Evaluation of Dry-Soil Infrared Techniques for Soil Organic Carbon Characterization
Open Access
- Author:
- Randhawa, Rupinder K
- Graduate Program:
- Agronomy
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- November 21, 2008
- Committee Members:
- Andrew Joseph Archibald, Thesis Advisor/Co-Advisor
Douglas D Archibald, Thesis Advisor/Co-Advisor
Heather D Karsten, Thesis Advisor/Co-Advisor
David Russell Hunter, Thesis Advisor/Co-Advisor - Keywords:
- Spectroscopy
Chemometrics
ATR
near infrared
mid-infrared
Soil organic carbon - Abstract:
- Soil organic carbon (SOC) is one of the most important soil quality parameter. It is a proxy for soil organic matter, which influences the physical, chemical and biological properties of the soil. Soil carbon also plays an important role in global C cycle. Understanding the dynamics of SOC is therefore, very important for monitoring soil quality. High density sampling is usually required to capture the spatial variability in SOC of an area which makes costly standard methods highly impractical. The standard procedures for SOC measurement are time consuming and expensive and thus discourage the detailed study of crop-soil management systems. Rapid and relatively in-expensive methods of SOC quantification are needed to improve our understanding of its dynamics and spatial and temporal variability in soils. Infrared spectroscopy in combination with multivariate analysis is a technique for rapid and in-expensive quantification of SOC. This technique is non-destructive and preserves the basic integrity of the soil system. It does not require expensive and time consuming sample pre-processing or the use of expensive and environmentally hazardous chemical extractants. In addition, the method has the potential to be used for determining carbon quality and simultaneously measuring several other soil properties. This ability of spectroscopic technique can be particularly useful in site- specific agriculture (precision agriculture) which requires extensive information on soil properties and conventional laboratory systems are too costly and labor intensive to generate the necessary data. Despite an increased interest in infrared spectroscopy especially MIR for measuring soil properties, its utility is not completely understood. There are some inconsistencies in the data and differences in opinions of various researchers about the performance of this technique and its applicability in management systems. The objective of this study is to evaluate the ability of different IR techniques to characterize SOC, identify the factors that limit the performance and develop ways to overcome those limitations. Specifically we compared the ability of two infrared regions (near-infrared (NIR) and mid-infrared (MIR)) and also two MIR optical sampling options (diffuse reflectance (DR) and attenuated total reflectance (ATR)). We used samples from three management studies which were very different in terms of diversity to develop SOC calibration models, using partial least squares regression (PLSR) analysis. In the first study we compared the ability of MIR and NIR spectroscopy to determine SOC. The accuracy of two methods as judged by root mean squares error of prediction (RMSEP) was similar for the two less diverse study sets (RMSEP = 0.12 - 0.14). For the more diverse study set error was smaller and correlation higher with the MIR region (RMSEP = 0.27) compared to NIR region (RMSEP = 0.31). SOC calibration models did not perform as well for diverse study compared to the other two which indicates that at the errors associated with calibration models depend on the sample set structure. There is a potential to improve the model accuracy in the MIR region by selecting the most relevant spectral regions. Although we could get slight improvement in model accuracy for all sets by eliminating part of the spectrum that was not associated with SOC, we could not get good performing models based just on the organic bands alone. For very diverse data sets, calibration performance can also be improved by using more than one calibration model. This was proved using samples from one study set where better calibration was achieved by using separate models for samples from upper and lower depths. Samples can be divided for separate calibrations either on the basis of known sample information (for example depth or site of sampling) or by performing un-supervised clustering. In general when dealing with a single set of soils, MIR proved to be more efficient than NIR when expressed as the percentage of study samples that are not needed in calibration, and the efficiency of both ranges were better with less diverse data sets. For adapting an existing calibration model to a new site, some samples from a new site have to be included in the updated calibration. We found that if the sample set for which the model was originally developed is diverse or is similar in mineralogy to the new study, lesser number of samples from the new set need to be included in calibration. However, better predictions can be obtained by building calibration model on some samples from that site instead of trying to extend an existing model especially in cases where the new site is more diverse or else is very different from the old site. The second study we evaluated a new optical sampling approach called ATR for SOC quantity and molecular functional group quality determination and compared it with DR. Most of the MIR spectroscopic studies reported in the literature used diffuse reflectance (DR) sampling technique. Although the technique is useful in mineral soils, the high absorbance could be a problem in case of soils very rich in organic matter. ATR-FTIR technique also has a higher potential to be used in situ. In contrast to most DR instrumentation, the ATR technique involves a sealed optics and therefore is less influenced by the temperature, dust and humidity of the measuring environement. Using two different types of soil collections we could show that ATR-FTIR calibrations perform nearly as well as DR-FTIR calibrations. The usable MIR spectral range for SOC determination is different for DR and ATR-FTIR. There are still unresolved issues about the best spectral range for quantifications and qualitative analysis of SOC. However two ranges were found to perform well for SOC quantitative analysis (3900-1240, 1750-540 cm-1). The better quality MIR spectral signature in the fingerprint region suggests that ATR may perform better than DR for SOC qualitative analyses. Although our diamond-ATR studies utilized a lab based FTIR, the same instrumental components are found in portable instruments. Since ATR optics is easily sealed we can expect similar so on-site performance as we achieved in the lab.