Mapping water salinity using Landsat-8 OLI satellite images (Case study: Karun basin located in Iran)

摘要

Water salinity is a key physical parameter that affects water quality, growth, and development of the aquatic vegetation and animals. The salinity of Karun River has been increasing due to some critical factors, e.g., severe climate condition and regional physiography, industrial sources, domestic and urban sewerage, irrigation of agricultural land, fish hatchery, hospital sewage, and high tide level of Persian Gulf. This study aimed at building regression models to ascertain the water salinity through the relationship between the reflectance of the Landsat-8 OLI and In situ measurements. Accordingly, 102 In situ measurements have been collected from June 2013 to July 2018 along the Karun River, subsequently measured data was divided into 70∶30 for training and test purposes. Besides, the Sobol' sensitivity analysis was applied to determine the best bands combination from the performance standpoint. The results of the Sobol' sensitivity analysis revealed that band 1- Coastal/Aerosol (0.433-0.453 μm), band 2-Blue (0.450-0.515 μm) and band 3 - Green (0.525-0. 600 μm) are the best combinations and showed that Landsat-8 OLI band 2 has the closest correlation with the salinity. Furthermore, to have a comprehensive investigation, the Ordinary Least Square (OLS), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP) methods were applied. The number of layers and nodes in the MLP neural network is optimized using the Genetic Algorithm (GA), and GA has selected four layers and thirty neurons per layer. The OLS, SVR and MLP + GA models resulted in values of R~2 and RMSE for test data, which are respectively obtained to be 0.68 and 411 μs cm~(-1) , 0.72 and 376 μs cm~(-1) and 0.78 and 363 μs cm~(-1). Therefore, MLP + GA has had the best performance and accuracy. However, the corresponding values are acceptable considering the fact that the range of field data extensively changes from 385 and 4310. Eventually, water salinity maps were prepared by OLS, SVR and MLP + GA methods to demonstrate the water salinity on 1 February 2015 and 5 September 2018, afterward change detection maps were prepared to assess the water salinity on 1 February 2015 and 5 September 2018. The change detection maps illustrate that the pertinent salinity on 5 September 2018 is lower as compared to the data obtained on 1 February 2015, because not only the rainfall has increased, but the cane sugar cultivation has also decreased which is one of the effective factors on the salinity of the water.