Research project SR/00/328 (Research action SR)
CONTEXT AND OBJECTIVES
Ultra-wideband ground-penetrating radar (GPR) technology has demonstrated its capability for fieldscale digital soil mapping of key soil properties such as moisture and roughness (Huisman et al. 2003, Lambot et al. 2008a, Minet et al. 2011). Amongst existing GPR systems and processing techniques for soil dielectric permittivity and correlated moisture retrieval, the method introduced by Lambot et al. (Lambot et al. 2004a, Lambot et al. 2006a) appears to be the most accurate and robust for real-time mapping. The method relies on full-wave modelling and inversion of the radar data, accounting for (1) wave propagation in 3D layered media through exact Green’s functions and (2) antenna effects including antenna-medium coupling. The method has been successfully used in a series of hydrogeophysical applications (e.g., Lambot et al. 2004b, Lambot et al. 2004c, Lambot et al. 2008b, Mahmoudzadeh Ardekani and Lambot 2014) and has further been coupled with soil hydrodynamic modelling (e.g., Lambot et al. 2006b, Jadoon et al. 2008, Lambot et al. 2009). The method has been recently generalized to near-field antenna conditions (Lambot and Andre 2014, Anh Phuong et al. 2014), for deeper soil characterization (e.g., root zone), and a lightweight radar system is being setup for close range Remotely Piloted Aircraft Systems (RPAS). Additionally, up-scaling of local time-lapse GPR soil moisture maps can now be widely tested, using freely available Sentinel-1 C-band SAR data sets. In that spaceborne remote sensing framework, site-specific in situ calibrations remain essential for operational retrieval of surface soil moisture. The calibration of backscattered synthetic aperture radar (SAR) data by classical in situ point measurements using soil sampling or invasive sensors remains an expensive and complicated task given the inherent spatiotemporal variability of soil moisture at the field scale. The difference in support scales between remote sensing methods and local sensors reaches several orders of magnitude, therefore making these two methods hardly comparable for properly calibrating SAR products given the inherent variability of soils. RPAS-based GPR provides a new way for bridging the gap between soil sampling and remote sensing and thereby improve the value of remote sensing data products (e.g., see SENSAR project, Belspo).
The ability of imaging spectroscopy to cover large surfaces in a single campaign and to study the spatial distribution of soil properties with a high spatial resolution represents a major opportunity for improving the monitoring of soils and the spatial predictions of physico-chemical soil properties. Several studies have demonstrated the potential of remote sensing technologies for determining soil organic carbon, soil salinity and clay content (Ben-Dor et al., 2002; Stevens et al., 2008). However, there are still major constraints to the widespread use of imaging spectroscopy for soil applications due to the inherent complexity of natural surfaces (Ben-Dor et al., 2009). There is ample evidence that the conditions of the soil surface significantly affect the spectral signal. In particular, soil moisture and roughness are subject to variations both in time and in space and induce changes in soil reflectance that approach or exceed the spectral response of intrinsic soil properties such as organic matter content, texture or CEC (Barnes et al., 2003). Soil moisture and roughness are known to induce a significant anisotropy on the directional distribution of the solar radiation scattered from bare soils (Croft et al., 2014, Chappel et al., 2006). The recent and rapid advances in field, RPAS-based and airborne spectroscopy give a window of opportunity to take up this challenge (Nocita et al., 2015). In particular, combining close range hyperspectral spectroscopy with GPR and Sentinel products appears as a promising solution given the complementary information provided by the sensors as well as the relevance of the characterization scales. Soil moisture and roughness estimates provided by GPR is expected to provide valuable information to constrain soil spectroscopy models and infer physico-chemical properties of interests such as soil organic content with a higher accuracy.
