These four volumes present innovative thematic applications implemented using the open source software QGIS. These are applications that use remote sensing over continental surfaces. The volumes detail applications of remote sensing over continental surfaces, with a first one discussing applications for agriculture. A second one presents applications for forest, a third presents applications for the continental hydrology, and finally the last volume details applications for environment and risk issues.
Coupling Radar and Optical Data for Soil Moisture Retrieval over Agricultural Areas
The spatio-temporal monitoring of soil moisture in agricultural areas is of great importance for numerous applications, particularly those related to the continental water cycle. The use of in situ sensors ensures this monitoring but the technique is very costly and can only be carried out on a very small agricultural area, hence the importance of spatial remote sensing that now enables large-scale operational mapping of soil moisture with high spatio-temporal resolution.
Radar data have long been used to estimate and map the surface soil moisture of bare soils [BAG 16b]. In fact, physical, empirical and semi-empirical models were developed to invert the radar signal to monitor the soil moisture at different spatial scales (intra-plot scale, plot scale, on grids of a few hundred m2 to a few km2). Over vegetated cover surfaces, the coupling of radar and optical data is often necessary to estimate the surface soil moisture. Optical data are complementary to radar data, and their interest lies in their potential to estimate the physical parameters of vegetation, for example Leaf Area Index (LAI) from satellite indices such as the Normalized Difference Vegetation Index (NDVI). These parameters make it possible to evaluate the contribution of the vegetation in the backscattered radar signal, to extract the soil contribution and to finally invert it in order to estimate the surface soil moisture.
To map the soil moisture in the case of vegetation cover, most studies use the semi-empirical Water Cloud Model (WCM) developed by Attema and Ulaby [ATT 78]. Generally, in this model, the total backscattered radar signal is modeled as the sum of (1) the backscattered signal from the soil multiplied by the two-way attenuation and (2) the direct reflected signal from the vegetation. In most studies, the contribution of vegetation has been expressed in terms of one physical parameter of vegetation (biomass, LAI, water content, vegetation height). The soil contribution is generally modeled as a function of soil moisture and surface roughness (for given instrumental parameters: incidence angle, wavelength and polarization). It can be simulated using a physical radar backscattering model (in particular the Integral Equation Model (IEM) [FUN 94]), or a semi-empirical backscattering model (e.g. Dubois [DUB 95] or Baghdadi [BAG 16a] models).
The objective of this chapter is to show how to map the surface soil moisture over agricultural plots (summer and winter crops) and grasslands using the free and open-source software QGIS (Quantum Geographic Information System), by coupling radar (Synthetic Aperture Radar (SAR)) and optical images acquired at high spatial resolution (~10 m × 10 m).
1.2. Study site and satellite data
The study site located near Montpellier in the South of France (Figure 1.1) is an agricultural area (15 km × 15 km). Figure 1.1 shows the layout, made using QGIS, of a satellite image acquired over the study site by Sentinel-2A (S2A).
QGIS functionality for layout:
- Project > New Print Composer >.
1.2.1. Radar images
Two Sentinel-1A (S1A) radar images in C-band (radar wavelength ~5.6 cm) acquired on January 19, 2017, and January 26, 2017, have been used. On January 19, 2017, the soils in the study site were dry (no precipitation for 19 days, a soil moisture around 11 vol.% is measured on a reference plot), whereas on January 26 the soils were very wet (a soil moisture around 30 vol.% is measured on a reference plot) due to the high rainfall that occurred over 4 days before the radar image acquisition (accumulation of 23 mm). S1A images are freely available from the Copernicus1 and Google Earth Engine2 websites. The Copernicus website offers raw images that require radiometric (passage of digital number into backscattering coefficient) and geometric calibration. The downloaded images from Google Earth Engine are already calibrated and ortho-rectified (WGS84 projection system).
