Integration of mid-infrared spectroscopy and geostatistics in the assessment of soil spatial variability at landscape level

Knowledge of soil spatial variability is important in natural resource management, interpolation and soil sampling design, but requires a considerable amount of geo-referenced data. In this study, mid-infrared spectroscopy in combination with spatial analyses tools is being proposed to facilitate landscape evaluation and monitoring. Mid-infrared spectroscopy (MIRS) and geostatistics were integrated for evaluating soil spatial structures of three land settlement schemes in Zimbabwe (i.e. communal area, old resettlement and new resettlement; on loamy-sand, sandy-loam and clay soils, respectively). A nested non-aligned design with hierarchical grids of 750, 150 and 30 m resulted in 432 sampling points across all three villages (730–1360 ha). At each point, a composite topsoil sample was taken and analyzed by MIRS. Conventional laboratory analyses on 25–38% of the samples were used for the prediction of concentration values on the remaining samples through the application of MIRS–partial least squares regression models. These models were successful (R2 ? 0.89) for sand, clay, pH, total C and N, exchangeable Ca, Mg and effective CEC; but not for silt, available P and exchangeable K and Al (R2 ? 0.82). Minimum sample sizes required to accurately estimate the mean of each soil property in each village were calculated. With regard to locations, fewer samples were needed in the new resettlement area than in the other two areas (e.g. 66 versus 133–473 samples for estimating soil C at 10% error, respectively); regarding parameters, less samples were needed for estimating pH and sand (i.e. 3–52 versus 27–504 samples for the remaining properties, at same error margin). Spatial analyses of soil properties in each village were assessed by constructing standardized isotropic semivariograms, which were usually well described by spherical models. Spatial autocorrelation of most variables was displayed over ranges of 250–695 m. Nugget-to-sill ratios showed that, in general, spatial dependence of soil properties was: new resettlement > old resettlement > communal area; which was potentially attributed to both intrinsic (e.g. texture) and extrinsic (e.g. management) factors. As a new approach, geostatistical analysis was performed using MIRS data directly, after principal component analyses, where the first three components explained 70% of the overall variability. Semivariograms based on these components showed that spatial dependence per village was similar to overall dependence identified from individual soil properties in each area. In fact, the first component (explaining 49% of variation) related well with all soil properties of reference samples (absolute correlation values of 0.55–0.96). This showed that MIRS data could be directly linked to geostatistics for a broad and quick evaluation of soil spatial variability. It is concluded that integrating MIRS with geostatistical analyses is a cost-effective promising approach, i.e. for soil fertility and carbon sequestration assessments, mapping and monitoring at landscape level.