Quantitative assessment of soil contaminants, particularly hydrocarbons, using reflectance spectroscopy

Abstract

Apparatus and method for efficiently assessing the results of reflectance spectroscopy on a soil sample to determine the presence of contaminants in the soil, by constructing a model based on analysis of known samples. The model may be constructed using an all possibilities approach and data mining techniques, on a range of samples, for example of different kinds of soil without pollutants and with different levels of pollutants. The Disclosure relates both to the construction of the model and to its use in the field in analyzing soil contaminants.

What is claimed is: 
     
         1 . Method for quantitative assessment of hydrocarbon contamination in soil using reflectance spectroscopy, the method comprising:
 obtaining a plurality of soil samples, and having respective hydrocarbon contaminations;   spectrally analyzing each sample to obtain spectroscopy data of said respective sample;   applying to said sample spectroscopy data a plurality of combinations from a set of preprocessing operations, each combination generating a respective mutation of the sample spectroscopy data;   using all of said mutations, building a model relating said preprocessed spectral data to respective contaminations, and   using said model with spectroscopy data of soil samples of unknown contamination to determine said unknown contamination.   
     
     
         2 . The method of  claim 1 , wherein said building a model comprises:
 using said mutations, extracting latent variables;   retrieving a first latent variable and predicting said different contaminations;   iteratively adding further latent variables, repeating said prediction and retaining respective latent variables if an improvement is measured in said prediction until an optimal number of latent variables that improve prediction is included, thereby providing a stable model.   
     
     
         3 . The method of  claim 2 , further comprising verifying said model, prior to said using, by assessing said respective initial model results by further predicting of contaminations of a verification set of additional soil samples with known contaminations. 
     
     
         4 . The method of  claim 1 , wherein said set of preprocessing operations comprises initial smoothing, multiplicative scatter correction, signal normal variate, absorptance, continuum removal, first derivative, second derivative, and final smoothing. 
     
     
         5 . The method of  claim 4 , comprising using between a third and a half of all possible combinations of said preprocessing operations. 
     
     
         6 . The method of  claim 4 , comprising using all possible combinations of said preprocessing operations of said set except for combinations that contain mutually incompatible preprocessing operations and combinations that provide results that are complex numbers. 
     
     
         7 . The method of  claim 1 , wherein said building said model further comprises using a model performance scoring parameter (MPS) defined as:
     MPS= %Stability+ NRPD−NRMSEP−NnLV     wherein said % Stability is obtained by dividing a number of stable models by a number of repetitions for each mutation, said NRPD is a normalized average of a ratio of prediction to deviation, said NRMSEP is a normalized root mean square error in prediction, and said NnLV is a normalized optimal number of latent variables.   
     
     
         8 . The method of  claim 1 , wherein said spectroscopy comprises visible or near infra-red spectroscopy. 
     
     
         9 . The method of  claim 8 , wherein said spectrally analyzing is carried out within a wavelength range lying between 350 and 2500 nm. 
     
     
         10 . The method of  claim 1 , wherein said set of different data mining operations comprises both linear and non-linear algorithms. 
     
     
         11 . The method of  claim 10 , wherein said set of different data mining operations comprises artificial neural networks, genetic algorithms, support vector machines, fuzzy logic, partial least squares, multiple linear regression, metric learning and principle component regression. 
     
     
         12 . The method of  claim 1 , further comprising arranging said pre-processed spectroscopy data into datasets using conditioned Latin Hypercube Sampling. 
     
     
         13 . The method of  claim 1 , comprising taking said plurality of soil samples from a plurality of soil types. 
     
     
         14 . A testing device for testing soil samples for contamination, the testing device comprising:
 a spectrometer or spectral imager;   a model relating a set of spectral parameters obtained from spectroscopy data to contamination properties of respective samples, the spectral parameters being obtained from said spectroscopy data following mutation of said spectroscopy data using a plurality of combinations of spectral preprocessing operations; and   an output for indicating contamination properties predicted by said model.   
     
     
         15 . A modeling device for building a model for testing soil samples for contamination by hydrocarbons, the modeling device comprising:
 a spectrometer or spectral imager;   a preprocessor, configured to apply to said sample spectroscopy data a plurality of combinations from a set of preprocessing operations, each combination generating a respective mutation of the sample spectroscopy data; and   a modeler, configured to use said mutations to build a model relating said preprocessed spectral data to respective contaminations.