Open AccessImproved water quality mapping based on cross‐fusion of Sentinel‐2 and Landsat‐8 imageries
Author(s) -
Rangzan Kazem,
Kabolizadeh Mostafa,
Karimi Danya
Publication year - 2020
Publication title -
iet image processing
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.401
H-Index - 45
eISSN - 1751-9667
pISSN - 1751-9659
DOI - 10.1049/iet-ipr.2019.1503
Subject(s) - mean squared error , wavelet transform , image fusion , fusion , correlation coefficient , hue , root mean square , computer science , biochemical oxygen demand , artificial intelligence , remote sensing , water quality , coefficient of determination , mathematics , wavelet , pattern recognition (psychology) , chemical oxygen demand , statistics , environmental science , image (mathematics) , geology , environmental engineering , ecology , linguistics , philosophy , wastewater , electrical engineering , biology , engineering
This study proposed methods based on Sentinel‐2 and Landsat‐8 cross‐fusion for improving water quality mapping (WQM). Therefore, four traditional fusion methods including intensity–hue–saturation, Gram–Schmidt transform, wavelet transform and Brovey transform and different scenarios of cross‐fusion have been implemented. The proposed cross‐fusion methods highly improved the correlation coefficient (CR) between the images and the water quality parameter (WQP). Considering the higher CR values, the created WQP maps showed very good accuracy, in which the root‐mean‐square error values were 0.03, 0.59, 0.96, 0.26 and 279.76 for potential hydrogen (PH), dissolved oxygen (DO), chemical oxygen demand (COD), biological oxygen demand (BOD) and electrical conductivity (EC) maps, respectively. Also, the effect of considering 1 px value or the mean of a 3×3 window of the input images for calculating the regression models on the accuracy of the final maps was tested. Only the best outputs for mapping PH and DO parameters were based on applying the mean of a 3×3 window. The results also showed that increasing the window size could increase the computational complexity and decrease the WQM accuracy. Comparing the output maps with the traditional maps confirmed the higher accuracy of the proposed methods.