Machine learning approaches for crop growth monitoring using multi-temporal and multi-variety remotely sensed data

For crop growth monitoring task, process-based model is widely used to simulating plants biophysical processes using agrometeorological data, soil properties and plant management. However there is limitation for these models mainly due to availability, uncertainty and spatial variation of model parameters. To solve these problems, empirical models built using machine learning algorithms was developed to predict sugarcane sucrose content based on multi-temporal spectral data with a combination of agrometeorological data. The performance of the built models is evaluated with comparison of statistical model and machine learning model with only spectral data. The result shows that the machine learning model has a better performance compared with statistical model. Moreover, in addition of multi-temporal imformations from both spectral and agrometeorological data there is an apparently improvement on sugarcane sucrose prediction accuracy.