NIR and DRIFT-MIR spectrometry of soils for predicting soil and crop parameters in a flooded field

The increasing popularity of site-specific management (SSM) calls for fast, inexpensive, simultaneous analyses of large numbers of soil variables. The objective of this study was to assess the potential of near infrared (NIR) and diffuse reflectance Fourier transformed in the mid-infrared range (DRIFT-MIR) spectrometry for predicting crop and soil parameters in a flooded California rice field. Two transects of 400 m each were left unfertilized, and 100 sample locations were established. Soil samples were taken in spring, and crop and weed samples at harvest. IR spectra were linked to total soil C and N, mineralizable N, P Olsen, effective cation exchange capacity (eCEC) and exchangeable cations (Ca, Mg, Na and K), as well as yield, N uptake, biomass and weed biomass using partial least squares regression (PLSr). The PLSr models were calibrated using 50 random observations, and validated using the remaining 50 observations. For soil, predictions for eCEC, Ca and Mg were the most accurate, with r2 values of 0.83, 0.80 and 0.90 for NIR and 0.56, 0.60 and 0.61 for DRIFT-MIR. Correlations for P Olsen were 0.71 and 0.55, and for mineralizable N 0.46 and 0.21, respectively. No significant correlations were found for total soil C or N. For crop parameters, only weed pressure (r2 of 0.55 and 0.44) and straw biomass (0.30 and 0.34) yielded significant correlations. The correlation with weed pressure was an indirect effect due to better competition by weeds compared to rice under low soil fertility levels. For most parameters, standard errors of prediction were lower than reported in the literature. This indicates that the small range of variability within a field might be the limiting factor in predicting these parameters. It also illustrates the limited use of correlation coefficients in PLSr model validations. We concluded that NIR spectrometry shows promise for SSM, although its predictive power for parameters may vary from site to site. Moreover, predictive models remain unique for specific agroecosystems, and therefore have to be calibrated for every area. The fast and accurate predictions for Ca and Mg concentrations in the soil could be especially important in diagnosing and combating grass tetany, which strongly depends upon Ca and Mg concentrations in the soil.