METHOD AND APPARATUS FOR DETECTING CHEMICAL COMPOUNDS IN SOIL

Abstract

A spectrometer probe is disclosed herein including a shaft having a first end and a second end, a fiberoptic bundle located within the shaft, the fiberoptic bundle having a first end and a second end, a mirror, a transparent window, a prism, a prism support, an elastically deformable material, an index-matching elastomer, wherein the prism is completely encompassed by the index-matching elastomer, and a penetration cone operatively attached to the second end of the fiberoptic bundle, the mirror located within the second end of the shaft, wherein the transparent window is substantially parallel with the fiberoptic bundle and the shaft, wherein the prism is angled at approximately a 45 degree angle in relation to the window and the fiberoptic bundle, wherein the prism is flush with the prism support, wherein the elastically deformable material is biasly connected to the prism support.

Claims

1. A spectrometer probe comprising: a shaft having a first end and a second end; and a fiberoptic bundle located within the shaft, the fiberoptic bundle having a first end and a second end, wherein the shaft has no light source or spectrocope located within the shaft, wherein there are no air gaps in the shaft.

2. The spectrometer probe of claim 1, further comprising: an elastically deformable material, wherein the elastically deformable material is biasly connected to a mirror support; and a mirror, wherein the mirror is located within the second end of the shaft.

3. The spectrometer probe of claim 2, wherein the mirror is angled at approximately a 45 degree angle in relation to the fiberoptic bundle.

4. The spectrometer probe of claim 3, further comprising: an index-matching elastomer, wherein the mirror is completely encompassed by the index-matching elastomer.

5. The spectrometer probe of claim 4, further comprising a penetration cone operatively attached to the second end of the fiberoptic bundle.

6. The spectrometer probe of claim 4, further comprising: a transparent window, wherein the transparent window is substantially parallel with the fiberoptic bundle and the shaft, wherein the mirror is angled at approximately a 45 degree angle in relation to the transparent window and the fiberoptic bundle; a probe sample window; and a light reflection insert, wherein the light reflection insert is attached between the probe sample window and the first end of the shaft.

7. The spectrometer probe of claim 4, further comprising: an outer sleeve having a first end and a second end, wherein the outer sleeve surrounds the shaft; an endcap attached to the first end of the outer sleeve; and a spacer tab, wherein the spacer tab is connected between a probe sample window and the end cap.

8. The spectrometer probe of claim 4, wherein the mirror is concave, wherein the concavity of the concave mirror is facing the second end of the fiberoptic bundle, wherein the transparent window is substantially parallel with the fiberoptic bundle and the outer sleeve.

9. The spectrometer probe of claim 8 further comprising an index-matching elastomer, wherein the mirror is completely encompassed by the index-matching elastomer, wherein the index-matching elastomer is silicone elastomer.

10. The spectrometer probe of claim 8, wherein the fiberoptic bundle is bonded to the outer sleeve with epoxy, wherein there are no air gaps and no etaloning.

11. The spectrometer probe of claim 1, further comprising: an elastically deformable material, wherein the elastically deformable material is biasly connected to a prism support; and a prism, wherein the prism is located within the second end of the shaft.

12. The spectrometer probe of claim 11, wherein the prism is angled at approximately a 45 degree angle in relation to the fiberoptic bundle.

13. The spectrometer probe of claim 12, further comprising: an index-matching elastomer, wherein the prism is completely encompassed by the index-matching elastomer.

14. The spectrometer probe of claim 13, further comprising a penetration cone operatively attached to the second end of the fiberoptic bundle.

15. The spectrometer probe of claim 14, wherein the prism is flush with the prism support, further comprising: a backscatter baffle having a first end and a second end, wherein the second end of the backscatter baffle extends beyond the second end of the fiberoptic bundle.

16. A method for determining chemical compositions in soil, the method comprising the steps of: providing a spectrometer probe comprising: a portable carrier; an actuator; a shaft having a first end and a second end, wherein the shaft has no light source or spectroscope located within the shaft; a fiberoptic bundle located within the shaft, the fiberoptic bundle having a first end and a second end; a penetration cone operatively attached to the second end of the fiberoptic bundle; using the actuator to insert the penetration cone into associated soil to a predetermined depth in the soil; passing light through the fiberoptic bundle; receiving light diffusively reflected back from the soil; and analyzing the diffusively reflected light with a spectrometer.

