Soil property estimation using visible and near-infrared reflectance

Site-specific management of soil physical and chemical characteristics for maintaining high yields within production systems while preserving soil quality has grown in use in recent years. This management includes measurement and mapping of soil variability across fields, which often requires a large number of soil samples. Additionally, traditional soil tests can be slow and/or expensive to complete, leading many farmers to skip soil tests completely. Recently, however, sensor-based approaches including reflectance spectroscopy have been proposed as quicker, easier alternatives. Therefore, the objectives of this study were 1) to investigate the capabilities of reflectance spectroscopy for the determination of both full profile and surface soil properties important in soil quality and 2) to determine if reflectance sensing at a limited number of wavelengths is a viable tool for measuring in-field variation of surface soil properties for site-specific management.;To address the first objective, soil samples were obtained from a research site which contained 32 plots with a wide variation in topsoil depth and assumed differences in soil quality. Soil samples from the 32 plots were scanned with a laboratory spectrometer in both field-moist and oven-dried conditions. Statistical calibrations were developed relating reflectance data to conventional lab analysis. Total carbon (C) and nitrogen (N) showed high estimation accuracy with an R2 of 0.97 and 0.91, respectively. Estimation accuracy of other soil variables was much more variable. Magnesium (Mg) and cation exchange capacity (CEC) exhibited good estimations (R2 = 0.83 and 0.74, respectively), while pH and saturated hydraulic conductivity (K s) did not (R2 = 0.07 and 0.37, respectively). This research showed that spectroscopic analysis of field-moist soil is a viable option for estimating several important soil quality factors. This is advantageous because sample preparation time is further reduced and because sensing field-moist samples is a step toward successful in-situ sensing.;To address the second objective, data were obtained in two central Missouri production fields with a two-band sensor capable of capturing mobile measurements. Based on variability observed in field mapping with the two-band sensor, points were selected for soil sample collection. Collected soils were scanned with both the two-band sensor and a spectrometer in the laboratory. Of the different data sources, the best prediction accuracy (R2 = 0.75) for soil organic carbon (SOC) was from in-field sensing with the two-band sensor. However, high accuracy was not maintained when calibration equations were applied in full-field mapping. In one field SOC was well-estimated with a root mean squared error (RMSE) of 6.9 g kg-1 but the other field had a RMSE of 18.4 g kg-1. This difference could be attributed to the areas selected for soil collection not being indicative of the overall variation within the entire field. Various combinations of wavelengths from the spectrometer were examined to determine optimum two-band indices for SOC estimation. There was high correlation with SOC in the near-infrared range, specifically when combining wavelengths greater than 1800 nm with wavelengths near 1400 and 1900 nm. When only considering combinations of wavelengths from <1000 nm, the highest correlation to SOC was when both were from the visible wavelength range. Additional data collection and analysis would be needed to determine if these same wavelengths were best for a wider range of soils.