Interactive comment on “ Multi-scale estimation of surface moisture in a semi-arid region using ENVISAT ASAR radar data ” by M . Zribi

First, we want to thank the reviewers and editor for their very constructive remarks. We thought they highly help us to seriously improve the paper. We have followed their suggestions and take into account most of their comments (answers to reviewers follow). This leads to an almost complete revision of the paper. We send injoined a file with answer to reviewers and a revised version of our paper. We add also two figures that we add to the first version. Kind regards Mehrez Zribi


Introduction
Soil moisture is a key parameter, influencing the manner in which rainwater is shared between the phenomena of evapotranspiration, infiltration and runoff (Engman, 1991;Beven et al., 1996;Koster et al., 2004).In the case of semi-arid and arid regions, this parameter is particularly important for irrigation management (Bastiaassen et al., 2000).In order to optimise and protect water resources, which are often very limited, an accurate estimation of the soil's water content is needed, in order to determine the expected evapotranspiration flux.Considerable efforts are thus devoted to improving Introduction

Conclusions References
Tables Figures

Back Close
Full the evaluation of evapotranspiration, and to understanding its relationship with the vegetation cover and the soil's water content (Simonneaux et al., 2007).Over the last twenty years, radar remote sensing has demonstrated its strong potential as a tool for soil moisture estimation (Ulaby et al., 1996;Le H égarat-Mascle et al., 2002;Moran et al., 2004;Wagner et al., 2007).In the particular case of SAR observations, soil moisture can be estimated with a high spatial resolution, which is not the case with other types of remote sensing measurement (Jackson et al., 1996;Baup et al., 2007;Rahman et al., 2008).
In the case of bare soil, the backscattered radar signal strongly depends on soil moisture and surface roughness (Zribi et al., 2006;Baghdadi et al., 2007).In the case of sparse vegetation, the return signal depends both on the vegetation's backscattering characteristics, and on the attenuation it introduces to backscattering from the soil (Binslish et al., 2001;Le H égarat-Mascle et al., 2002).In the case of dense vegetation, such as that of forests, the soil contribution is generally very weak in the C band, particularly at high incidence angles (Ulaby et al., 1986).Many models have been developed to describe the physics of interactions between the radar signal and the surface or vegetation parameters.For bare soils, various theoretical and empirical approaches have been developed (Fung et al., 1992;Oh et al., 1992;Dubois et al., 1995;Zribi and Dechambre, 2003;Baghdadi et al., 2006;Thoma et al., 2008).Among these, the "linear approach" linking surface soil moisture to calibrated and validated SAR (Synthetic Aperture Radar) measurements (SIRC, ERS, RADARSAT, ASAR, TerraSAR-X . . . ) is widely used (Quesney et al., 2000;Zribi et al., 2006;Paris et al., 2010).In some studies, e.g. that of Woodhouse and Hoekman (2000), the estimation of soil moisture is based on a simple model using an incoherent combination of vegetation and bare soil contributions, weighted according to their respective relative surface areas (as a percentage of the total surface area).The backscattered contribution from the vegetation is determined using physical or empirical models (Wigneron et al., 1999).
Because of the high spatial variability of soil moisture in the studied region, resulting from variable convective phenomena causing the rainfall to be strongly localized in Introduction

Conclusions References
Tables Figures

Back Close
Full small areas, and as a consequence of the presence of a large fraction of irrigated areas, we propose a methodology in which soil moisture is estimated from high-resolution SAR radar data.Our approach in this paper is based on ENVISAT ASAR radar data, acquired simultaneously with in situ measurements of surface parameters (moisture, roughness and vegetation).The developed methodology proposes the use of high spatial resolution mapping, firstly over non-irrigated olive tree fields, followed by mapping over wheat fields.
The present paper is organised as follows: Sect.2 presents the data collected from the Kairouan plain region (Tunisia) under study: the database including satellite and ground-truth measurements is discussed.In Sect.3, the proposed methodology for soil moisture retrieval is described.The derived results, including the validation of soil moisture estimations and mapping, are then presented.Finally, our conclusions are gathered in Sect. 4.