17. The method of claim 16, the spectrometer probe further comprising: a mirror or a prism; an elastically deformable material, wherein the elastically deformable material is biasly connected to a mirror support or prism support; an index-matching elastomer, wherein the mirror or prism is completely encompassed by the index-matching elastomer, wherein the mirror or prism is flush with the mirror support or prism support; and a backscatter baffle having a first end and a second end, wherein the second end of the backscatter baffle extends beyond the second end of the fiberoptic bundle.

18. The method of claim 17, the spectrometer probe, further comprising: a transparent window, wherein the transparent window is substantially parallel with the fiberoptic bundle and the shaft, wherein the prism is angled at approximately a 45 degree angle in relation to the transparent window and the fiberoptic bundle; a probe sample window; and a light reflection insert, wherein the light reflection insert is attached between the probe sample window and the first end of the shaft.

19. The method of claim 18, the spectrometer probe, further comprising: an outer sleeve having a first end and a second end, wherein the outer sleeve surrounds the shaft; an endcap attached to the first end of the outer sleeve; and a spacer tab, wherein the spacer tab is connected between the probe sample window and the end cap.

20. The method of claim 19, wherein the index-matching elastomer is silicone elastomer, wherein the fiberoptic bundle is bonded to the outer sleeve with epoxy, wherein there are no air gaps surrounding the prism and no etaloning.

Description

III. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present teachings are described hereinafter with reference to the accompanying drawings.

[0027] FIG. 1 shows a spectrometer probe;

[0028] FIG. 2A shows a cross-section of the probe;

[0029] FIG. 2B shows a cross-section of the probe;

[0030] FIG. 3 shows an exploded view of the probe;

[0031] FIG. 4A shows a cross-section of the probe;

[0032] FIG. 4B shows an enlarged cross-section of the probe end;

[0033] FIG. 5 shows a prospective view of a probe carrier; and

[0034] FIG. 6 shows a 3D map.

IV. DETAILED DESCRIPTION

[0035] With reference now to FIG. 1, spectrometer probe 10 is shown having an outer sleeve 12, a protector endcap 14, a spacer tab 16, a probe sample window 18, a light reflection insert 20, a top spacer tab 22, a protector tube top cap 24, near-infrared (NIR) probe cable 26, read fiberoptic cable 28, and light source fiberoptic cable 30. FIG. 2A shows a cross section of probe 10 having fiber bundle 34, concave mirror 36, backscatter baffle 38, and sapphire window 40. FIG. 2B shows a cross section of probe 10 having fiber bundle 34, index-matching elastomer 42, sapphire prism 44, and penetration cone 46. FIG. 3 shows an exploded view of the probe 10 showing the outer sleeve 12, the sapphire prism 44, the penetration cone 46, support prism 48, sapphire window 40, set screws 52, inner sleeve 66, inner sleeve top 54, grooves 56, and set screws 58. FIG. 4A shows the fiber bundle 34, the sapphire prism 44, and epoxy 60 to bond the fibers in the tube. FIG. 4B shows the penetrating cone 46, the sapphire prism 44, a prism support 64, and spring 62.

[0036] With continuing reference to FIGS. 1-4B, the read fiberoptic cable 28 and light source fiberoptic cable 30 are located next to each other, but combine to form the fiberoptic bundle 34 that passes through the NIR probe cable 26 and down through the probe 10 inside outer sleeve 12. The NIR probe cable 26 is held in place by top spacer tab 22 and protector tube top cap 24. The fiber bundle 34 extends through the inner sleeve 12 to the protector end cap 14. At the end of the probe 10, near the penetration cone 46, there is the spacer tab 16, the probe sample window 18, and the light reflection insert 20. In one aspect of the present teaching, as shown in FIG. 2A, the probe 10 has a concave mirror 36, backscatter baffle 38, and sapphire window 40. Light from the light cable 30 travels through the fiber bundle 34, reflects off of the mirror 36 and travels through the sapphire window 40. The backscatter baffle 38 helps prevent backscatter from the light.

[0037] With continuing reference to FIGS. 1-4B, the probe 10 does not experience etaloning because of the solid-state optical path along with the index-matching elastomer 42 between the prism 48 and the prism support 64. The probe 10 has an inner sleeve 66 inside the outer sleeve 12. Near the inner sleeve top 54, grooves 56 have been cut into the inner sleeve 66 in order to fill with epoxy for better stabilization. The set screws 52, 58 also help stabilize the probe 10. In one aspect of the present teaching, a spring 62 is added behind the prism support 64 in order to create pressure, and allow fluctuation without having an air gap. Air gaps create wavelength interference and create spurious data. Additionally, index-matching elastomer 42 is utilized such that the prism 44 is completely surrounded by, and floating in, the index-matching elastomer fluid 42. The prism 44 is flush with the surface of the probe 10 and sealed. In one aspect of the presenting teaching the index-matching elastomer can be a silicone elastomer, but it is to be understood that any index-matching material can be chosen using sound engineering judgment. In one aspect of the present teaching, the spring 62 could be substituted with an elastically deformable material, such that pressure and fluctuation are still accomplished without using a spring.

