Evaluating the performance of multi-temporal synthetic-aperture radar imagery in land-cover mapping using a forward stepwise selection approach

Radar images are a supplement or alternative to optical images, especially in tropical regions where cloud cover is a challenge. The present study evaluated the performance of using multi-temporal synthetic-aperture radar images in land-cover mapping in Binh Duong province, Vietnam. Two experimental...

Teljes leírás

Elmentve itt :
Bibliográfiai részletek
Szerzők: Mucsi László
Bui Dang Hung
Dokumentumtípus: Cikk
Megjelent: 2023
Sorozat:REMOTE SENSING APPLICATIONS : SOCIETY AND ENVIRONMENT 30
Tárgyszavak:
doi:10.1016/j.rsase.2023.100975

mtmt:33741208
Online Access:http://publicatio.bibl.u-szeged.hu/27462
Leíró adatok
Tartalmi kivonat:Radar images are a supplement or alternative to optical images, especially in tropical regions where cloud cover is a challenge. The present study evaluated the performance of using multi-temporal synthetic-aperture radar images in land-cover mapping in Binh Duong province, Vietnam. Two experimental cases were investigated: Case 1 used only multi-temporal radar images, whereas Case 2 used a combination of multi-temporal radar images and one optical image. A set of 24 Sentinel-1 images and one Landsat-8 image acquired in 2020 were processed. A forward stepwise selection approach based on a random forest algorithm and a six-class classification scheme were used to determine the best combination of images. In Case 1, the 16-date combination gained the best result with an overall accuracy (OA) of 76.6%. Considering the trade-off between efficiency and cost, the seven-date combination (OA = 76.1%) could be the optimal integration. Compared to using single-date radar images, the OA in Case 1 was improved by 9.5%, and the producer's and user's accuracies were improved by 4.68%–33.33%. Meanwhile, in Case 2, the combination of one optical and seven radar images gave the best result (OA = 83.7%). It had at least 7.1% higher OA than Case 1. However, its OA was at least 2.4% lower, and the producer's and user's accuracies of most classes were reduced when compared to using only a single Landsat-8 image. Overall, the findings of this study confirmed the effectiveness of using multi-temporal radar images in land-cover mapping. However, the effectiveness of the combination of radar and optical images needs further elucidation.
Terjedelem/Fizikai jellemzők:11
ISSN:2352-9385