Description
In March 2002, the European Space Agency launched the ENVISAT platform, carrying ASAR in its suite of instruments.Compared with ERS/SAR, this instrument has an extended measurement capacity, due to its multiple operating modes (Rosich, 2002).
In particular, it has a greatly improved measurement repetition rate, with less than three days between two successive images taken at two different incidence angles, as opposed to a 35-day repeat cycle for ERS/SAR.In the present study, we chose to use the narrow observation mode, which generates high-resolution data (12.5 m × 12.5 m pixel spacing).Acquisitions were made between 2008 and 2010, at three different incidence angles ( 18• "IS1", 23 • "IS2" and 27 • "IS3") in co-polarized, alternating HH and VV polarization mode.Details of the SAR image characteristics are provided in Table 1.A large number of SPOT/HRV images was acquired simultaneously with the radar soundings.SPOT/HRV is a multi-spectral optical sensor, with two bands in the visible domain, one in the near infrared, and one in the medium infrared.These proved particularly useful for the mapping of land use and vegetation dynamics.

Radar data
Absolute calibration of the ASAR images was carried out, to transform the radar signals (digitized values) into a backscattering coefficient (σ

SPOT data
The SPOT/HRV images were firstly geo-referenced.Radiometric and atmospheric corrections were then applied in order to estimate the reflectance of the vegetation canopy.
Finally, for each image, the Normalized Difference Vegetation Index (NDVI) was estimated.This index, given by the ratio between the difference between the visible and near-infrared channels, and the sum of these two channels, is related to the green vegetation photosynthetic activity (Rouse et al., 1973).

Surface moisture
Moisture measurements were taken simultaneously with the satellite acquisitions.The in situ collection of soil was extremely important in this experiment, as it was needed to validate the soil moisture retrieval algorithm.For each field, we made approximately twenty measurements at the time of each satellite acquisition.These were made using a handheld Thetaprobe, and by means of gravimetric measurements at depths between 0 and 5 cm.Thetaprobe measurements are calibrated with gravimetric Introduction

Conclusions References
Tables Figures

Back Close
Full measurements.Table 2 illustrates moisture values over field tests during different ground campaigns.

Soil roughness
Roughness measurements were made using a pin profiler (total length of 1 m, and resolution of 2 cm).In order to guarantee suitable precision in the roughness computations, approximately 10 profiles were recorded for each field.As the surface height profile is considered to be ergodic and stationary, we can compute the correlation function for each profile (Zribi et al., 1997), and derive two statistical parameters: the rms height (vertical scale of roughness), and the correlation length (l ) which represents the horizontal scale over which similar roughness conditions are detected.The rms height values are approximately equal to 0.7 cm for wheat fields, and are generally greater than 1.5 cm for olive tree fields, as illustrated in Table 3.

Vegetation covers
In order to characterise the vegetation covers, we considered three types of measurement.For the non-irrigated olive fields, we measured the distances between trees and the size of the trees in a large number of test fields.Distance between olive trees is of approximately 20 m, and the mean projected surface area of an adult olive tree, is approximately 16 m 2 (Fig. 2).
In the case of wheat fields, we implemented two types of measurement:

Leaf area index data
The Leaf Area Index (LAI) is defined as the total one-sided area of leaf tissue per unit ground surface area.According to this definition, the LAI is a dimensionless quantity characterizing the canopy of an ecosystem.During the 2008/2009 agricultural season, the LAI was derived from hemispherical digital photography based on analysis of the canopy gap fraction (Duchemin et al., 2008).These measurements were applied to Introduction

Conclusions References
Tables Figures

Back Close
Full each wheat field, on different days during the vegetation season.Irrigated wheat fields are generally characterised by a higher LAI than non-irrigated wheat fields.Before the end of March, the highest observed LAI was approximately 2. At the end of April we observed the highest vegetation density, with its maximum generally lying in the interval (4-6).Table 4 illustrates measurement values over wheat test fields.

Vegetation water content (VWC) data
The VWC was measured several times in five fields during the 2009 vegetation cycle (Table 4).For each field, measurements were made at three locations, each having a 1 m 2 surface area.The above ground biomass was removed, and wet and dry weights were used to compute the VWC.A mean value was computed from the three measurements.