The general objective of the RAPAS project is twofold: (1) use RPAS-based GPR to calibrate Sentinel-1 SAR products for surface soil moisture and to estimate root-zone soil moisture, and (2), to disentangle soil surface moisture and roughness effects from hyperspectral data using a new close range RPASbased platform. In that respect, the RPAS integrated platform will permit to bridge the scale gap between the laboratory or local scale and airborne and spaceborne remote sensing, in order to improve digital soil mapping of key soil properties such as organic content and clay content in addition to soil moisture. The complementary data provided by lightweight spectrometer and ultra-wideband GPR will be merged in a combined statistical-mechanistic data fusion framework.
The four specific objectives of RAPAS are defined as follows:
The first specific objective is to perform RPAS-based GPR measurements over the RAPAS test sites (including BELAIR in the Hesbania region in central Belgium) to provide high-resolution maps of both surface soil moisture and root-zone soil moisture, plus possibly roughness amplitude. These data will be acquired simultaneously with the spectrometer and SAR acquisitions. Acquisitions will be repeated in time over the same fields for different moisture/roughness conditions. Time-lapse measurements will in particular benefit from the Sentinel-1 short revisit time. During the field campaigns, ground-truth soil samples will be collected for both surface and root-zone volumetric soil moisture. This objective entails the following scientific questions: (1) what are the optimal frequency ranges to consider to provide both root-zone (lower frequency ranges) and surface (higher frequency ranges) soil moisture as well as to provide the most suitable information for unravelling spectrometer data from moisture and roughness effects, (2) which type of antenna(s) to use for optimal signal-to-noise ratio and lightweight (e.g., dipole vs horn or Vivaldi), (3) how to optimally set up the GPR-RPAS system and flight (defining resolution, flying height, etc.), and (4) what is the optimal strategy to calibrate low-frequency radar antennas. GPR acquisitions will also be used for the validation and the improvement of the SAR data processing. The short revisit time and free availability of data of Sentinel-1 will permit to extend the methods introduced in previous SENSAR project (Belspo, Stereo II). Finally, during the field campaigns, ground-truth soil samples (surface and root-zone) will be collected for comparison with the larger scale RPAS-GPR estimates and validation for soil moisture and roughness.
The second specific objective is GPR data processing, namely, inversion of the radar data to provide root-zone and surface soil moisture, and roughness amplitude. First, the global reflection and transmission complex functions of the antenna will be determined, considering both the antenna separated from the RPAS system and the antenna mounted on the RPAS system. The effect of the RPAS system on the antenna characteristic functions will then be determined. Antenna calibration measurements will be performed over a lake representing an infinite surface (a relatively large homogeneous medium is required for the low frequencies, < 100 MHz). Second, a specific inversion strategy will be developed for the particular RPAS-based radar configuration. The objective here is to simultaneously retrieve moisture and roughness, which is theoretically possible given the ultra-wide bandwidth. Indeed, although surface soil roughness does not significantly affect the backscattered radar signal when operating at sufficiently low frequencies, higher frequencies are affected. The time-lapse acquisitions will further contribute to disentangle moisture-roughness effects, as surface roughness is not expected to significantly change over short periods of time. Insights regarding characterization depth will also be provided through numerical experiments and frequency analyses. The final GPR results will be compared to the ground-truths for validation.
The third specific objective is with respect to satellite SAR imagery: to perform a tentative up-scaling of RPAS-based GPR soil moisture information, using GPR measurements as calibrating and validating ground-truths on Sentinel-1 SAR imagery. GPR-derived surface soil dielectric permittivity maps will be used for calibrating the SAR data using state-of-the-art inversion models (Piles et al. 2009, Pierdicca et al. 2010, Ramirez-Beltran et al. 2010, Verhoest et al. 2007). The SAR model will account for roughness, but not for vegetation. The direct use of dielectric permittivity values instead of volumetric/gravimetric water content is expected to remove a large part of the uncertainties usually associated with SAR and GPR data processing, i.e., the uncertainties due to the use of a petrophysical model to relate permittivity to water content. This specific objective situates in the direct continuation of the SENSAR project and involves two major improvements: (1) the access to readily available Sentinel-1 data with short revisit time and (2) the RPAS-GPR that will permit characterization to a scale closer to the spaceborne remote sensing scale compared to the previously used quad-based GPR platform.
