By Juwita Nirmala Sari (Indonesia)
In this work, we aimed to improve the smartphone-based SOC prediction by incorporating various area information which are soil properties; sand, silt, clay, and environmental conditions; DEM (Digital elevation model ), land cover, precipitation, slope, temperature, and vegetation. The improved smartphone-based SOC prediction was compared with two existing SOC maps namely SoilGrids and S-World. We employed multiple linear regression techniques to develop our model with soil color, soil properties, and environmental conditions as covariates to predict SOC content. Our study found that the improved model was proven to be sufficiently accurate. The robustness of the smartphone-based SOC prediction model was observed to be improved when the soil properties and environmental data were added to the model indicated by its R2 and RMSE. Some variables like lightness, sand, silt, DEM, temperature, and vegetation were found to be the major factors influencing the prediction of SOC. In comparison to the existing SOC map, we observed small differences between the prediction residuals with both SoilGrids and S-World. However, our model predictions were closer to the observed laboratory values and its residuals were close to zero suggesting that it was better than the existing SOC maps. Based on our observation, we also found that there was inconsistency in predicting SOC. Some of the identified causes were missing sample picture, unbalance lightness, lower phone battery, and shifting of the reference color paper during the measurement.
Keywords: Soil Organic Carbon (SOC); smartphone-based method; soil properties; environmental data.