Diagnostic Screening and Prevalence Assessment of Soil Stability Related Problems Using Infrared Spectroscopy in Lake Victoria Basin, Kenya

Abstract

Application of infrared spectroscopy (IR) in soil studies is well established. However, there has been little focus on examining IR for soil stability pedotransfer purposes. This study aimed to evaluate the use of IR in diagnosing soil stability related problems and assessing their prevalence with a case study of Lake Victoria Basin (LVB) in Kenya. Specifically to develop alternative IR-based models for screening soil stability related properties and compare these with predictions using conventional soil properties, validate the IR-based models using independent datasets, assess prevalence of stability related problems in two sentinel sites, and assess indices of soil stability functional attributes most appropriate for screening stability problems using IR-based models. Two samples sets representing different soils were used for the study. A model calibration set (n = 136) was obtained following a conditioned Latin hypercube sampling, and a validation set (n = 120) using spatially stratified random sampling. Spectral measurements were obtained for air-dried (< 2 mm) and for ground (< 0.5 mm) soil sub-samples using multipurpose analyzer and Tensor 27 spectrometers for near infrared (NIR) and mid-infrared (MIR) ranges, respectively. Soil laboratory data were also obtained for wet aggregation indices (WSA): macro, micro and unstable fractions from two different wet-sieving pretreatments. Soil properties were screened for prediction of the WSA using Classification and Regression Tree analysis. The WSA were calibrated to the soil-based predictors and to smoothed first derivative spectra and to spectra wavelet transform variables using partial least squares regression (PLS). WSA threshold values developed using soil predictors and spectra were used to diagnose soil stability related problems and assess their prevalence in Lower Nyando (LNY) and Homa Bay (HB) sentinel sites. Key soil predictors were: soil organic carbon and pH water (macro), water dispersible clay (micro) and exchangeable sodium (unstable). Coefficient of determination (R2) for full cross validation PLS and IR-methods was: 0.6 (macro and unstable); 0.4 (micro) fractions. The R2 for soil-based PLS was: 0.3 (micro); 0.5 (unstable). Independent testing of IR-methods gave R2 and RPD (ratio of prediction deviation) (R2, RPD) as follows: macro (0.81, 1.4); unstable (0.65, 1.1). The LNY and HB sites indicated 66 % low stability prevalence, and 80 % of the sites were at risk. The study showed that IR-based predictors are superior over soil properties for stability transfer purposes. That 70 to 80 % of the soils in the sites had low stability problems and the risk of stability related problems increased with soil depth. Soil wet stable macro aggregate at10 and 50 %, stable micro at 20 and 40 %, and unstable fraction at 70-65 and 20-40%, define low and high stability, respectively. The models developed can be used to diagnose and assess prevalence of soil stability related problems in the LVB in Kenya and other regions. Further efforts should, however, widely test similar soil property predictor sets, aggregation indices and IR to: (i) validate established soil-based predictors, (ii) counter variability from sample provenance for improved model geographic transferability, and (iii) assess suggested performance improvement with calibration spiking.