The fourth specific objective is related to the quantification of the spectral behaviour of soil properties under real-life conditions where soil moisture and roughness are dynamic in space and time. Time lapse multi- and hyperspectral data will be acquired simultaneously with the SAR acquisitions (see objective 1) at the scale of airborne/satellite products. This specific objective involves the following scientific tasks:
1) To develop and evaluate a RPAS-platform to acquire simultaneously positional, radar and spectral information 2) To investigate the relationship between soil moisture and roughness and soil spectral properties and 3) To demonstrate how this information can be used to improve the quality of soil property estimation from hyperspectral data. Within this project, we will specifically focus on soil organic carbon as a key soil property. During field campaigns, ground-truth soil samples (soil organic carbon and spectra) will be collected and confronted with the larger scale RPAS-derived information for validation purposes.
METHODOLOGY
Study site
The study area is the central part of the loam region of Wallonia, Belgium, and covers an area of about 40 by 5 km. The area is characterized by a uniform soil cover (mostly Luvisols) with very little variation in soil texture (about 10% clay, 75% silt) and organic carbon (about 1 – 1.3 %C). Arable land is the main land use in the area (about 70%) and bare soil surfaces with a broad range of soil roughness conditions (from seed bed to freshly ploughed soils) are present throughout the year. The region is part of the BELAIR ‘Hesbania’ study site and is therefore regularly monitored by an airborne hyperspectral sensor (APEX). We will conduct measurements on several fields along a soil roughness gradient in the study area. These measurements will be ideally taken in the same period as the APEX flights.
WP1: RPAS-GPR measurements
Time-lapse RPAS-GPR field campaigns over the test sites will take place simultaneously with the spectrometer and SAR Sentinel-1 acquisitions. Ground-truth soil samples will be collected as well during one or two campaigns, for validation of the RPAS-based GPR estimates of soil moisture (root-zone down to 0.8-1 m and surface down to 0.05 m). Roughness of the soil surface will also be characterized (method to be determined, including, e.g., photogrammetry or needle/laser profiler). GPR data acquisitions will be performed following the method of Lambot et al. (Lambot et al. 2006a, Lambot et al. 2004a, Lambot and Andre 2014). The GPR system will be set up using a lightweight handheld vector network analyser (VNA) connected to an ultra-wideband transmitting and receiving antenna. The optimal antenna type and operating frequency bandwidth will be determined through numerical experiments and trials. Root-zone information will be provided with frequencies ranging from 20-100 MHz, while surface properties will be estimated using frequencies ranging from 0.8-5.4 GHz, including the C band of Sentinel-1 radar. To cover both ranges, it is likely that two flights with different radar setups will have to be performed for a field campaign (to be investigated). The present weight of the system is about 5.250 kg, including the VNA, antenna, GPS and controlling micro-computer, which is a reasonable weight for flying with an octocopter-type RPAS. This weight is expected to be further decreased to about 3 kg with a new VNA presently under test. The flying altitude will be around 5-6 m, to ensure a sufficient signal-to-noise ratio and high-resolution. The RPAS system is the RPAS-X8 from RC Take Off, including 8 motors of 1050 W distributed over 4 arms. The flying autonomy is expected to be around 15 minutes. Mounting a GPR on a RPAS for digital soil mapping is very innovative and probably a world first considering the advanced processing algorithms that will be used to process the data (see WP3). The characterization scale and resolution of the system is of particular value for bridging the scale gaps between classical soil probing, ground- based geophysical mapping and airborne and spaceborne remote sensing. In that respect, and following the results of previous projects (HYDRASENS and SENSAR, Belspo, Stereo II), such a RPAS-based GPR platform is expected to be the required tool for validating and improving remote sensing data products.
