Estimation of Soil Organic Carbon for Sustainable Agriculture using Deep Learning

The organic carbon percentage is concomitant indicating the mineralization of nutrients and the ability of the soil to hold nutrients cations, structural stability, and water holding capacity. It is necessary to know the quantity of carbon for healthy soil and avoid the production related problems which can affect the sustainable agriculture model. In existing approaches, to quantitively calculate soil carbon, sample collection and in-situ laboratory testing are performed. In this work, a novel framework is proposed which is based on Partial Least Square Regression and Long Short-Term Memory networks to quantify soil organic carbon from the LUCAS dataset. Samples of LUCAS dataset are used as input to this framework. The samples are pre-processed by PLS to reduce their dimensions. These pre-processed samples are then passed to the LSTM, a Deep learning framework to build an efficient prediction model. The proposed framework performed more accurately, and its effectiveness is shown by comparing it with existing regression models.