Land use
Land use validation was carried out in March 2009, with different fields being selected from the studied region (more than 150 fields) with two parts, a first one for the identification of empirical NDVI limits between different types of vegetation classes, and a second one for the validation of our approach to land use classification.
Land use mapping is based on a decision tree, using three types of satellite data: four SPOT images, SRTM data and finally two radar images.We established eight classes of land use: non-irrigated olive trees, irrigated olive trees, irrigated winter vegetables, irrigated summer vegetables, bare soils, urban areas, mountainous areas, water cover and areas of coastal salt flats "sebkhas".In the case of vegetables, as previously mentioned, we considered two classes, one for winter and the other for summer.We used empirical NDVI thresholds with the images acquired at the end of December 2008 (NDVI > 0.4) and during July 2009 (NDVI > 0.3).In fact, during these two periods, only irrigated vegetables presented a high NDVI.For the wheat classes (irrigated or nonirrigated), we made our analysis on two different dates, the first at the beginning of the Introduction

Conclusions References
Tables Figures

Back Close
Full cycle (in December 2008), and the second at the end of the vegetation development period (April 2009).The distinction between irrigated and non-irrigated wheat is based on a NDVI threshold equal to 0.5, since the irrigated class has a higher NDVI.Irrigated and non-irrigated olives are separated using a K-mean approach, based on a single optical SPOT image.The DTM provided by the Shuttle Radar Topography Mission (SRTM, http://srtm.usgs.gov/)allowed certain zones to be eliminated from our land use analysis.We excluded mountainous areas with an altitude greater than 300 m.We also identified water cover and urban classes.Validation of these remotely sensed classifications, based on ground verification over more than 100 fields with different types of land uses, reveals an accuracy of around 94%. Figure 3 illustrates the results of our land use mapping for the 2008-2009 season.The non-irrigated olive tree class covers 43% of the studied site, and the wheat class corresponds to 12% of the surface area of the studied site.

Methodology and results
Our approach to soil moisture estimation and mapping is carried out on two types of land use: non-irrigated olive areas and wheat fields, which represent the two most important land use classes.

Introduction
For the purposes of surface soil moisture estimation, we used the IS1, IS2, IS3 configurations, corresponding to low incidence angles of less than 30 • .The aim of this approach was to limit the influence of vegetation and soil roughness, thereby increasing the accuracy of the moisture estimations.
As shown in the following expression, the signal received from the non-irrigated olive tree areas can be written as the incoherent sum of two contributions, weighted by Introduction

Conclusions References
Tables Figures

Back Close
Full their respective percentages of terrain coverage.The first of these corresponds to the bare soil contribution, whereas the second corresponds to that from the olive trees, which composed on three terms: the vegetation contribution, the bare soil contribution including the influence of olive vegetation attenuation, and the contribution resulting from interactions between the soil and the vegetation.
where C is the fraction of the non-irrigated olive field surface covered by olive trees.
Using the estimated distance between olive trees of approximately 20 m, and the mean projected surface area of an adult olive tree, i.e. approximately 16 m 2 , we derive for different incidence angles lower than 30 • a value between 4% and 10% for C fraction.We make therefore the assumption that the influence of the vegetation on the radar signal is negligible.Based on different developments over disperse vegetation cover (Wagner et al., 1999;Zribi et al., 2006), the radar signal could be modelled with a linear relationship between radar signal and moisture, as: Where α is related to the parameter C and to the attenuation due to the olive trees characteristics.
g is a function of soil roughness and vegetation cover effects on radar signal.Mv is volumetric soil moisture.The inversion process is based on four successive steps: 1. Normalisation of all radar images to the same incidence angle (20 • ). 4. Application of the inversion approach to the studied site.Moisture estimations are computed within a 100 × 100 pixel grid positioned over each olive tree pixel, in order to avoid local heterogeneity effects.

Normalisation of the radar signals to an incidence angle of 20
• .
Normalisation of the ASAR data is based on the interpretation of radar signal data, for different incidence angles, recorded over large olive tree AOIs.These areas are selected to be in the olive tree class, and only those radar images recorded on very dry dates are considered, in order to eliminate noise contributed by soil moisture effects.As shown in Fig. 4, we observe a decrease in radar signal with increasing incidence angle.
The angular dependence of backscattering coefficient is modelled with a mathematical function (Baghdadi et al., 2001), resulting in a high correlation coefficient (respectively R 2 equal to 0.56 and 0.6 for HH and VV polarisations): Using this fit, the data was normalised to a θ ref equal to 20 • incidence angle using the following expression: We retrieve b respectively equal to 5.5 and 6.3 for HH and VV polarisation.