[0038] With continuing reference to FIGS. 1-5, the probe 10 is connecting to a portable carrier 70, which has an actuator 72. The actuator 72 pushes the probe 10 into the ground. The penetrating cone 46 is designed to provide better case for ground penetration. The fiber bundle 34 takes and carries the light to and from the soil. The light is analyzed using diffuse reflective spectroscopy to detect various components in the soil. The probe 10 uses a reflective prism 44 and a side-looking probe sample window 18 to analyze the soil. The fiber bundle 34 can be anchored to the penetrating cone 46 at a single point instead of multiple points. The probe 10 can be made of stainless steel. In one aspect of the present teaching, spectra of the soil are taken every 2.5 cm to measure soil organic carbon. The spectra are thus taken at various depths of the soil. Light passes through the sapphire window 40 to strike the soil. Some light (I(?)) is diffusely reflected back to reach the spectrometer (not shown), which separates the different wavelengths (?) of the light. The intensity of the reflected light from the soil (I(?)) is compared to the light reflected from a white test blank (Io(?)). At each ?, the effective absorbance is calculated as A(?)=log.sub.10[I.sub.0(?)/I(?)]. A plot of the absorbance A(?) versus wavelength ? constitutes a spectrum and is the input data for the present teaching. In one aspect of the present teaching, the spectrometers are sensitive to the ranges 500-1100 nm, and 1100-2500 nm, which in tandem can sense nearly the entire visible and near infrared spectrum.

[0039] With continuing reference to FIGS. 1-5, a spectrometer (not shown) can be modular, above ground, and connected to the probe 10 with a fiberoptic connector. In one aspect of the present teaching, up to three spectrometer modules covering different wavelength regions could be connected via suitable wavelength splitter fiber cables.

[0040] With reference now to FIGS. 1-6, the total soil organic carbon can be determined using the above-referenced probe 10, by taking various samples at different depths and horizontal distances. Once the data are collected, a 3D map can be created to show the location of the soil organic carbon, and the relative concentration locations of the soil organic carbon.

[0041] With reference now to FIGS. 1-6, an algorithm is used to analyze all of the spectrometer data to create a 3D map of the soil organic carbon like the map shown in FIG. 6. The various concentrations can be shown in various colors using isosurfaces. 3D maps can be generated for many different chemicals aside from the soil organic carbon. The probe and spectrometers enable the rapid and accurate mapping of soil organic carbon in three dimensions.

[0042] In one aspect of the present teaching, for each site, the probe insertion locations for which soil property data existed are randomly divided into training (60% ), validation (20% ), and test (20% ) sets. All depth increments of a probe insertion are placed together into a set. The training set is used to develop the model, and the validation set is used to routinely check model performance. The test set is used to evaluate performance of the final model on novel data. Model fit is evaluated using root mean square error (RMSE). For comparison across sites, the ratio of performance to interquartile distance (RPIQ) is used, calculated by dividing the interquartile distance (difference between the 75.sup.th and 25.sup.th percentile values) by the RMSE. An RPIQ value >2.0 is often assumed to indicate excellent performance of a model. Also calculated is the coefficient of determination (R.sup.2).

[0043] In the modeling step, spectral data are preprocessed by resampling and normalization using the standard normal variate. Ancillary data derived from digital elevation models using standard geomorphometric indices are also included as model input. All input variables are standardized by removing the mean and dividing by the standard deviation. The highly multicollinear variable set is reduced using recursive feature elimination with cross-validation, which excludes variables that are least informative for a regressor. A comprehensive search is conducted through a number of regressors, including support vector regression, partial least squares regression, random forest regression, and AdaBoost regression, and hyperparameter settings using five-fold cross-validation with the training set, choosing the model with the lowest RMSE of cross-validation. For SOC concentration separate models are developed for each site and for the combination of all sites. For some sites BD is also modeled using the same methods.

[0044] For SOC concentration and stock, metrics for individual sites and for site combinations exceeded the rule of thumb value of RPIQ ?2.0 on the test set for model performance. Metrics for the SOC per-sample models also generally meet or exceed the goal of R.sup.2?0.8.