Figure 1.1. Study site located 5 km east of Montpellier. The background of the map is an optical image acquired by the satellite S2A. The geographical coordinates are in UTM (Universal Transverse Mercator), zone 31 N. For a color version of the figure, see www.iste.co.uk/baghdadi/QGIS2.zip
Radar images used in this chapter have been downloaded from the Google Earth Engine website. Each image is a stack of three bands: band 1 corresponds to the backscattering coefficient in VV polarization (in decibel (dB) scale), band 2 is the backscattering coefficient in VH polarization (in dB) and band 3 contains the local incidence angle relative to the ellipsoid (in degrees). The two bands corresponding to backscattering coefficients in VV and VH polarizations have been transformed into linear scale. Part 1 of the flowchart (Figure 1.2) shows the processing performed on radar images.
QGIS functionality to transform the first two bands of radar images in linear scale:
- Raster> Raster Calculator >.
1.2.2. Optical image
One optical image acquired by the satellite S2A on October 15, 2016, is used. Ideally, it is preferable to use an optical image at an acquisition date near to that of each radar image. This optical image, freely accessible via the website of the land data center Theia3, covers an area of 110 km × 110 km. The Theia website provides S2A data corrected from atmospheric and slope effects (processing level 2A). S2A images are downloadable from the Theia website in the form of 13 separate spectral bands. The projection system associated with the S2A images downloaded via the Theia website is also the UTM.
To facilitate the use of an optical image, three spectral bands in the visible (bands 2, 3 and 4) and one in the infrared domain (band 8) are first stacked. The optical image is then clipped to adjust the spatial extent of the optical image to the surface of the study site (15 km × 15 km). Next, the clipped image is reprojected into a WGS84 geodetic system to be in the same projection system as the radar images. Finally, an NDVI image is calculated from the reprojected optical image using the spectral bands corresponding to the red and infrared (respectively, band 4 and band 8). The second part of the flowchart (Figure 1.2) shows the processing performed on optical image.
QGIS functionality for stacking the four spectral bands:
- Raster > Miscellaneous > Build Virtual Raster >.
QGIS functionality for clip stacked bands:
- Raster > Extraction > Clipper >.
QGIS functionality to reproject the image:
- Raster > Projection > Warp (Reproject) >.
QGIS functionality to calculate the NDVI image:
- Raster > Raster Calculator >.
1.2.3. Land cover map
A land cover map4 produced by the scientific expertise center of Theia is used to extract the crop plots and grasslands. This map is a thematic raster file with values between 11 and 222, where each value corresponds to a type of land cover5. The projection system associated to the land cover map of Theia is Lambert-93. The land cover map is first clipped to adjust the spatial extent of the study site. Next, the clipped map is reprojected in the WGS84 geodesic system to have the same projection system as that of radar and optical images. Part 3 of the flowchart (Figure 1.2) shows the processing done with the land cover map.
In this section, the steps that lead to the production of soil moisture maps on crop areas and grasslands are described. First, an inversion approach using neural networks is developed. The networks are trained using a simulated dataset of radar backscattering coefficients obtained from the WCM. In WCM, the IEM calibrated by Baghdadi et al. [BAG 06] is used to simulate the soil contribution. The application of neural networks on real satellite data requires the identification of crop and grassland zones. These zones have been extracted from the land cover map available on the study site. Next, an NDVI image calculated from the optical image is used to partition these zones into homogeneous segments (intra-plot scale). Finally, the soil moisture maps are produced by applying the developed inversion approach on each homogeneous segment.
1.3.1. Inversion approach of radar signal for estimating soil moisture
The soil of agricultural areas is covered for a long period of the year by vegetation. An approach that considers the effects of vegetation on the backscattered radar signal for estimating the soil moisture is therefore indispensable for accurate estimation of the soil moisture.
The WCM defines the backscattered radar signal in linear scale (s0tot) as the sum of the contribution from the vegetation (s0veg), the contribution of soil (s0soil) attenuated by the vegetation (T2 s0soil) and multiple soil-vegetation scatterings (often neglected): [1.1] [1.2] [1.3]
- - V1 and V2 are the vegetation descriptors: biomass, vegetation water content, vegetation height, LAI, NDVI (in this chapter, V1 = V2 =...