A computational model for soil fertility prediction in ubiquitous agriculture

The application of sophisticated sensors to measure soil composition and plant needs are a tendence in precision agriculture. In any case, prediction models are built using machine learning algorithms. The goal is to make farming more efficient and productive with minimal impact on the environment. The present article proposes an architectural model that evaluates the soil's fertility and productivity through context history with Partial Least Squares Regression. Also productivity prediction of a wheat planted area was performed using climatic events between the years of 2001 and 2015 resulting a mean square error of calibration (RMSEC) of 0.20 T/ha, mean square errors of cross-validation of 0.54 T/ha with a Pearson coefficient (R-2) of 0.9189. For the prediction of organic matter and clay, the best results obtained were a R-2 of 0.9345, RMSECV of 0.54% and R-2 of 0.9239, RMSECV of 5.28%, respectively.