USE OF UNMANNED AERIAL VEHICLE DATA IN NEAR-INFRARED REGION TO ESTIMATE WATER QUALITY OF MIHARU DAM RESERVOIR, JAPAN

摘要

Lake Sakurako is a reservoir of the Miharu dam in Fukushima Prefecture, Japan. The water quality of the small lake becomes significantly worse during the summer, owing to the occurrence of blue-green algae. Taking into account the limited water quality monitoring data available for the lake, we previously used a fuzzy regression analysis (FRA) of water quality measurements and water conditions that appear in near-infrared (NIR) data collected by unmanned aerial vehicles (UAVs). Then, fuzzy c-means (FCM). which shows relative differences in water quality, was also applied for the analysis. Furthermore, we investigated a noise removal process using a non-local mean (NLM) filter and demonstrated that the process provides more detailed information regarding the lake's water quality. However, a comparison of classification results with respect to differences in analysis methods has not yet been conducted. Therefore, this paper describes the differences in classification results obtained by both FRA and FCM using an NLM filter. Water quality data sampled at 20 points synchronized with UAV data were acquired. Five water quality parameters were directly measured. The analysis method adopted is comprised of preprocessing. NLM filtering. FRA. fuzzy level slices, and FCM. FRA assumes that the differences between observation data and the model prediction indicate the system fuzziness. thus revealing the relation between the input and the output. FRA was carried out for each combination of data: the UAV NIR data and the measurements of water quality parameters. For FCM. first, the study area was divided into two classes: C1 and C2. The initial point of C1 was selected from an average value of 2% from the minimum value of the study area, and the initial point of C2 was selected from an average value of 2% from the maximum value of the study area. Then, the degree of belonging to C2 was divided into preset levels. The results suggest that the application of both FRA and FCM using an NLM filter to understand the water quality is more effective than the simple FRA method. When the presence of blue-green algae was high, it became clear that FRA using an NLM filter can estimate the water quality more accurately than FCM on a 256-level gray scale.