Improving the resolution of UAV-based remote sensing data of water quality of Lake Hachiroko, Japan by neural networks

Remote sensing techniques for periodically collecting global data have been widely used for water quality monitoring. Satellites with a ground resolution of 10 to 30 m obtain information over broad areas, making it difficult to use them to evaluate local water quality. Methods for improving satellite data resolution allow for water quality monitoring over both wide and local areas. However, previous studies have failed to target water bodies that undergo drastic changes; moreover, they have not sufficiently examined features contributing to resolution improvement. This study proposes a resolution improvement method using a neural network, which performs learning so that the output matches the high-resolution data when the target pixel and its surrounding pixels in the low-resolution data are input. Moreover, the band ratio of data obtained from an unmanned aerial vehicle was used in the learning process as an input feature. We investigated the (i) band ratio providing highly accurate resolution improvement and (ii) application of a new resolution improvement method to the estimation of suspended solid conditions for water quality parameters. Finally, the proposed method was compared with the bicubic method for validation. The results indicate that the estimated map at the band ratio B/R in the resolution improvement data created via the proposed method can be used to greatly improve the resolution in areas with high levels of suspended solids, compared to the water quality estimation maps created using the bicubic method.