METHOD FOR SOIL POLLUTION ANALYSIS

Description

[0001] The present invention relates to the field of soil contamination analysis by organic pollutants.

[0002] Sources of soil contamination are varied: the use of fertilizers, pesticides, discharges from an industrial site, proximity to an incinerator, a waste storage site, discharges of drug residues from livestock manure, hydrocarbons... Soil, land, garden or playground pollution can come from chemical agents: dioxins, PCBs or extremely dangerous toxic metals.

[0003] The origin of these pollutions can be accidental (spills or occasional deposits of pollutants due to negligence, malfunction of an industrial installation, accident of an installation or of a vehicle transporting polluting materials), with a large quantity of pollutant spilled, or chronic (continuous supply of contaminants by leakage or leaching, the cumulative effects of which can be greater and more insidious than those of an accidental pollution).

[0004] Polluted soil can cause poisoning when consuming fruits and vegetables from the garden. The bioaccumulation of pollutants in soils by plants and animals makes them among the most dangerous pollutants for human and animal health. They have the particularity of contaminating the food chain (dioxins, PCBs, radioactivity, etc.). Soil pollutants can also cause skin and respiratory irritation. They are also responsible for cardiac and neurological disorders, loss of fertility, developmental problems in the fetus, and are a contributing factor to certain cancers.

[0005] To assess the presence of contamination and to identify and quantify the nature of the pollutants, it is common practice to take samples, for example by coring, and submit these samples to a physicochemical analysis laboratory. The "Methodological Guide for the Analysis of Polluted Soils," published in February 2000 under reference BRGM / RP-50128-FR, presents a detailed description of the techniques for analyzing polluted soils.

[0006] Because these analyses require high-level scientific resources and skills, it has also been proposed to automate all or part of these analyses.

[0007] For a given site, the standard steps for soil analysis involve: 1. Bring in a survey company to take core samples; 2. Send samples to analysis laboratories; 3. Wait for the analysis results; 4. Interpret these analyses in the form of maps; 5. Write a recommendation report.

[0008] The minimum duration for such an analysis is 6 to 8 weeks. Following these analyses, the report often reveals significant uncertainties due to heterogeneity in the soil studied, thus recommending a new sampling campaign. Indeed, the initial sampling is often insufficient because the distances between sampling points are too great (cost considerations). It is then necessary to repeat the process described in the previous problem, one or more times. The delay caused by this issue is at least one month.

[0009] Finally, during the remediation process, the site is excavated by the remediation company, and unforeseen contamination may be detected. This leads to a work stoppage, and the process resumes with re-sampling and analysis as described above. Meanwhile, the excavated material is temporarily stored on-site until the nature of the contamination is determined and it is sent to the appropriate treatment facilities. In this case, an additional delay of at least 15 days is observed, along with significant extra costs related to the downtime of workers and machinery, as well as the reprocessing of the initially undiagnosed contaminated soil. State of the art

[0010] The prior art is known in the article by Bin Zou, Xiaolu Jiang, Huihui Feng, Yulong Tu, and Chao Tao, "Multisource spectral-integrated estimation of cadmium concentrations in soil using a direct standardization and Spiking algorithm," published in Science of the Total Environment, Volume 701, 2020, 134890, ISSN 0048-9697. This article concerns the field of low-level and large-scale satellite remote sensing, and more specifically, the study of the exact spectral response of cadmium (Cd) in soil. It presents a new method combining direct standardization (DS) and Spiking algorithms to integrate multisource spectra in order to improve the accuracy of Cd concentration estimation. This article focuses on the analysis of the presence of the heavy metal cadmium in soil samples.This article offers no insights into the characterization of unknown pollutants, particularly organic pollutants, in a core sample taken in the field. We are also familiar with the publication "Scafutto, Rebecca & Souza Filho, Carlos. (2016). Quantitative characterization of crude oils and fuels in mineral substrates using reflectance spectroscopy: Implications for remote sensing. International Journal of Applied Earth Observation and Geoinformation. 50. 221-242. 10.1016 / j.jag.2016.03.017." concerning the environmental monitoring of oil and fuel leaks using multispectral, hyperspectral, and ultraspectral proximal and far-field remote sensing.This publication is based on the measurement of the near- and short-wave infrared spectral reflectance properties of several mineral substrates impregnated with crude oil, diesel, gasoline, and ethanol using principal component analysis (PCA) and partial least squares (PLS) regression. These characteristics were used for the qualitative and quantitative determination of the impregnated contaminant in the substrates. Specific wavelengths, where key absorption bands occur, were used for the individual characterization of the oils and fuels. The intensity of these characteristics can be correlated with the abundance of the contaminant in the mixtures. The grain size and composition of the impregnated substrate directly influence the variation of the spectral signatures. Also known is the article "Kopel, Daniella & Brook, Anna & Wittenberg, Lea & Malkinson, Dan. (2015). Spectroscopy as a Diagnostic Tool for Urban Soil."Water, Air, and Soil Pollution. 226. 10.1007 / s11270-015-2442-2. » concerning spectral activity (SA) detection in a structured hierarchical approach to identify dominant spectral features. The developed method is adopted by several production tools, including continuum suppression normalization, guided by polynomial generalization, and spectral likelihood algorithms: orthogonal subspace projection (OSP) and iterative spectral mixing analysis (ISMA).

[0011] We also know the article "Estimating the soil clay content and organic matter by means of different calibration methods of vis-NIR diffuse reflectance spectroscopy"

[0012] Said Nawara, et al. Soil & Tillage Research 155 (2016) 510-522, concerning the estimation of soil clay and organic matter content using visible and near-infrared reflectance spectroscopy (VNIRRS). The main objective is to optimize calibration methods to improve the accuracy of soil property predictions. Three multivariate regression techniques are compared: partial least squares regression (PLSR), support vector regression (SVR), and multivariate adaptive regression splines (MARS).

[0013] The study is carried out on 102 soil samples collected in the El-Tina plain in Egypt, scanned with a portable spectroradiometer covering a spectrum from 350 to 2500 nm. Different spectral preprocessing methods are evaluated, including Savitzky-Golay smoothing, spectral derivation and normalization. Disadvantages of prior art

[0014] Prior art solutions do not allow for a sufficient level of accuracy and reliability when the analysis is carried out directly on site by a spectrometric method and in particular by hyperspectral imaging.

[0015] Furthermore, to meet field needs, it is important to be able to provide qualified information on pollutant content according to depth, in order to optimize field treatment, and solutions based on the analysis of images taken remotely are poorly suited. Solution provided by the invention

[0016] To overcome these drawbacks, the invention, in its most general sense, relates to a method for analyzing soil contamination by pollutants, particularly organic pollutants, as defined in claim 1. The method according to the invention comprises: a training sequence consisting of analyzing a plurality of reference samples and recording in a training database: a) the spectral reflection signature acquired by spectral analysis; b) the known values ​​of variables representing the contaminants present in each of said reference samples; c) the known values ​​of variables representing the substrates of each of said reference samples; a calibration sequence of field analysis equipment with respect to said first equipment, said field equipment comprising a light source and a spectral sensor; analysis sequences of a soil sample from a geological site consisting of acquiring the reflection and photoluminescence signature of said sample using said field equipment thus calibrated.and to proceed with the estimation of the pollutant characterization by processing said signature with a machine learning engine using data from the database created during the training sequence.

[0017] Advantageously the analysis system according to the invention comprises a probe having at least one optical fiber to transmit light between the analysis area of ​​said probe on the one hand, and the light source, in particular Xenon or Halogen, and said at least one sensor on the other hand.