[0045] A per-sample model, applied to the full site data set, results in a lattice of measurements at each depth interval in each probe insertion. One method of visualizing the lattice is to map the soil property at regular depth intervals. Alternatively, the same data set can be used to visualize the soil property in 3D, as with the SOC concentration isosurfaces. Similar visualizations can be produced for any of the per-sample predicted soil properties.

[0046] SOC stock is estimated and mapped by summing the modeled per-sample SOC stock at each probe insertion to the maximum depth of the probe insertion.

[0047] Soil spectroscopy requires a database that widely samples the soil variability within the study area. The relationships between spectra and soil properties can be both spatially dependent and highly non-linear, and it is difficult to construct a calibration set that adequately reflects the immense variation found in soils. Establishment of the minimum change in per-site SOC stock detectable is tested, including a formal assessment of uncertainty associated with each sampling and modeling step.

[0048] A separate rubric can be developed for each target area. The target is that mean ?RMSE will be better than ?10%.

[0049] Data collection activities are organized around two different site types, intensive and extensive. Intensive sites are used to produce a single-site model, test accuracy and precision of SOC stock estimation, and create 3D maps of SOC stock, while the extensive sites are designed to efficiently extend calibration into previously uncovered portions of attribute space, even though in isolation they are not adequate for site-specific modeling or 3D mapping.

[0050] Initially, in a field of about 5 ha, probe insertions are in a grid pattern with a spacing of 15 m (?225 probe insertions and a density of ?50 ha.sup.?1). A soil core is obtained at about 25% of the probe insertions (?60 cores, or 360 samples at 6 depths/core). A stratified random sampling design is used to determine the grid points at which cores are taken.

[0051] At each extensive site, between 20 and 100 ha in size, 15 probe insertions are obtained and the corresponding 15 cores (?90 samples at 6/core). Specific locations are chosen by stratifying the site to attribute space characteristics, and randomly sampling five points within each of three strata. Density depends on site size and strata distribution.

[0052] At both site types, the probe is inserted to a depth of about 90 cm or the maximum depth allowed by the soil. Spectral data are acquired at 1 cm intervals in the surface 15 cm, at 2.5 cm between 15 and 60 cm; and at 5 cm intervals thereafter. Soil cores (3.8 cm diameter) are extracted in a plastic liner to a depth of 1 m or as deep as reached by the probe. Cores are divided at 7.5, 15, 30, 45, and 60 cm. After outlier eliminating spectra are matched to the corresponding depth interval and averaged within the interval for modeling. Samples are divided into training (60% ), validation (20% ), and test (20% ) sets using stratified random selection. To maintain independence of the validation and test sets, all samples from a given soil core are assigned to the same set.

[0053] For any given site, the baseline is the model trained only on data collected at that site. Regional and global models, trained on data from broader geographical regions, are evaluated with the usual accuracy metrics (RMSE, R.sup.2, and RPIQ), but also by comparing the RMSE of validation of the broader model to that of the local site model. The primary metric is ?RMSE, calculated as ((RMSE.sub.a?RMSE.sub.b)/(RMSE.sub.b* 100) where RMSE.sub.a is the RMSE of the broader model applied to the individual site validation set and RMSE.sub.b is the RMSE of the site-specific model. ?RMSE are continuously tracked at the site, regional, and global levels, with the target being ?RMSE (broad vs. site) better than ?10%. To track model performance on novel sites, the metric ?RMSE is calculated for models developed both including and excluding data collected at the site from the training set, targeting ?RMSE (excluded vs. included) better than ?10%. To establish the minimum change in per-site SOC stock that is detectable using the present teaching, for each of the intensive sites a formal analysis of measurement uncertainty is conducted, with a target of precision sufficient to detect an increase of 0.3 Mg C ha.sup.?1.

[0054] An exhaustive search is conducted through a variety of model types and hyperparameters, evaluating model performance using RMSE, R.sup.2, and RPIQ. Also included are (1) additional regressor types (e.g., 1D convolutional neural networks); (2) new methods of spectral preprocessing; (3) moisture correction; and (4) dimensionality reduction (e.g., wavelets). Also incorporated is ancillary data representing terrain parameters. The target metric will consistently exceed RPIQ >2.0 and R2>0.8.

[0055] Non-limiting aspects have been described, hereinabove. It will be apparent to those skilled in the art that the above methods and apparatuses may incorporate changes and modifications without departing from the general scope of the present subject matter. It is intended to include all such modifications and alterations in so far as they come within the scope of the appended claims or the equivalents thereof.

[0056] Having thus described the present teachings, it is now claimed: