landuse types in this region can be classified into agricultural land, with irrigated crops (corn, alfalfa,
Natural land is generally made up of typical Mediterranean vegetation like shrubs and bushes, as well as Mediterranean oaks. Large coniferous extensions of
In this respect, the performances of new SAR systems must then be verified. For this purpose, within a large olive groves area, eight sampling units (SU) of approximatively 50 × 50 m, have been identified in order to retrieve hydrological and biophysical vegetation canopies reference values (
The SAR image used in this research has been acquired by the system onboard RADARSAT 2 which operates in full polarimetric mode, with three polarization states (HH, HV y VV) available; these have been processed as a single look complex (SLC) product. Given the specific pattern and architectural properties of olive groves, and the different vegetation constituent’s dimensions, the microwave frequency of this system (5.4 GHz—C Band), as well as its resolution and polarimetric capabilities, appear to be suitable for the aims of this research. The acquisition of this image has previously been programmed, so that soil field surveys and SAR image are temporally coincident. In order to have optimal climatic and soil moisture conditions, the early days of autumn have been chosen for this purpose. Satellite data have been finally acquired on 9 October 2008, with an incidence angle of 38.5° at the scene centre and a nominal resolution of 8 m in range and azimuth.
The available polarization states have been filtered with a ‘Lee speckle filter’ with a window width of 5 × 5 cells. Then the backscattering coefficients have been derived from the scattering matrix
Soil and vegetation measurements have been driven on the 8 sampling units (SU) introduced in Section 2.1, where the corresponding physical and biophysical parameters were respectively measured by a specific method and instrument.
Soil physical parameters comprise both, surface roughness and moisture content measurements. Each roughness measure has systematically been along a direction parallel to the nominal flying line of the satellite. A one meter length 5 × 5 cm gridded tablet has been employed for this task. Once it has been inserted into the soil surface, leveled and oriented, a picture has been recorded with a digital photographic device, so that this original position might be geometrically restored and the profile digitized by means of image processing techniques. This method enables to derive the statistical parameters of the soil surface such as the standard deviations of height displacements (σ) and correlation lengths (
Soil moisture sampling has been carried out by means of a metallic cylinder of 5 × 5 cm in diameter and in height, which has been in turn inserted into the first 5 cm of the soil surface. The extracted samples were held in a plastic bag with hermetic closure, and subsequently brought to the laboratory where they were weighted and dried in oven at 60 °C for 70 h. Then, each sample has been weighted again. Finally, 14 of these samples have been retained for successive texture analysis. All these samples have been used to derive soil moisture parameters such as gravimetric moisture, bulk density of the material and volumetric moisture. The latter, together with soil texture information have been finally used to compute the soil dielectric constant by means of the DobsonPeplinsky model [
Vegetation biophysical variables have been derived by means of hemispherical photographs. This process has been undertaken in each sampling unit where soil roughness and moisture have also been determined, so that these new values may be spatially consistent with the previous ones. This technique allows computing variables such as true and effective LAI’s, Average Leaf Angle (ALA) and fraction cover for individual or group of plants. Regarding acquisition conditions, some properties of the instrument and vegetation cover must also fulfilled [
The sampling methodology is based on the strategy proposed in [
Each set of photographs has been processed using CAN_EYE software version 1.4 [
There are two main theories for modeling vegetation canopies in the microwave domain which take into account soil and vegetation contributions separately,
Each microwave scattering model used for describing a vegetation canopy needs to incorporate a simulation of the soil surface beneath the vegetation layer, which is modeled as a dielectric surface. This task can be undertaken by means of many existing models. Amongst these, the Oh and Dubois surface models, referred to as semiempirical [
Generally, Radiative Transfer models for vegetation canopies take into account contributions of the soil surface (S) and vegetation (V) separately, as well as multiple interactions between them, which are produced between soils, trunks (Tr) and/or primary branches. However, for the sake of simplicity, within the framework of this work, these interactions have not been considered, and as they are not considered to be a dominant term in the copolarized returns [
Therefore, according to this principle the total backscattering coefficient will be expressed by:
A first characterization the vegetation canopy has been undertaken using the Rayleigh vegetation backscattering model (
For a continuous set of observation incidence angles, this expression enables one to compute a raster layer representing the soil contribution of the observed canopies using known values of transmissivity and albedo. This is also the first step for extracting soil moisture content values (m_{v}), which can also be carried by means of Oh and/or Dubois inversion procedures. This process requires that both polarizartion states (
The second model taken into consideration is based on the model specified in [
This second model has been selected and further adapted, due to its convenience for integrating the physical and biophysical variables considered in this study,
For this purpose, it is suggested to take as soil contribution a soil raster layer derived by means of the inverted model represented by
In this section, the results achieved for soil physical and vegetation biophysical values are first presented. Then, the simulation of the backscattering coefficient by means of soil and vegetation microwave scattering models is also described and analyzed. Finally, LAI values are inferred from an inversion procedure and compared to field reference values.
