Automatic Extraction and Filtering of OpenStreetMap Data to Generate Training Datasets for Land Use Land Cover Classification

Cidália C. Fonte,Diogo Duarte,Ismael Jesus,Joaquim Patriarca

doi:10.3390/rs12203428

Cidália C. Fonte, Diogo Duarte + Show 2 more

Open Access

https://doi.org/10.3390/rs12203428

Copy DOI

Abstract

This paper tests an automated methodology for generating training data from OpenStreetMap (OSM) to classify Sentinel-2 imagery into Land Use/Land Cover (LULC) classes. Different sets of training data were generated and used as inputs for the image classification. Firstly, OSM data was converted into LULC maps using the OSM2LULC_4T software package. The Random Forest classifier was then trained to classify a time-series of Sentinel-2 imagery into 8 LULC classes with samples extracted from: (1) The LULC maps produced by OSM2LULC_4T (TD0); (2) the TD1 dataset, obtained after removing mixed pixels from TD0; (3) the TD2 dataset, obtained by filtering TD1 using radiometric indices. The classification results were generalized using a majority filter and hybrid maps were created by merging the classification results with the OSM2LULC outputs. The accuracy of all generated maps was assessed using the 2018 official “Carta de Ocupação do Solo” (COS). The methodology was applied to two study areas with different characteristics. The results show that in some cases the filtering procedures improve the training data and the classification results. This automated methodology allowed the production of maps with overall accuracy between 55% and 78% greater than that of COS, even though the used nomenclature includes classes that can be easily confused by the classifiers.

Highlights

Knowledge regarding the Earth’s surface and its’ use for human activities is critical for several applications, such as climate change monitoring and forecast [1,2], habitat conservation and planning [3,4], population mapping [5,6], urban planning [7], policy making [8], among others [9,10]
The results show that, for both study areas, the overall accuracy of the training datasets increases from Training Data 0 (TD0) to Training Data 1 (TD1) and from TD1 to Training Data 2 (TD2), showing that the filtering process is removing regions incorrectly included in the original data
This paper presented an automated methodology to obtain land use land cover (LULC) maps with the classification of Sentinel-2 multispectral images using training sets extracted from OSM

Summary

Introduction

Knowledge regarding the Earth’s surface and its’ use for human activities is critical for several applications, such as climate change monitoring and forecast [1,2], habitat conservation and planning [3,4], population mapping [5,6], urban planning [7], policy making [8], among others [9,10]. Training data are central within the LULC map generation process, as their quality and representativeness of the classes will determine the quality of the classification result Such training data is usually generated by human photointerpretation of higher resolution imagery, such as very high spatial resolution satellite or aerial imagery, and/or from field surveys, which are costly and time-consuming processes [18,19]. This is a major limitation when seeking to automatically generate LULC maps and fully explore the potential of satellite imagery with temporal resolutions of just a few days, such as by using the imagery collected by both satellites of the Sentinel-2 constellation. It is desirable to develop methodologies that enable the automatic generation of training data, either by using already available sets of ancillary information, adopting a data driven approach or utilizing a mix of the two

Objectives

Methods

Results

Conclusion