[0018] According to an optional variant, the said equipment also includes a means of physico-chemical analysis. Description of a non-limiting example of the invention

[0019] The present invention will be better understood upon reading the following description, with reference to the attached drawings concerning non-limiting examples of embodiment, where: [ FIGURE 1 ] there figure 1represents the hardware architecture of an example implementation of the invention [ FIGURE 2 ] there figure 2 represents the functional diagram of the invention [ FIGURE 3 ] there figure 3 represents the functional architecture [ FIGURE 4 ] there figure 4 represents the optical diagram of a variant embodiment of the optical system. General context of the invention

[0020] The invention relates to the characterization of soil samples by coring to determine the qualitative and quantitative presence of constituents of interest (pollutants, including plastics, hydrocarbons, metals...) to provide a real-time measurement tool for the nature of pollutants on polluted land recovery sites.

[0021] The goal is to provide a solution that, ideally in real time and on-site, delivers a detailed pollution map to facilitate and secure the implementation of selective sorting of excavated material based on its pollution level, directly on the construction site using a real-time measurement tool. The result is a three-dimensional map of the terrain providing geolocated information on pollution, in x,y coordinates with a resolution ranging from a few square meters to several hundred square meters, and in z-depth, enabling informed decisions regarding optimal remediation measures. Optionally, core sampling analysis can be supplemented by remote analysis using a hyperspectral camera to quantify pollutants delivered to analysis centers for real-time monitoring of the contaminated soil they receive.Thus, the proposed system guarantees the quality of the soil, identifies its pollutants, in order to allow its recycling.

[0022] To this end, the invention enables the instrumentation of drilling machines and construction equipment to allow for the integration of real-time analyses and to accelerate the process by eliminating iterative steps and sources of financial uncertainty.

[0023] Thus, when pollution not initially identified is detected, real-time measurements would allow the excavated material to be directed to the corresponding recovery channel without the delay of laboratory analysis (optimization of real-time decision-making). Hardware architecture

[0024] The equipment according to one embodiment of the invention consists of a field device (1), portable in the described example, comprising a backpack-type case containing a broad-spectrum light source, for example a xenon lamp (10), and a power supply, for example batteries. The light source (10) is connected by an optical fiber (2) to a lance (3) which transmits light to the ground and reflects the light to sensors (11) whose electrical signals are transmitted to a computer which can be housed in the case or attached in the form of a tablet (4) to the handle (5) of the lance (3).

[0025] The sensors are hyperspectral sensors with a sensitivity range between 100 microns and 200 nanometers. They can be a hyperspectral camera, a multispectral camera, or an assembly of sensors forming a composite multispectral sensor.

[0026] The lance includes one or more optical fibers to transmit the light emitted by the source (10) to a measuring end, and to collect the reflected light or fluorescence light from the sample towards the end of the lance (3). The fiber may include a lens or collimating optics at its end.

[0027] The tablet (4) has a touch interface (6) equipped with a geolocation module (13) (GPS or Galileo) and a 5G SIM mobile phone card.

[0028] This equipment also includes a computer (12) and radio communication means that allow, firstly, the processing of data and the provision of a real-time diagnosis of soil condition and pollution (lithological characteristics, type of pollutants, quantity of pollutants, 3D soil mapping, etc.); and secondly, the transmission of data to a cloud storage space (30). The data is available locally for real-time decision-making but also on an online platform to allow the project manager, equipped with a connected terminal (20) and located at the control center, to monitor operations live and collect data acquired in the field. Implementation of the invention

[0029] A core sample is taken by an operator equipped with the aforementioned equipment at a site suspected of contamination. The drilling location is probed by coring or tapping with an instrument equipped with a geolocation module, and its geographic coordinates are recorded. Several sections approximately 1 meter deep are taken at the drilling site, and multiple boreholes are drilled at the location.

[0030] Spectral analysis of all core samples is performed. This involves removing the lengthwise split half-cores and acquiring data from several simultaneous half-cores, made possible by the rapid image acquisition speed. Extracting sub-images corresponding to a half-core is facilitated by the placement of digitally recognized QR codes at the corners of the sections. Imaging can be performed on-site using a camera in a vehicle or container. A pre-trained predictive model can then be used, provided there is sufficient data in the database for the model to produce a real-time diagnosis.

