A comparison of global and regional open datasets for urban greenspace mapping

Greenspace has positive influences on urban environment and human health, and thus it is desirable to acquire data for (urban) greenspace mapping. Nowadays, global and regional open land-use/land-cover datasets have become essential sources for greenspace mapping, but few studies have quantitatively compared them. To fill this gap, this study carries out a quantitative comparison of six global and regional open datasets (CGLS-LC100, CLC, GLC30, UA, FROM-GLC10 and OSM) for greenspace mapping. First of all, the most appropriate land-use/landcover classes selected as greenspace are analyzed for each open dataset; then, different open datasets are evaluated and compared in terms of five measures (accuracy, precision, recall, F1-score and green coverage rate). Five urban areas in UK are chosen as study areas. Two categories of reference datasets are used for evaluation, including an Ordnance Survey (OS) greenspace dataset in UK and a number of sampling points classified by referring to Google Earth. Results show that: the OSM dataset performs the best, while comparing with the OS dataset (characterized by a narrowly interpreted greenspace); and the FROM-GLC10 dataset performs the best, while comparing with the sampling points (characterized by a broadly interpreted greenspace). Moreover, by using these two open datasets, most quantitative results are close to or higher than 80 %, in terms of the accuracy, precision, recall and F1-score; in most cases there also is the smallest difference between using these two open datasets and corresponding reference datasets, in terms of the green coverage rate. These findings have benefits for researchers and planners to choose an appropriate open dataset for greenspace mapping.

Automated summary

The study compared six global and regional open datasets for greenspace mapping, finding that the OSM dataset performed best when compared to the OS dataset, and the FROM-GLC10 dataset performed best when compared to sampling points. Most quantitative results using these two open datasets were close to or higher than 80%, indicating their effectiveness for greenspace mapping.