According to the procedure specified in Section 2.3.1, soil surface roughness has been assessed by means of a 5 × 5 cm gridded one meter length plate inserted into the soil surface and recorded on digital pictures. For this purpose, 37 pictures have been registered at singular spatial positions distributed regularly throughout the sample units. Furthermore, these pictures have been processed digitally using a semiautomatic methodology, which enables the acquisition of highly dense profiles describing very precisely the soil surface (green lines in
Each one meter length profile is characterized by at least 400 regularly distributed vertices, whose spatial coordinates allow deriving the vertical displacements of the surface. In turn, these values have been used to compute the standard deviation of these displacements (σ) and the correlation length (
According to the Rayleigh criterion, for the radar frequency used in this study, surface standard deviations lower than 0.7 cm will be designated as smooth, and conversely rough, for σ higher than this quantity. An example of the surfaces observed in some sampling units is given in
As a consequence of these results, the gathered soil surface values for the sampling units appear to lie within the validity range of Oh and Dubois models, so they may be integrated into these soil surface scattering simulation models.
For soil moisture determination, the results reveal extremely low soil moisture content values. A mean volumetric moisture value of m_{v} = 10% has been reached, with a minimum value of m_{v} = 4.5% and a maximum of m_{v} = 14%. Although, these differences might appear very high, this range of values exhibits unfortunately very low moisture content.
These variations may be due to errors introduced by the measurement device and method, as well as the spatial distribution of the soil bulk density in the sampling units, which in turn could be produced by the compaction effects resulting from the tillage techniques, as it is stated in [
Soil texture knowledge is also important since it determines the properties of water retention and transmission of the soil. Soil textural classes are based on the proportions of sand, silt and clay expressed in percentages. In this work, 14 soil samples have been retained for deriving representative soil texture classes. For those not included in this analysis, their class has been assigned based on the nearest determined texture. For this purpose, a conventional methodology based on Stokes Law has been applied. According to the USDA soil texture classification scheme, almost all analyzed samples belong to class clay, although it has been found than in some sampling units the textural class ‘silty clay’ is also present. Volumetric soil moisture as well as soil texture are needed as input variables for dielectric models. The DobsonPeplinsky model has been selected and applied in order to compute the complex dielectric constant required for the soil surface models introduced in Section 2.3.1.
Biophysical values of the vegetation canopy have been retrieved using the methodology depicted in Section 2.3.2. First, in order to minimize optic distortions produced on canopy elements due to far extreme observation angles, it is necessary to eliminate these effects from photographs, which is done by reducing the field of view to from [0°–90°] to [0°–60°]. Moreover, due to illumination conditions, it is not possible to distinguish properly between green and non green elements such as branches, twigs,
Then a segmentation process must be carried out. This operation is considered to be as the most critical in the extraction chain, since the final result will depend on the approach taken to discriminate between the different classes contained in these pictures [
The segmentation operation is carried out by means of a supervised training, where image values are assigned to their corresponding class applying the
These average P_{0} representations depict the geometry/architecture of each observed canopy and reveal the grouping (clumping) effect of the vegetation constituents (mainly leaves in this case). Thus, each vegetation cover is supposed to have a specific behavior to the electromagnetic waves. In this sense, lighter grey scales values indicate a greater transmissivity of the vegetation cover, while the darker indicate a higher opacity. This is the rationale for assessing to which extent the grouping of vegetation components, represented by effective LAI, has an influence on the characterization of the backscattering coefficient, in comparison to the regular distribution of leaves described by true LAI.
