Experience: Quality Assessment and Improvement on a Forest Fire Dataset

AbstractSpatio-temporal data can be used to study and simulate the movement and behavior of objects and natural phenomena. However, the use of real-world data raises several challenges related to its acquisition, representation, and quality. This article presents a data cleaning process, based on consistency rules and checks, that uses geometric operations to detect and remove outliers or inaccurate data in a spatio-temporal series. The proposal consists of selecting key frames and applying the process iteratively until the data have the desired quality. The case study consists of extracting and cleaning spatio-temporal data from a video tracking the propagation of a controlled fire captured using drones. The source data was generated using segmentation techniques to obtain the regions representing the burned area across time. The main issues concern noisy data (e.g., the height of flames is highly variable) and occlusion due to smoke. The results show that the quality assessment and improvement method proposed in this work can identify and remove inconsistencies from a dataset of more than 22,500 polygons in just a few iterations. The quality of the corrected dataset is verified using metrics and graph analysis.