VNIR Soil Spectroscopy for Field Soil Analysis

Abstract

The advent of affordable, ground-based, global positioning information (GPS)-enabled sensor technologies provides a new method to rapidly acquire georeferenced soil datasets in situ for high-resolution soil attribute mapping. Our research deployed vehicle-mounted electromagnetic sensor survey equipment to map and quantify soil variability (∼50 ha per day) using apparent electrical conductivity as an indirect measure of soil texture and moisture differences. A portable visible-near infrared (VNIR) spectrometer (350-2500 nm) was then used in the field to acquire hyperspectral data from the side of soil cores to a specified depth at optimized sampling locations. The sampling locations were derived by statistical analysis of the electromagnetic survey dataset, to proportionally sample the full range of spatial variability. The VNIR spectra were used to predict soil organic carbon (prediction model using field-moist spectra: R 2 = 0.39; RPD = 1.28; and air-dry spectra: R 2 = 0.80; RPD = 2.25). These point values were combined with the electromagnetic survey data to produce a soil organic carbon map, using a random forest data mining approach (validation model: R 2 = 0.52; RMSE = 3.21 Mg C/ha to 30 cm soil depth; prediction model: R 2 = 0.92; RMSE = 1.53 Mg C/ha to 30 cm soil depth). This spatial modeling method, using high-resolution sensor data, enables prediction of soil carbon stocks, and their spatial variability, at a resolution previously impractical using a solely laboratory-based approach.

Soil spectroscopy with the use of chemometrics, machine learning and pre-processing techniques in soil diagnosis: Recent advances–A review

“…The results obtained confirmed that the proposed approach provided reliable estimates with a large coefficient of determination R² and small predictions errors. In [52] the performances of infield estimation of SOC using a portable Vis-NIR spectrometer and moist soils with a laboratory NIR instrument on dried soil samples were compared. The model on airdry spectra outperformed the one obtained from fresh samples in terms of correlation between predicted and measured SOC values.…”

Comparison of Vis–NIR on in situ, intact core and dried, sieved soil to estimate clay content at field to regional scales

“…For prediction of SOC with the same set of cores, calibrations based on dried intact core measurements had better prediction statistics than those on derived processed samples, but field-moist, intact-core calibrations performed poorly (Morgan et al, 2009). For various locations in New Zealand, Kusumo et al (2008) and Roudier et al (2015) reported satisfactory predictions of total carbon (RMSE = 1.21% and RPD = 2.01 for Kusumo et al (2008); RMSE = 0.38% and RPD = 2.62 for Roudier et al (2015)), whereas Hedley et al (2015) estimated SOC from intact cores at field conditions with an RMSE of prediction of 0.66% and RPD of 1.28. Viscarra Rossel et al (2009) examined several soil pit and road cut profiles with an analytic spectral device (ASD, ASD Inc., Boulder, CO, USA) contact probe and reported slightly more accurate predictions for clay content in situ (RMSE = 7.9%) than for laboratory Vis-NIR (RMSE = 8.3%).…”

“…()), whereas Hedley et al . () estimated SOC from intact cores at field conditions with an RMSE of prediction of 0.66% and RPD of 1.28. Viscarra Rossel et al .…”

Integration of NIR on a multi-sensor platform to improve soil resource assessments

“…In the last few decades, technological progress has provided new and improved sensors able to tackle soil complexity and variability, and Vis–NIR spectroscopy has played a pioneering role. 3 I have been fascinated during my studies by the ease, speed, low cost and minimal pre-processing requirements of soil spectroscopy to enable analysis of extensive sample datasets. The array of information embedded in a single spectrum and the ease and rapidity of analysis are extremely valuable new ways to study soil complexity.…”