An approach for simplifying the process of Neural Network model creation in the domain of land usage classification is presented in this work. The proposed method is a complete pipeline for cleaning the data using minimal supervision and subsequently creating a crop specific pixel level temporal model. A first step in the domain of Big Data is controlling the integrity of the data for the purpose of model creation. Traditional methods are typically based on data specific heuristics to decide what information can be used for training the model. These approaches require significant effort in hand-crafted filtering methods, and often lack sufficient structure for formalisation and generalisation.