Optimizing Remote Sensing-Based Level–Area Modeling of Large Lake Wetlands: Case Study of Poyang Lake

Remote sensing-derived level-area models have been widely used in inundation analysis of large lakes. The current study aimed to optimize the model for Poyang Lake, the largest freshwater lake in China, where the hydrological connections are highly dynamic and complex. The inundation data delineated using 217 MODIS images between 2003 and 2005 together with concurrent water level data were used to analyze the level-area model accuracy and its associated influential factors. It has been demonstrated that the primary model uncertainty was introduced by the image selection in terms of both magnitude and temporal distribution. The results from random sampling simulations indicate that at least 40 remotely sensed images are required to assure a stable linear regression model. In addition, the selection of gauging stations, where the water level measurements were collected, could serve as another error source to the model. If the model input (water level) changes between different gauging stations, the variability of the output (inundation area) could reach to 144.49 km2. Moreover, the model performance could be improved through the matched regression functions, where the average improvement among different regression functions is 134.44 km2. Of the 40 selected models, the logistic regression based on the lake's inundation patterns appears to be the best, resulting in an R2 of 0.98 and uncertainty of 100.45 km2. This report describes the first attempt in which the logistic function has been used in level-area models development.