Roughness and vegetation effect reduction
In order to limit roughness and vegetation effects, we computed the difference between each raw data image and a reference image taken under dry conditions at the beginning of the vegetation season (21 December 2008), with a moisture content of approximately 5% over the studied site.
In the case of the olive tree fields, we observed very small variations during the vegetation cycle, due in particular to the olive trees being evergreen.We thus consider, Introduction

Conclusions References
Tables Figures

Back Close
Full as an initial hypothesis, that the vegetation has an approximately constant effect on the radar signal.
If we now consider a reference image, with a roughness R 1 and moisture content Mv 1 and a data image with a roughness R and moisture content Mv, As for surface roughness, the olive fields generally have a tillage corresponding to ploughed soil with an rms height of around 1.5-3 cm, as shown in ground measurements.Only small variations could be observed after rainfall events.However, the soil is ploughed at different times during the year, which induces low variations on rms heights.For such roughness levels the backscattered radar signals are nearly saturated, as illustrated in IEM simulations of backscattering coefficients realized in its validity domain with rms height lower than 1.1 cm (Fig. 5).The subtraction of a reference image is therefore sufficient to considerably reduce the influence of roughness in the observed pixels, even for cases where there are small differences in roughness between the two images.We can thus simplify the above expression to:

Validation of the proposed algorithm
Validation of the proposed algorithm is based on a comparison between ground-truth (gravimetric, and handheld Thetaprobe) measurements and estimations derived from ENVISAT ASAR data, for data acquired in 2010 and moisture conditions ranging from dry to wet, over olive test fields (P4bis, P10, P12).The resulting RMSE is equal to 3.8% for the HH and 4% for the VV polarisations, as illustrated in Fig. 7. Figure 8 illustrates a good coherence between soil moisture estimations with HH and VV radar signals, with an RMSE equal to 2% and bias equal to 1.6% over tested fields.The accuracy of this outcome demonstrates the robustness of the proposed algorithm, in spite of its simplicity.Our decision to develop an inversion algorithm, for olive trees only, considerably reduces the influence of roughness and vegetation on the soil moisture estimations.It is thus possible to apply this validated model to each ENVISAT ASAR image, to produce soil moisture maps over fields in the non-irrigated olive tree class.

Mapping of soil moisture
In order to eliminate the effects of local terrain heterogeneities (due to soil texture, vegetation dispersion heterogeneity, discontinuities between fields, etc.) in the processed radar signal, the soil moisture was estimated over large cells defined by 100 × 100 pixel areas (about 1 km 2 ).For each resulting cell, the soil moisture estimation is applied only if more than 25% of the cell's pixels belong to olive tree fields.The value of the computed moisture can be then considered to be representative of the whole cell.
To validate these estimations, the ground-truth measurements taken within the same cell are averaged.When the inversion is applied to the HH and VV radar signals, we observe similar results for both polarisations.In order to increase the precision of our estimations, we took the mean value of the two polarisations as the final result in the mapping process.
In Fig. 9, soil moisture maps are shown for three different dates.These maps are directly related to the temporal and spatial variability of the precipitation over this region.Introduction

Conclusions References
Tables Figures

Back Close
Full For example, on date 9 December 2009, dry soil is observed over the full studied site, with a low moisture content of around 10%.Indeed, no rainfall was recorded during the 15 days preceding the acquisition of this satellite image.In the case of the image taken on 11 April 2009, strong spatial variability of the surface moisture can be observed.In fact, a rainfall event arriving from the West occurred during the afternoon of 11 April.
In the Eastern part of this image, the soil moisture remained low.The third image in this figure provides the moisture map produced one day later, on 12 April, showing generalised rainfall throughout the studied site, associated with a global increase in soil moisture with a mean value of around 25%.
Our approach allows the moisture to be estimated over approximately 50% of the studied site.It is presented particularly in the South East, where irrigated agriculture is absent.The interest of the choice of this class of land use is evident, since the computed moisture has only a small sensitivity to roughness and vegetation, both of which are affected by very limited changes from one year to another.This type of algorithm can thus be applied each year, with no need for it to be adapted to variations in local conditions.