[0031] Subsample selection is performed using imaging data. Statistical methods, both supervised (Machine Learning) and unsupervised (end-member extraction, feature extraction, novelty detection, etc.), are used to select areas for subsampling in order to represent the heterogeneity of the geological formations present at the site. Spectrometer data can be used as an alternative. Subsampling can be carried out immediately (in the case of volatile pollutants) or later in the laboratory (the half-cores are hermetically sealed with plastic film and stored in a cold room for preservation).

[0032] Chemical analyses involve extracting pollutants using various methods, including solvent extraction (water, hexane, ether), agitation, solid-phase microextraction, microwave extraction, headspace extraction, etc. Analytical techniques include chromatography (ICP-AES, ion chromatography, HS-GS / MS). Analyses are performed on-site (in a container) or in the laboratory, on subsamples of core samples or on provided and / or non-selectable samples (e.g., those collected by auger).

[0033] Predictive model training is performed using a database and reference samples. The first processing step involves converting raw data from the reflectance sensor. This step is called normalization, referring to the current method which uses raw data measured on a reference material with >99% reflectance (Spectralon®) and electronic noise data measured without an illuminant (source) to normalize the sample data between these two spectra (i.e., 0 and 100% reflectance). To eliminate these measurements prior to data acquisition, a model is generated from the prior recording of this raw reference data. Raw data measured on eight reference materials (from 2% to 99% reflectance) and electronic noise are used; a model can be trained for each combination of device parameters.A second processing level predicts variables of interest (soil composition, presence of pollutants, and pollutant quantity) based on the reflection of a sample. Several training databases are used: published training databases (pure spectral compound libraries, e.g., the USGS Spectral Library), data produced on artificial samples produced in the laboratory, or data produced on samples analyzed in the laboratory. If a batch of analyzed samples comes from a specific site, a model can be trained on that batch only, or that batch can be used to improve a pre-trained model based on an existing database using transfer learning.

[0034] Data processing is applied to imaging data. Data from analyzed subsamples allows for the interpretation or refinement of an initial interpretation of the core samples. The model generated with the spectrometer data enables real-time analyses, including analyses referenced to data from COFRAC-certified laboratories. The coupling of on-site imaging and spectrometry addresses both the diagnostic phase of a site and the subsequent construction phase. Selective sorting of excavated soil based on its waste classification is possible. Quality control of recycled soil at the landfill is performed using provided samples, either artificial or semi-artificial (samples provided diluted or artificially enhanced to cover a wider range of concentrations and improve the model).

[0035] Mapping the results constitutes a third level of interpretation based on a machine learning model. Interpolation of variables of interest measured at different boreholes in the space of the map generates a map. Gaussian process modeling methods are used. Geophysical data measured at the time of drilling are used in the mapping via data fusion methods.

[0036] In-situ measurements can be considered. Data acquisition using a fiber optic spectrometer allows for probing the soil and producing a profile of variables of interest as a function of depth. (Note: a distinction has been made between in-situ and on-site: Imaging can be carried out on-site with a camera in a vehicle or container. The term "in-situ" refers to the use of equipment including a fiber optic probe with one or more portable spectrometers, as previously described.) Functional diagram

[0037] The first step (100) consists of taking a core sample of a length determined by the depth of soil to be analyzed using the probe (4). The core sample is geolocated by the GPS module (13) of the equipment.

[0038] The next step (200) consists of carrying out an analysis of the core over the length of the core by measuring the hyperspectral reflection measured by the sensors (11) when illuminating with the source (10).