As a result of these gap fraction representations, the following parameters are derived: Clumping Index (
An additional parameter of interest that must be also mentioned is the canopy or plant cover fraction, which is defined as the fraction of ground covered by vegetation [
Once the physical values of the soil surface are available, the next phase consists in simulating the soil surface backscattering coefficient. Due to the volumetric soil moisture results derived for the study area only the Dubois model is assessed. First, a sensitivity analysis of the behavior of this model to the range of the sampled moisture and roughness values has been undertaken. For this purpose, a set of plotted curves [
Thus, the outlined model show slight different behaviors depending on the polarization used, which might also produce different results, but in both cases, the Dubois model appears appropriate for the soil simulation, as the existing set of soil physical values fall within the valid limits of this model. Only roughness values approaching
For the horizontal polarization state (
In turn, for the horizontal polarization state (
For this canopy type and observation conditions, this result shows that the simulated soil surface values are in accordance with the image measured backscattering coefficients, as the volumetric scattering contribution of the vegetation layer produces an additive effect on the returned signal and therefore the image measured values must be higher than those returned only by the soil surface. Consequently, this resulting dataset of the soil simulation process is definitely considered as the most suitable for characterizing the vegetation canopy, and will be then used for assessing the vegetation microwave scattering models selected for this study (
Regarding the satisfactory results reached by this simulated vertical polarization state, they appear to be in agreement with the reported results given in [
Once the simulation of the soil surface has been performed, the considered type of vegetation canopy is evaluated by means of the microwave scattering models introduced in Section 2.3.1, which are assessed in the following sections using a non linear regression based on the ‘LevenbergMarquadt’ algorithm, where the convergence criterion for the sum of squares has been always set to 10^{−8}. All statistical analyses have been carried out at a 5% significance level.
First, a Rayleigh Model (
Thus, the values obtained for these two variables are Γ = 0.91 and ω = 0.35, which indicate a very high transmissivity and a mediumlow albedo or reflectivity.
The results attained for these two variables may be in agreement with the considered canopy type, since olive groves are characterized by large gaps between rows of trees and by a low density of vegetation constituents, thus leading to this high transmissivity of the canopy. In turn, for reflectivity, which might be produced by volumetric scattering, its value is not as significant as it might occur for dense vegetation canopies. In this framework, it is not been possible to assess these results by means of direct methods, which in addition are spatially and temporally specific, and therefore they are only meaningful under these particular conditions. However, the knowledge of Γ and ω, and the corresponding system incidence angles, allow applying
Usually, WCM models, using Leaf Area Index as a descriptor of the vegetation canopy, do not specify which of the two variables addressed in this paper,
While there are some differences, these two reported cases are verified to converge properly. For LAI_{true} a coefficient of determination r^{2} = 0.83 (pvalue = 1.56 × 10^{−8}), has been reached after 14 iterations, whereas an r^{2} = 0.89 (pvalue = 3.24 × 10^{−10}) has been achieved for LAI_{eff} after seven iterations. The adjusted model parameters are A = 0.276 and B = 0.071 when true LAI is taken into account, while A = 0.405 and B = 0.115 when effective LAI is considered. These results indicate that in the first case (LAI_{true}) the olive grove canopy is very thin with a very low attenuation, while in the second case (LAI_{eff}) these values are higher, which appear to be more realistic and according to this vegetation cover. Moreover, as the coefficient of agreement is better for effective LAI, the characterization of this type of canopy by means of this biophysical variable is proven to be more acceptable, and might also imply that the grouping effect of the vegetation constituents (or their geometrical properties), is better taken into account by this type of microwave scattering models, as it has been suggested in [
Once parameters A and B are then derived,
In a temporal framework, these documents might be systematically generated and used for environmental or agronomic monitoring, such as analysis and management of radiation and energy exchanges, as well as for subsequent measurement of the canopy photosynthesis [
Finally, the results of this inversion process have been analyzed by comparing inverted values against field measured values. For this purpose, eleven LAI_{eff} and LAI_{true} field observations not included in the regression analysis have been used as reference values. These reference values are also distributed regularly throughout the 8 sampling units.