Introduction
Following an estimation of soil moisture related to precipitation effects, carried out over non-irrigated olive tree fields, we propose a methodology with high spatial resolution analysis over wheat fields.In fact, because of limited fields scale (generally lower than 2 ha), and high spatial variability of moisture between irrigated and non-irrigated wheat fields, we need to realize moisture estimation in higher spatial resolution.In this case, the inversion algorithm is based on three steps: -Correction of vegetation effects using a first-order radiative transfer model.Introduction

Conclusions References
Tables Figures

Back Close
Full -Analysis of the relationship between surface moisture and the processed radar signal corrected for vegetation effects.
-Application of an inversion approach over wheat fields.The moisture estimations are computed for each cell within a 5 × 5 grid over each wheat field.

Vegetation correction
In order to estimate the soil moisture over fields covered by vegetation, we first need to eliminate the vegetation's influence on the backscattered radar signal.We propose to use the water-cloud model developed by Attema and Ulaby (1978).For an incidence angle θ, the backscatter coefficient is represented in the water cloud model by the expression: where τ 2 is the two-way vegetation transmissivity.The first term represents scattering due to the vegetation; the second term is linked to multiple scattering effects, and the third term represents the soil scattering attenuated by the vegetation cover.The second term can be neglected in the case of wheat scattering (Ulaby et al., 1986).Expression (Eq.9) can thus be simplified to: and where VWC is the vegetation water content (kg/m 2 ).Introduction

Conclusions References
Tables Figures

Back Close
Full A and B are parameters which depend on the type of canopy.This formulation represents a first-order solution for the radiative transfer equation through a weak medium, where multiple scattering is neglected.The parameters A and B are estimated empirically, using our ground-truth measurements and radar signals measured over wheat test fields (P4, P6, P7, P9 and Pst2) in a part of radar acquisition dates.

Relationship between soil moisture and bare soil radar signals
For bare soil backscattering, we consider a simple relationship between moisture and radar signal.
This expresses a linear relationship between soil moisture and the radar signal in dB.
The slope of this relationship is estimated from measurements acquired just before the vegetation starts to develop.From the end of December until the end of January, the soils are bare, with no vegetation cover on the wheat fields.The constant term β is dependent on the roughness effect which must be precisely estimated to reduce errors in moisture estimation (Alvarez-Mozos et al., 2009;Lievens et al., 2009).After sowing, the farmers do not till the soil again before harvesting.Our roughness ground measurements indicated the presence of smooth soils with an rms height approximately between 0.6 and 0.8 cm.It is reasonable to assume that for some wheat fields roughness could have a small decrease throughout our period of inversion.IEM simulations show approximately a 2 dB decrease of backscattering coefficient, at low incidence angles, for surfaces with a rms height going from 0.8 cm to 0.6 cm (Fig. 5).Our hypothesis of a constant mean β value for all wheat fields during period of inversion could then introduce a supplementary maximum error in volumetric moisture estimation of about 3% due to ± 1 dB error in roughness effect.
In order to retrieve variations of this constant as a function of incidence angle, we determined the variations in radar signal as a function of incidence angle, at different dates under dry conditions, over wheat fields.In fact, the angular variations in Introduction

Conclusions References
Tables Figures

Back Close
Full radar signal are related only to roughness effects, as illustrated in IEM simulations of backscattering coefficients, with different soil roughness and moisture values (Fig. 10).
Figure 11 presents the angular dependence of HH polarisation data, based on radar acquisitions over a large number of wheat fields at different incidence angles.As for olive trees fields, we use the function (Eq. 3) to model radar signal variation function of incidence angle.
For the proportionality between soil moisture and radar signal, we considered a constant value, independent of incidence angle.This was confirmed by Integral Equation Model (IEM) (Fung et al., 1992), simulations showing very little change in radar signal sensitivity to moisture, for the range of incidence angles used in our analysis, i.e. between 15 • and 30 • (Fig. 12).
The backscattering model over bare soil thus becomes:

