The influence of urban stractures, vegetation cover, and utilized data in urban impervious surface mapping from multi-source data

ABSTRACT
Given the significant impact of expanding impervious surfaces on the urban environment, obtaining accurate and up-to-date information about impervious surfaces (IS) is important in urban planning and sustainable management. Remote sensing data, especially aerial photos, have a high potential to provide the mentioned information and have been successfully used in recent years. Despite the widespread use of these data in urban impervious surfaces (UIS) mapping, the reliability and accuracy of the final map still need further investigation. Therefore, in this study, using a detailed and precise classification scheme, drone, and Sentinel satellite data, the impact of three parameters including utilized data, urban structure, and vegetation canopy on the output maps accuracy was evaluated. The results showed that all three factors are of great importance and may cause significant uncertainty in the output maps. Vegetation cover can lead to up to 10% underestimate in the IS. Additionally, changes in urban structure in different areas and changes in the utilized data can also result in a 20% change in the overall accuracy. Results from this work can be used to provide a proper understanding of the reliability of remote sensing products and depict directions for future methodological development
Extended Abstract
Introduction
Urban development and the resulting changes in land cover can have a significantly negative impact on natural habitats, microclimates, and urban hydrological systems. This is primarily due to the replacement of natural surfaces with impervious ones, such as streets, buildings, and parking lots which prevent water infiltration. In line with the sustainable development of cities, it is of great importance to have accurate and detailed information of these surfaces to monitor their growth and study their impacts. Remote sensing has proved useful to provide spatially explicit information about urban areas. Remote sensing data includes various types, such as images with high spectral, high spatial or high temporal resolution. Each has its own previledge and shortcomings. These data has complementary nature; as an example, backscattering values in Sentinel-1 polarimetric bands govern by the surface roughness and moisture content or reflectance in RGB bands deponds on the color. Although many studies have focused on extracting this information from remote sensing data, key issues like the accuracy of final maps, the factors affecting it and the spatial distribution of errors have not been thoroughly addressed. Therefore, this study aims to evaluate and quantify the impact of some various influential factors including vegetation, urban structure, and the data utilized.
 
Methodology
In order to allow a quantitative comparison, this research was conducted in two distinct blocks under varying condition conditions in the city of Ahvaz located in the southwest of Iran. For this purpose, initially, the UAV data for two blocks was classified using the classification scheme with 18 subclasses and the random forest algorithm. Subsequently, the UAV RGB image was fused with the Sentinel 1 data, and features such as NDVI, NDBI, and texture were extracted using a feature-based algorithm. The random forest classification algorithm was done.
The results from various classifications were assessed and compared using 1911 random training and validation samples.  Additionally, the tree canopy map derived from the classification was intersected with the map of streets, residential blocks and park impervious surfaces, which was prepared using the municipality database, and the results were evaluated.
 
Results and discussion
The results showed:

When using only drone data, the random forest algorithm achieved an overall accuracy and Kappa coefficient of 0.826 and 0.816, respectively, in block A. However, in block B, these accuracy parameters were significantly lower, at 0.612 and 0.587. Given that the data, algorithm and pattern used in both blocks are identical, the decrease in accuracy could be attributed to the changes in the urban structure relative to the flight lines or differences in the height distribution between the two areas.
The fusion of Sentinel and UAV data fusion resulted in a significant improvement in classification accuracy, raising the overall accuracy from 0.612 to 0.8. An assessment of the subclass’s accuracy revealed that the paved sidewalk subclass experienced the highest accuracy improvement, with a 54% increase.
Due to the vegetation canopy obscuring impervious surfaces, 9% of the Block A and approximately 3% of the Block B streets have been incorrectly classified as vegetation.
Expressing the error caused by vegetation as a fraction of the canopy area rather than the impervious surfaces makes the stated error percentage more accurately.

 
Conclusion
The main conclusions of this study can be summarized as follows:

By utilizing the multi-source data (RGB UAV, spectral indices from Sentinel-2, polarimetric bands of Sentinel-1 and textures), impervious surfaces can be extracted with a high level of detail (17 classes) and acceptable accuracy. Because, these data has complementary nature and the joint use of them will increase the classification accuracy.
Irregular urban structures can increase classification errors due to the significant changes in features direction relative to the view angle, especially in the case of short-range photogrammetry. Therefore, the effect of urban structure on impervious surface classification accuracy depends on the utilized dataset.
Urban impervious surfaces are frequently underestimated because they are covered by tree canopies. The errors caused by these obscured surfaces vary across different parts of the city and do not follow a specific pattern.

-The results of this work can enhance our understanding of the reliability of remote sensing products and provide guidance for future methodological developments.
 
Funding
There is no funding support.
 
Authors’ Contribution
Authors contributed equally to the conceptualization and writing of the article. All of the authors approved the content of the manuscript and agreed on all aspects of the work declaration of competing interest none.
 
Conflict of Interest
Authors declared no conflict of interest.
 
Acknowledgments
 We are grateful to all the scientific consultants of this paper.