Rapid assessment of soil condition in Kenya through development of near infrared spectral indicatators
Soil degradation in Kenya has environmental and economic impacts. Large area assessments are needed to quantify and diagnose problems of soil fertility and environmental degradation and target sustainable land management interventions, such as agroforestry, and to measure impacts of interventions. Near-infrared reflectance spectroscopy (LAIRS) is a low cost, rapid and robust method for characterizing soil condition. The World Agroforestry Centre (ICRAF) has compiled an extensive spectral library of several thousand soil samples from Kenya. However, soil test data of properties determined using wet chemistry is available for only a subset of samples in the database because the methods are too expensive to measure on large numbers of samples. In addition it is tedious to develop separate individual calibrations for each soil property and region. Soil spectra integrate information on a number of soil physical and chemical components and it is against this background, that this study was carried out to derive integrated spectral indictors of soil condition for Kenya based on near infrared spectra. The specific objectives were to summarize the main variation in reflectance spectra in the soil samples into simple metrics, summarize the main variation in the soil chemical and physical properties and relate the spectral metrics to the individual soil chemical and physical properties and their combined principal components. Eight hundred and forty three soil samples (0-20 and 045cm depths) from different parts of Kenya that had complete physical and chemical data were randomly selected. The soil properties tested were organic carbon, pH, exchangeable Ca and Mg, extractable P and K, sand and clay content. Soil spectra were recorded on these samples using a Fourier Transform infrared spectrometer. The absorbance peak heights of the three principal spectral absorption features after baseline correction were proposed as the basis for the spectral condition index. The soil properties were related to the absorbance peak heights using Partial Least Squares Regression (PLS) regression. A second set of relationships was developed by relating the soil properties to the full spectrum using the Bruker Quant 2 method, which is also based on PLS but uses all the wavebands. The peak heights displayed higher correlation with soil properties after baseline correction - Continuum Removal (CR) as compared to before CR. The peak heights predicted exchangeable Ca which is a key soil fertility parameter moderately well, with calibration and cross-validated r2=0.60. Sand, exchangeable K and extractable P however, had the poorest correlations with the spectral peak data. The full spectrum (use of Quant 2 method) provided effective predictions for the individual soil properties: ExCa (r2 =0.86), ExMg (r2 =0.74) and pH (r2 =0.61). The method was also most robust in predicting the first principal component of the soil properties(r 2=0.76). Better models to predict soil condition from spectral metrics are obtained by using the full wavelength range, as opposed to restricting the models to the use of the three peak regions (using the PLS method), using basic peak height and width information related to the key absorption features. The study showed that the full spectrum method was most robust in developing spectral indicators of soil condition in Kenya based on the first principal component of the soil conditions.