|Title||Estimating soil properties from smartphone imagery in Ethiopia|
|Author(s)||Aitkenhead, M.J.; Poggio, L.; Wardell-Johnson, D.; Coull, M.C.; Rivington, M.; Black, H.I.J.; Yacob, G.; Boke, S.; Habte, M.|
|Source||Computers and Electronics in Agriculture 171 (2020). - ISSN 0168-1699|
|Department(s)||ISRIC - World Soil Information|
|Publication type||Refereed Article in a scientific journal|
|Keyword(s)||Carbon - Ethiopia - Field monitoring - Neural networks - Soil|
The links between soil properties and smartphone imagery were investigated for 273 samples in the Halaba area of south-west Ethiopia. The aim of this was to explore the possibility of using a smartphone-based system to estimate soil properties in the field, without the need for sampling and laboratory analysis. This presents an opportunity to develop low cost soil assessment in remote locations. Imagery and associated site characteristics were captured using an ODK (Open Data Kit) interface developed specifically for the project. Two types of model linking image information to soil properties were explored, backpropagation neural networks (NN) and partial least squares (PLS). Models were generated with colour alone, spatial covariates alone and a combination of colour and spatial covariates. Two sets of data, for soil chemistry and soil physical properties, were modelled. For both NN and PLS models, estimation accuracy for chemical properties was consistently higher using colour and spatial covariate information together rather than colour or spatial covariates alone. For physical properties a similar pattern was seen but this was less clear, and estimation of physical properties was less successful based on statistical model validation.