The Extraction Methods Of Very Small Inland Water Bodies Based On Sentinel-2A Images

As an important component of surface water,very small inland water bodies(area less than 0.001 km~2)play an important role in microclimate regulation,greenhouse gas cycle.Accurate estimation of their area and number is essential for a comprehensive understanding of water resources state and water body greenhouse gas emissions.The use of remote sensing technique can achieve rapid and accurate extraction of water body information,and provides better technical background support and data foundation for water body research.By using Sentinel-2A satellite image and taking Chuzhou where very small water bodies are widely distributed as study objective,this research evaluated the accuracy of extracting inland water bodies based on both pixel and object-based classification methods.To explore the potential of Sentinel-2A remote sensing data in water extraction,the difference and applicability of estimation derived from two different spatial resolution images of Sentinel-2A and Landsat 8were also compared and analyzed.By using the optimum method,the area and number of very small water bodies in four seasons for the year 2018 of Chuzhou were evaluated and their seasonal variation characteristics were further analyzed.The results show that:(1)Based on the pixel classification method,the water body information extracted by the single-band threshold method is the most inaccurate.The"salt and pepper phenomenon"of the water body index method and K-Means classification method is obvious.The maximum likelihood method and decision tree classification method have relatively high accuracy,but there is still the phenomenon of misclassification and leakage of water bodies.Compared with the pixel-based methods,the classification rule set of the object-oriented classification method brought up in this study can effectively avoid the noise caused by buildings and vegetation shadows with an overall accuracy of 89.1%,therefore this method can accurately and effectively extract very small water bodies.Its producer's,user's and overall accuracy as well as Kappa coefficient is 92.6%?86.3%?89.1%and 0.9 respectively;(2)The spatial resolution is an important factor which affects the accuracy of very small water body extraction.In 2018,by using the object-oriented classification method,the very small water body area and number is 36.53 km~2and 68980 respectively in Chuzhou based on Sentinel-2A images,while the area and number is 5.43 km~2 and 6029 extracted by Landsat 8 image.Compared with Landsat 8,the accuracy of Sentinel-2A results is generally higher.The Kappa coefficient of the former is 0.9,which is 0.18 higher than the latter.It can well characterize the structure of ground features,and its ability to extract ultra-small water bodies is much better than Landsat 8.(3)The total area of water body in Chuzhou is the largest in summer and the smallest in winter.The area and number of small water bodies for the year of 2018 of Chuzhou shows distinct seasonal variation with the highest one occurred in summer and lowest in winter.But there is a sharp decline in spring and winter.Meteorological conditions such as precipitation and evaporation as well as anthropogenic activities such as the seasonal rhythm of aquaculture activities in Chuzhou are possible main reasons.