Use of Multiple Sensor Satellite Images to Predict the Concentration of Chlorophyll-a within Wetlands using Artificial Neural Networks
Abstract
The use of remote detection and monitoring using satellite images is increasing to monitor algal blooms. The advantage of remote detection and monitoring is spatial continuity of a wide range of water systems, but there are limitations in acquiring images of the region of interest due to the weather conditions, satellite revisit cycle and coverage. To address these limitations, multiple sensor satellite images of different resolutions has been proposed. This study used multiple sensor satellite images to predict the concentration of the chlorophyll-a within 15 wetlands of Nakong river watershed, South Korea, using Artificial Neural Networks (ANN). The concentration of chlorophyll-a taken at 25 monitoring stations from 2018 to 2021 was set as the target variable of ANN. The multiple sensor satellites used in the study were Sentinel-2 and Landsat-8 images acquired during the corresponding period. After visual processing, 121 Sentinel-2 and 33 Landsat-8 images were used. Atmospheric correction of satellite images was performed using ACOLITE. We identified the image pixel spatially matching with observation stations, and the nine pixels including neighboring pixels were averaged to make input variables for ANN. To demonstrate the necessity of multiple sensor satellite images, we developed three ANNs: ANN#1 (input variables from multiple sensor satellites), ANN#2 (input variables from only Sentinel-2), and ANN#3 (input variables from Landsat-8). ANN#1, #2, and #3 had 154, 121, and 33 samples, respectively. The prediction capacity of three ANNs were quantified using the determination of coefficient (R2) and root mean squared error (RMSE). This poster will present comparative analyses of three ANNs, emphasizing how multiple sensors outperforms single sensor on predicting the concentration of chlorophyll-a within wetlands.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H15J0910L