Validation of moisture estimation
Validation of the proposed algorithm is based on comparisons between ground-truth measurements made in test wheat fields (P4, P6, P7, P9, Pst2) characterised by different soil moistures, ranging between dry and wet conditions, and estimations for the state of vegetation development derived from ENVISAT ASAR radar signal acquisitions, made in 2009 and 2010.The results are illustrated in Fig. 13.We observe more validation points in HH polarisation because of the use of one ASAR image with just this configuration.
The resulting rms error is equal to 5.3% and 6.4%, in the respectively HH and VV polarisations.Although this accuracy could be considered to be adequate, in the case of irrigated fields we often observed a high spatial variation of the soil's moisture content.In addition, our measurements were often carried out within a three hour period before or after the site was overflown by the satellite.Some differences could arise due to a high evaporation rate, and in some cases it is possible that our ground-truth Introduction

Conclusions References
Tables Figures

Back Close
Full measurements were affected by irrigation which commenced during the satellite measurements.Finally, as discussed in the last section, the hypothesis of a constant roughness effect could increase the rms error.

Mapping of soil moisture over wheat fields
For the studied site, application of the inversion algorithm requires some information related to the vegetation's water content.For this reason, we developed an approach based on the interpretation of SPOT satellite optical measurements, linking VWC to LAI and then to NDVI index estimations.Figure 14 illustrates the relationship between measurements of water content and LAI over different test fields.We observe a good correlation between the two variables, with R 2 equal to 0.61.Therefore, knowledge of the LAI values can be used to estimate the vegetation's water content (VWC), using the following equation: For the LAI estimations, we made use of the NDVI vegetation index derived from SPOT images acquired during the full vegetation cycle.We proposed a relationship between NDVI and LAI estimations for wheat, based on a large database of ground and SPOT/HRV satellite measurements.This expression is: with NDVI ∞ = 0.75, NDVI soil = 0.15 and k = −1.24.
For LAI < 2, we observe an increase in the LAI with NDVI indices.For higher values of LAI, the estimation becomes more complex, with saturation of the NDVI values resulting in reduced accuracy for the LAI estimations.In order to make reliable estimations of the vegetation moisture content, allowing accurate vegetation corrections, we ran the inversions only for the period between January and March, for which the LAI were still not high (lower than about 1.5).The expressions for water content estimation Introduction

Conclusions References
Tables Figures

Back Close
Full could then be applied with good accuracy.In the case of dense vegetation cover, it is very difficult to retrieve the soil moisture with sufficient precision.This is also an intrinsic limitation of the use of C-band SAR data, since the radar signal is strongly attenuated by the vegetation.Our process thus involves, firstly NDVI mapping from SPOT satellite images, from which the LAI and then vegetation water content are deduced over wheat fields.Finally, after applying corrections for the influence of vegetation, we derive the soil moisture.
All pixels in the wheat class of land use are considered to be valid candidates for soil moisture estimations.However, a radar signal from a minimum number of neighbouring pixels is required to avoid adding speckle noise to the results.We thus considered 5 × 5 pixel windows (about 0.4 ha) for the computation of effective radar signals in the wheat class, which were then used to estimate the soil moisture.
Figure 15 illustrates the resulting soil moisture maps, computed over wheat fields at different dates.For wet days corresponding to rainfall events, such as that of 16 January 2009, a high soil moisture value can be observed for all wheat fields.For dry dates such as 7 March 2009, we observe different moisture values.Increasing moisture values can be observed over irrigated fields.On 24 December 2008 date, non-irrigated wheat fields are found to have soil moisture of approximately 6%.For irrigated wheat fields, the values are generally higher; even very high moistures (around 40%) can be observed in some cases.The variability of these moisture observations is in complete agreement with the land use classification, distinguished by two classes, i.e. nonirrigated and irrigated wheat.This type of mapping process, if enhanced by means of high temporal monitoring, could become a very useful tool for the regional analysis of irrigation and water consumption, particularly in semi-arid areas with limited water resources.

Final moisture mapping
Figure 16 provides an illustration of our mapping process in a small area of our studied region, in 7 March 2009, in which moisture map computed for non irrigated olive tree Introduction

Conclusions References
Tables Figures

Back Close
Full fields and wheat fields are combined.Differences in moisture level can be observed between the two classes.The mean moisture level in the olive tree fields is approximately equal to 10%, as opposed to 15% for the wheat class fields.This difference is not due to irrigation alone, but also to differences in soil texture.In particular, the percentage of sand in the soil of olive fields is higher than in wheat fields, such that soil moisture of the former decreases more rapidly after rainfall events.

