A Bayesian approach to mapping the uncertainties of global urban lands

Global distribution of urban lands is one of the essential pieces of information necessary for urban planning. However, large disagreement exists among different products and the uncertainty remains difficult to quantify. We applied a Bayesian approach to map the uncertainties of global urban lands. We demonstrated the approach by producing a hybrid global urban land map that synthesized five different urban land maps in ca. 2000 at 1-km resolution. The resulting hybrid map is a posterior probability map with pixel values suggesting the probability of being urban land, which is validated by 30-m higher resolution references. We also quantified the minimum and maximum urban areas in 2000 for each country/continent based on subjective probability thresholds (i.e., 0.9 and 0.1) on our hybrid urban map. Globally, we estimated that the urban land area was between 377,000 and 533,000 km2 in 2000. The credible interval of minimum/maximum urban area can help guide future studies in estimating urban areas. In addition to providing uncertainty information, the hybrid map also achieves higher accuracy than individual maps when it is converted into a binary urban/non-urban map using a probability threshold of 0.5. This new method has the ability to further integrate discrete site/location-based data, local, regional, and global urban land maps. As more data is sequentially integrated, the accuracy is expected to improve. Therefore, our hybrid map should not be regarded as a final product, but a new prior product for future synthesis and integration toward a “big data” solution.