Marco Dinis , Pedro Tarroso

Remote sensing, particularly from satellite observation, has become the standard tool for monitoring the planet with a steady increase of data production, and has seen wide application in ecosystem services analysis and management. Many remote sensing applications involve image classification, using methods from simple regressions to complex machine learning approaches that require advanced user expertise. Many approaches rely on a single model, which fails to account for the uncertainty inherent in the stochastic nature of single-model methods. This work introduces SpartANN – an open-source tool for image classification combining Artificial Neural Networks and ensemble modelling approaches, allowing users the flexibility to customize model parameters, identify areas of model congruence and quantify prediction uncertainty. We demonstrate the flexibility of SpartANN by performing a cloud-cover classification exercise on Sentinel-2 images and discuss the success of classification outputs and advantages of SpartANN.