Conclusions
Numerous studies have been published on the topic of soil moisture estimation over bare soil, or over land with one type of vegetation, in one spatial direction.The present study describes an approach for the two-dimensional estimation of soil moisture, a first one only linked to rainfall, using fields in the "non-irrigated olive tree" land use class.
Radar data has been used to clearly identify soil moisture variations resulting from rainfall events.A relationship is established between ground-truth measurements and backscattered radar signals.The proposed inversion approach is based on three main steps: -Normalisation of the ENVISAT ASAR data to one incidence angle.
-Reduction of roughness effects through the subtraction of a reference image corresponding to the 2-D signal obtained on a dry day.
-Implementation of an empirical relationship, enabling the soil moisture to be derived from the processed radar signals.
The validation of this approach has been demonstrated to have good accuracy in terms of moisture estimation.Moisture mapping using this process is shown for several dates, revealing various temporal and spatial variations, linked only to rainfall events.This estimation is proposed at a cell resolution of 100 × 100 pixels.The approach developed for fields in the non-irrigated olive tree class (about 43% of used land) allows nearly Introduction

Conclusions References
Tables Figures

Back Close
Full all areas of the studied region to be covered, from which a quantitative and precise estimation of the spatial variability of soil moisture can de derived.This parameter is essential for rainfed agriculture which represents 90% of agriculture areas in Tunisia.
In fact, understanding surface processes over these areas and vegetation resistance, particularly during frequent drought periods or because of climate change, is essential.
A second type of moisture estimation is proposed over wheat fields.The principal objective of this estimation is to identify a relationship between moisture variability and irrigation in the studied region.The methodology developed for this application is based on two steps: -Correction for vegetation effects using a simple first-order radiative transfer model.This correction is based on the relationships established between vegetation water content and optical satellite measurements (SPOT/HRV data).
-Determination of a linear relationship between ground moisture measurements and processed bare soil radar signals.
Good agreement is found between the inversion results and the ground-truth measurements, with a mean rms error of about 5.8%.Moisture mapping over wheat fields allows those fields that are irrigated, and thus characterised by generally higher moisture values, to be clearly identified, particularly during dry periods.If associated with temporal monitoring at a high repetition rate, this mapping process could make it possible to implement regional control of irrigation politics.In future studies, we plan to use the soil moisture maps described here, together with other satellite data sources, in particular those describing vegetation dynamics, to estimate the 2D distribution of evapotranspiration effects over the studied site.Introduction

Conclusions References
Tables Figures

Back Close
Full  Full  Full  Full  Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | N, 35 • 45 N) (Fig.1).The climate in this region is semi-arid, with an average annual rainfall of approximately 300 mm per year, characterised by a rainy season lasting from October to May, with the two rainiest months being October and March.As is generally the case in semi-arid areas, the rainfall patterns in this area are highly variable in time and space.The mean temperature in Kairouan City is 19.2 • C (minimum of 10.7 • C in January and maximum of 28.6 • C in August).The mean annual potential evapotranspiration (Penman) is close to 1600 mm.The landscape is mainly flat.The vegetation in this area is dominated by agriculture (cereals, olive trees, and market gardens).Crops are various and their rotation is Discussion Paper | Discussion Paper | Discussion Paper | typical of semi-arid regions.The aquifer of the Kairouan plain represents the largest basin in central Tunisia.It is fed by the infiltration of surface waters during floods in the natural regime, or at the time of dam releases since the construction of the Sidi Saad and El Haouareb dams.Surface and groundwater streams are drained into Sebkha Kelbia, a large salt lake.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2. Reduction of roughness and vegetation effects by subtracting the radar data recorded on a dry date, at the beginning of the season, from the data used for the soil moisture estimation.3. Defining a relationship between the ground soil moisture and the processed radar signals from different olive tree test fields.Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Figure 6
Figure6illustrates the linear relationship found between ground surface moisture measurements and radar signals over different test fields.Each point corresponds to a set of two measurements (ground-truth measurement, radar signal) recorded for different test fields.A strong correlation can be seen between the two types of data, for HH and VV polarisations, with a correlation coefficient R 2 equal to 0.67 and 0.53 respectively.The measured moisture contents range between 5% and 22%.
Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Table 1 .
Characteristics of ENVISAT ASAR and SPOT data used in this study.

Table 4 .
Leaf Area Index and vegetation water content measurement.