[0039] The data obtained during the analysis step are recorded locally and in the cloud, and are then subjected to a step (300) of selecting a subset of samples to perform either model training (steps 400, 500, 600), or on-site analysis (steps 450, 550, 650). Variant of the learning stage

[0040] Learning can be shared from laboratory analyses, with equipment equipped with a high-performance hyperspectral camera, to record the spectral signatures of a large number of reference samples, and provide a database accessible to a plurality of field equipment equipped with less powerful and less expensive sensors.

[0041] To account for technical and optical differences, each field instrument is calibrated using reference samples whose spectral signature has been previously recorded in the database. A correction function is then calculated to allow the database content to be used with equipment different from that used for the initial analysis.

[0042] The samples are distinguished on the one hand by the nature of the substrate, and on the other hand by the nature of the pollutants present.

[0043] The substrates are characterized by meta-descriptors based on variables such as: The chemical nature of the mineral and organic constituents, the water content, the oxide content, the pH, the particle size, the membership of one or more mineral classes according to the Strunz classification, the redox potential.

[0044] The reference substrate can be characterized by physicochemical analyses. It can also be prepared from predetermined components to prepare substrates by assembly.

[0045] Reference pollutants are characterized by their chemical composition.

[0046] Next, for each reference sample, the spectral signature is recorded by exposing it to illumination from a light source, such as a xenon lamp. Both reflected light and photoluminescent light are captured within a wavelength range extending from thermal infrared to ultraviolet UVC. The data are recorded for each sample along with a reference sample identifier and its physicochemical characteristics.

[0047] According to a preferred alternative, spectral acquisition of the sample or the entire core is carried out first, then a (sub-)sample (or several) is extracted for physico-chemical analysis. Spectral acquisition

[0048] There figure 4This represents the optical diagram of an alternative embodiment of the optical system. The configuration consists of two separate channels operated by a fiber bundle connected to a xenon lamp (10) which irradiates soil samples (60) and collects the reflected light in each channel by means of an optical switch (50). A monochromator (51) placed in the optical path provides a secondary beam for fluorescence excitation.

[0049] The first reflectance channel is designed to collect photons simultaneously on two separate spectrometers: A spectrometer (61) in the visible and ultraviolet band. A spectrometer (62) in the infrared band. The second channel is intended for the collection of fluorescence photons in the UV-Visible-PIR range: the monochromatic incident light is selected by means of a monochromator connected to the Xenon lamp and the photons are collected on the UV-Visible-PIR spectrometer (61).

Claims

1. A method for analyzing, by means of hyperspectral analysis of a reflection and photoluminescence, soil contamination by pollutants, in particular organic pollutants, in order to determine the qualitative and quantitative presence of constituents of interest constituting pollutants and comprising in particular plastics materials, hydrocarbons and metals, characterized in that said analysis is carried out by means of a first item of equipment by illuminating a sample using a light source and by at least one spectral sensor sensitive to a spectrum ranging from thermal infrared to ultraviolet, and in that it includes: - a sequence for learning, consisting in analyzing a plurality of reference samples, and for recording the following in a learning database ∘ a) the spectral signature of reflection acquired by spectral analysis o b) known values of variables representative of the contaminants present in each of said reference samples ∘ c) known values of variables representative of substrates of each of said reference samples - a sequence for calibrating a piece of field analysis equipment with respect to said first piece of equipment, said piece of field equipment including a light source and a spectral sensor, - sequences for analyzing a soil sample from a geological site, consisting in acquiring the reflection and photoluminescence signature of said sample using said piece of field equipment thus calibrated, - and in estimating the characterization of pollutants by processing said signature by means of a learning engine that uses the data from the database created during the learning sequence.

2. The method for analyzing soil contamination by pollutants, in particular organic pollutants, according to claim 1, characterized in that, during the analysis of a site, at least one core sampling operation is carried out, and in that the analysis of a plurality of samples distributed over the height of the core is carried out in order to characterize the contaminants at various depths.

3. The method for analyzing soil contamination by pollutants, in particular organic pollutants, according to claim 1, characterized in that it further includes steps of physical and / or chemical analysis of at least some of said samples.