Classification of Aquaculture Waters through Remote Sensing on the Basis of a Time-Series Water Index

Abstract

Zhang, C.; Gao, L.; Lu, Z.; Liu, H.; Zhu, H., and Tang, K., 2022. Classification of aquaculture waters through remote sensing on the basis of a time-series water index. Journal of Coastal Research, 38(6), 1148–1162. Coconut Creek (Florida), ISSN 0749-0208. Accurate spatial distribution information on aquaculture is crucial for scientific aquaculture management, postdisaster evaluation, and water environment protection. At present, large-scale aquaculture areas can be quickly and accurately monitored using remote-sensing technology. However, most previous studies gave importance only to extracting aquaculture areas and ignored the classification of aquiculture types. Although several researchers classified aquaculture modes, the waters in aquaculture areas were unclassified. Moreover, classification of the aquaculture target is usually performed on a single remote-sensing image, which lacks time-series information. To examine existing issues, the current study selected the coastal marine aquaculture areas in the southern Lvushunkou District of Dalian. First, a normalized difference aquaculture water index (NDAWI) was constructed on the basis of the measured data through a spectral feature analysis. Second, NDAWI was extracted from Sentinel-2 images to construct a time-series data set, and the random forest classification method was applied to classify kelp and wakame aquaculture waters. Last, the experimental outcomes of the NDAWI data set were compared with those of other time-series data sets. Results showed that the overall accuracy of the NDAWI time-series data set was 92.11%, which showed an improvement of 7.68%, 17.66%, 4.50%, and 11.50% compared with the values for normalized difference chlorophyll index, normalized difference vegetation index, and greenness component of tasseled cap transformation data sets and single-phase images, respectively. The NDAWI time-series data set constructed in this study exhibited a dramatically improved classification accuracy compared with other time-series data sets. It has a high application value for identifying relevant aquaculture areas.