WP2: RPAS-based hyperspectral measurements
The quick changes in soil moisture at the soil surface and strong vertical gradients make it very difficult to collect soil moisture samples at the same time as optical data using traditional methods. Although it is well known that soil moisture has a significant influence on spectral reflectance of soil, it remains challenging to isolate its effect from spectral data. By using an RPAS platform that integrates GPR and spectral measurements, we thus propose an innovative solution for this problem. We will use the same RPAS system as for the GPR measurements. The RPAS is equipped with a professional gymbal that has a ‘target lock‘ function to keep the sensor pointed at a fixed point on the ground or that can be stabilized along three axes. Multispectral images will be acquired at nadir using the Tetracam Mini MCA6 with filters that are suitable for the prediction of soil organic carbon. An Ocean Optics VIS-NIR sensor will be used to collect the soil radiance over the near-infrared-visible range. A cross-calibration between the Tetracam, OO and an ASD Fieldspec spectrometer will be performed under natural illumination conditions. An incident light sensor will be used to estimate soil reflectance values as a fraction of the detected incident radiation. For the multi-spectral images, ground control points will be placed at the surface. The workflow is then based on Grenzdörffer and Niemeyer (2011) and includes triangulation, radiometric and vignetting reduction and ortorectification. Since existing UAV-altimeter and GPS systems are imprecise (i.e. +/- 1m), an important development will be the integration of a lightweight RTK-GPS in the RPAS that will be synchronized with the RPAS GPR and spectral sensors. This will allow for an accurate estimation of the RPAS position, heading and altitude with centimetre precision and will inform the georeferencing process of the sensor data. As described above, the time-lapse spectral measurements will be synchronized with the GPR field campaigns.
WP3: GPR data processing
GPR data will be processed using full-wave inversion of the radar data, using the approach of Lambot et al. (Lambot et al. 2004a, Lambot et al., 2006a, Lambot and André 2014). This method inherently maximizes radar information retrieval capabilities in terms of both quantity and quality, relying on an intrinsic solution of Maxwell’s equations fully describing the antenna and its coupling with a 3D planar layered medium. The near-field generalization of the method is patented PCT/EP2012/055416 (WO 2012/130847 A1 - “Method and device for characterization of physical properties of a target volume by electromagnetic inspection” by S. Lambot (UCL)). In this project, the optimal radar measurement height to be defined will be either in the near- of far-field. Inversion will be focused in the time domain on the surface reflection to estimate soil surface permittivity, with a characterization depth depending on the considered frequency range. Information on roughness will be obtained following a frequency analysis based on Rayleigh criterion (Lambot et al. 2006c) as well as using inversion in the frequency domain using the recent model of Jonard et al. (Jonard et al. 2012). Time-lapse analyses of the maps will permit to strengthen the estimates and provide insights in temporal stability of soil moisture patterns in relation to soil topography and type. Calibration of the antenna will be either performed with the method of Lambot et al. (2004) of Lambot and André (2014), depending on the antenna-soil distance. In all cases, it will involve measurements at different heights over a water plane (lake) for the lowest frequency range and at different distances from a copper plane in laboratory conditions for the highest frequencies.
WP4: Hyperspectral data processing
The soil moisture and roughness information provided by the GPR system will be used to improve our understanding of these effects on soil spectral data. In this WP, we aim to construct a stratification of hyperspectral datasets to provide finer estimates of key soil physico-chemical properties whereby we will mainly focus on soil organic carbon content. To this end, we will compare the accuracy of soil organic matter prediction models (building on the work of Aldana-Jague et al. (In Prep) before and after the application of a stratification methodology based on (i) a spectral index (i.e., using only optical information; e.g., Haubrock et al, 2008) and (ii) on the RPAS-GPR measurements (WP3). The stratification approach will be informed by the empirical observations in our study sites (WP1 & WP2).
Finally, we will explore the possibility to improve the prediction of soil organic matter from airborne hyperspectral campaigns (BELAIR Hesbania study site).
