Title: Sampling and Calibration Requirements for Optical Reflectance Soil Property Sensors for Korean Paddy Soils Authors
|Lee, K - SUNGKYUNKWAN UNI S. KOREA|
|Lee, D - SUNGKYUNKWAN UNI S. KOREA|
|Jung, I - NATL INST AGRIC ENG S KOR|
|Chung, S - CHUNGNAM NATL UNI S. KORE|
Submitted to: Journal of Biosystems Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: August 25, 2008
Publication Date: September 5, 2008
Citation: Lee, K.S., Lee, D.H., Jung, I.K., Chung, S.O., Sudduth, K.A. 2008. Sampling and calibration requirements for optical reflectance soil property sensors for Korean paddy soils. Journal of Biosystems Engineering. 33(4):260-268. Interpretive Summary: Measuring the variation in soil properties within fields is an important component of precision agriculture. For many soil properties, it is difficult to obtain enough data to accurately characterize their spatial variation, due to the cost of traditional sampling and laboratory analysis. Sensors that can estimate soil properties without the need for sampling are a promising alternative. One technology that has received considerable attention in this regard is optical reflectance sensing in the visible and near infrared (NIR) wavelength bands. An important question to answer in applying reflectance sensing for soil analysis is how many, and what type of, calibration soil samples are required. To help answer that question, we collected multiple soil samples from 39 fields representing 14 distinct soil series that account for 74% of the total paddy field area in South Korea. We measured the reflectance characteristics of these soils in the laboratory and used statistical techniques to relate the reflectance to laboratory-measured soil properties. We found that it was important to include calibration soil samples that had characteristics (such as soil type) closely related to the soils under analysis. This could be done either with a very broad, built-in calibration or by creating modified calibrations specific to the soils being studied. These results provide information that instrumentation engineers and researchers can use to help develop and apply new in-field soil sensing technology.
Technical Abstract: Optical diffuse reflectance sensing has potential for rapid and reliable on-site estimation of soil properties. For good results, proper calibration to measured soil properties is required. One issue is whether it is necessary to develop calibrations using samples from the specific area or areas (e.g., field, soil series) in which the sensor will be applied, or whether a general “factory” calibration is sufficient. A further question is if specific calibration is required, how many sample points are needed. In this study, these issues were addressed using data from 42 paddy fields representing 14 distinct soil series accounting for 74% of the total Korean paddy field area. Partial least squares (PLS) regression was used to develop calibrations between soil properties and reflectance spectra. Model evaluation was based on coefficient of determination (R2) and RPD, the ratio of standard deviation to RMSEP. When sample data from a soil series were included in the calibration stage (full information calibration), RPD values of prediction models were increased by 0.03 to 3.32, compared with results from calibration models not including data from the test soil series (calibration without site-specific information). Higher R2 values were also obtained in most cases. Including some samples from the test soil series (hybrid calibration) generally increased RPD rapidly up to a certain number of sample points. A large portion of the potential improvement could be obtained by adding about 8 to 22 points, depending on the soil properties to be estimated, where the numbers were 10 to 18 for pH, 18-22 for EC, and 8 to 22 for total C. These results provide guidance on sampling and calibration requirements for NIR soil property estimation. Specifically, they suggest that obtaining calibration samples from within a specific soil series is important for obtaining good soil property estimates. Depending on the particular soil series and property of interest, about 20 or fewer samples are sufficient to obtain most of the potential gain in accuracy.