Characterisation of selected soil properties using remote sensing techniques

Abstract

Many conventional laboratory methods are used to characterise spatial and temporal variation of soil properties in order to understand soil quality for different purposes. Currently there is a high demand for accurate soil information by land users. Therefore there is a need to develop a rapid, inexpensive, non-destructive and accurate technique that could compensate or replace conventional laboratory methodologies. Remote sensing has the potential to serve as an alternative approach to characterise soil properties due to its advantages over conventional laboratory methods such as it is rapid, non-destructive and it has low cost. The objectives of this study were to: (i) evaluate the ability of proximal soil sensing to characterise soil properties namely organic matter, soil moisture content, macronutrients, soil texture, cation exchange capacity (CEC), and pH. (ii) Identify bands of relevance from proximal soil sensing (300-2400 nm) that can provide acceptable reflectance variation for different levels of selected soil properties. (iii) Evaluate the performance of models developed from multispectral space-borne image in characterising selected soil properties. In this study spectroradiometer (proximal sensor) and worldview 2 satellite images (space-borne) were the two remote sensing techniques used to collect information about soil at Syferkuil experimental farm of the University of Limpopo. Visible and near infrared spectral data of 98 soil samples were collected at the study site using Analytical spectral device (ASD) field spectroradiometer. Spectral reflectance from spectroradiometer and those extracted from worldview 2 satellite image were used to develop prediction models of selected soil properties using Partial least square regression (PLSR). Bands of relevance were also identified from PLSR models developed from spectral data acquired by spectroradiometer. The results showed that estimation accuracy of PLSR models developed using spectral data from proximal soil sensing were excellent (Category A) for clay, sand, soil organic matter (SOM), and soil moisture content, while good prediction accuracy (Category B) was observed for other soil properties such as silt, ammonium, nitrate, active acidity (pHw), calcium, magnesium, phosphorus, potassium, sulphur, CEC, and reserve acidity (pHKCl). Then, relevant bands which contributed greatly in the prediction of these soil attributes were selected from the electromagnetic spectrum, the range was from 451 nm to 2400 nm. These bands fall within visible, shortwave infrared and near-infrared x

regions of electromagnetic spectrum. In addition all selected soil properties were approximately quantitatively estimated using spectral data from satellite image. Based on the results obtained it can be concluded that proximal soil sensing has the ability to predict selected soil properties with various accuracies and it can be used as an alternative technique to characterise soil properties of South African soils. Soil predicting models developed from proximal soil sensing data also showed that there are bands of relevance within spectral range of 451 nm to 2400 nm. However more work is required for space-borne sensing before it can be used as one of the soil characterisation methods since its prediction accuracy was low as compared to that of hyperspectral proximal soil sensing. Keywords: Space-borne sensing; proximal soil sensing; soil characterisation.