WP5: Prospective RPAS-GPR soil moisture up-scaling
This WP will be dedicated to tentative up-scaling of surface soil moisture derived from GPR based on Sentinel-1 data. The global idea is to use the GPR measurements as sample measurement or probe areas to calibrate the SAR data and then extend the measurements to similar zones at a regional scale, including GPR validating test sites and/or test dates. To this purpose, two kinds of Level-1 IW Sentinel-1 data products will be used; SLC and GRD products. SLC are Single Look Complex products issued from SAR focalization. They are still in acquisition geometry (slant range – azimuth) and contains both the amplitude and the phase of the signal. This latter information is the one used to perform interferometric measurements. GRD products are ground projected detected values (backscattering amplitude). Ground projection is performed on a flat Earth surface and does not take local topography into account. Data is over-sampled on a regular grid of 10 x 10 meters for a multilook resolution of about 20 x 20m. For both products, Look-Up Tables (LUT) are provided to convert digital numbers in backscattering values (sigma nought). In both cases, a first step will consist in geoprojecting Sentinel-1 data to locate GPR measurements within the images. The nowadays CSL InSAR Software (CIS) allows handling fully SLC data and will be adapted to handle GRD products in order to geoproject them conveniently using an external DEM. Second, Sentinel-1 image time series will be considered, considering the first one as reference.
This reference will be chosen with an acquisition date corresponding to the GPR measurement campaign.
Then, taking advantage of results issued from SENSAR project funded by Belspo, reference image will be calibrated in order to associate soil moisture and backscattering values. Areas for which GPR measurements are available will then be delineated and characterized with respects to their statistics (average backscattering value and variance). If working with SLC data, interferometric coherence on successive data of the time series will also be used to characterize parcels (coherence level). Areas having both the same statistics within the reference image and the same coherence level in the first available interferometric pair will be considered as identical with respect to both humidity and roughness. This will allow extending GPR measurements spatially. Areas keeping a same high coherence from time to time (Sentinel-1 revisit time is of 12 days) will be considered as stable with respect to roughness, while areas loosing coherence with time can for example be subject to vegetation or global phenologic stage variation. For those areas that can be considered as stable with respect to roughness, model issued from SENSAR can be used and allow extending GPR maps within the time series. In this way, both spatial and temporal up-scaling of humidity measurements issued from GPR campaign can be tentatively performed. Validation can be performed spatially keeping one or several GPR measurements probe points as validation values. Temporal validation can be performed if a second GPR campaign can be made available on the same zones.
WP6. Dissemination of results
Throughout the project, the results will be presented in conferences and scientific events. The results will also be shown on the project website (http://sites.uclouvain.be/gprlouvain/projects.html) and made available to potential users (subject to a specific agreement during the project). Three original journal papers are envisaged during the project (one per partner): (1) Validation of RPAS-GPR for root zone soil moisture mapping, (2) Calibration of Sentinel-1 SAR products for soil moisture using RPAS-GPR, (3) Calibration of the hyperspectral model using GPR soil moisture and roughness estimates.
EXPECTED SCIENTIFIC RESULTS
The anticipated results and corresponding deliverables for each WP are as follows:
WP1: Setup of the RPAS-GPR. Time-lapse, low and high-frequency GPR data over the test sites. Groundtruths for root zone and surface soil moisture and soil surface roughness.
WP2: Setup of the RPAS based spectrometer and positioning/orientation accessories. Hyperspectral
data over the test sites, corresponding in particular to the GPR campaigns. Ground-truth samples for soil carbon content.
WP3: GPR root-zone and surface soil moisture maps acquired with the RPAS system.
WP4: Hyperspectral soil carbon content maps acquired with the RPAS system.
WP5: GPR-calibrated Sentinel-1 soil moisture maps.
WP6: Project website, data available for potential users, publications.
EXPECTED PRODUCTS AND SERVICES
Potential services subsequent to RAPAS are the RPAS-based mapping of soil properties to support precision agriculture (e.g., irrigation, soil property assessment) and hydrological & soil eco-system modelling through the improvement of Sentinel-1 surface soil moisture products.
If Sentinel-1 tentative up-scaling reveals to be sufficiently efficient, expected products are regional soil moisture mapping and soil moisture monitoring. Therefore, if successful this kind of product can be used to develop soil moisture monitoring services.