In
However, for effective LAI values, generally, the trend between measured and inverted values is verified to be quite satisfactory. However, some discrepancies may still be appreciated, as it might be the case for observation P7 (
Therefore, this achieved result confirms again the good performance of effective LAI, which reveals to be a more pertinent variable compared to true LAI for characterizing the vegetation canopy by means of microwave scattering models. Furthermore, transmissivity and albedo values obtained by the Rayleigh model for these particular sites and at this specific time are proven to be also acceptable. Nevertheless, the statements and results issued from this work must still be verified using a higher number of observations, both for assessing the vegetation microwave scattering models and for analyzing the achieved results. As a final remark, in the case of effective LAI, using the full dataset of available observations for the adjustment of
In this paper, a methodology for characterizing vegetation canopies by means of microwave scattering models and optical means has been assessed. For this purpose, soil surface information has been acquired by classical methods, while the biophysical values of the vegetation layer have been derived by the hemispherical photography technique, which has not been extensively used for providing ancillary data in the field of radar systems applications.
For the soil dielectric surface simulation, the Dubois Model achieves the best results (r^{2} = 0.88). Given the soil surface conditions exhibited by the sampling units of this study and the properties of the radar system, the vertical polarization state at this particular band (
Regarding the canopy biophysical variables, this study has distinguished between effective and true LAI’s. Accordingly, a modified version of the Xu and Steven Model has been applied and analyzed, which shows a good capability for assimilating true and effective LAI’s derived from the suggested technique. For this model, effective LAI is proven to return a better coefficient of determination (r^{2} = 0.89). Therefore, the knowledge of the variables related to canopy architecture improve the application of these type scattering models, and the sensor used for acquiring this information (hemispherical photography) provides suitable values for the involved biophysical variables. Furthermore, the convenience of an inversion procedure for generating thematic documents that represent the spatial distribution of LAI values is analyzed. For the observed canopies, under homogeneity conditions, the discrepancies between inverted and measured values are minimal, which confirms that the behavior of this type of model is appropriate under certain conditions of homogeneity.
As a concluding remark, this work proves that the suggested approach has the potential of retrieving biophysical parameters from SAR data. Nevertheless, additional efforts must be still done in order to extend this methodology to wider areas, other vegetation canopies typologies and biophysical variables. Furthermore, these new studies must be performed in a multitemporal framework, with a radar multifrequency dataset, as well as with other measuring sensors for deriving common sets of values for physical and biophysical variables, so that the results can in turn be compared and assessed.
The authors wish to thank C. Gonzalez of the Department of Photogrammetry of the ETSIA (Engineering School of Agronomy, Technical University of Madrid—UPM) for his interest in this study and support, and to R. Garcia for proving the tablet for soil roughness determination and his very valuable suggestions on this topic. The authors would to thank S. Ormeño and R. Espejo for their valuable help and positive comments on soil characterization methods. We are very grateful to the Soil Department of the ETSIA (UPM) for providing its laboratory facilities for soil moisture and texture analysis. Thanks are due to the anonymous referees for their very valuable comments and suggestions.
Geographical situation of the study area.
Location of olive grove study areas and corresponding sampling units (SU).
RADARSAT 2 image, PAULI combination (B 
Example of a set of transects describing a sampling unit/vegetation canopy.
(
Example of soil roughness profiles (in green) and associated parameters (‘
Soil volumetric moisture values for each sampled location.
Edition of photographs, (
Hemispherical view of a single tree, (
(
(
(
(
Agreement between measured and simulated values using Rayleigh Model for the vertical polarization state
Regression analysis results between measured and simulated
Effective LAI map derived by means of the proposed inversion procedure.
True LAI, measured
Effective LAI, measured
Summary of Gap Fraction and LAI values for the Sampling Units (SU).
1  SU1/VC1  0.41  0.22  1.5  0.05  0.08  0.10  3°28′55″W  40°13′47″N 
SU2/VC2  0.47  0.34  1  0.02  0.12  0.12  3°28′56″W  40°13′44″N  
SU3/VC3  0.40  0.19  1  0.03  0.07  0.12  3°28′58″W  40°13′45″N  
SU4/VC4  0.40  0.26  0.94  0.03  0.10  0.12  3°28′58″W  40°13′47″N  


2  SU5/VC5  0.32  0.27  0.71  0.02  0.08  0.10  3°28′51″W  40°13′45″N 


3  SU6/VC6  0.26  0.15  0.47  0.06  0.07  0.10  3°29′34″W  40°13′44″N 
SU7/VC7  0.28  0.22  0.54  0.03  0.10  0.10  3°29′31″W  40°13′39″N  


4  SU8/VC8  0.41  0.22  1.2  0.02  0.08  0.10  3°29′13″W  40°13′49″N 