科研之友
微信
新浪微博
Facebook
分享链接
摘要
The surface water salinity of Florida Bay has been deteriorating due to anthropogenic interventions and natural disturbances for almost over a century, yet astoundingly, this issue perceiving limited exposure. Driven by this, this study developed a spatially weighted optimization model (SWOM) to predict the dry and wet season's surface water salinity in Florida Bay. The modeling was carried out using in situ salinity observations from United States Geological Survey (USGS) for 1998-2001 coupled with processed Landsat TM 5 images for corresponding dates. A mathematical optimization formulation was developed for the model, which was solved using a gradient-based optimization solver, following the clustering of the study area into a set of 3 km * 3 km grids. To train the model so as to infer optimum values of decision variables, the optimization process was simulated for each 3 km pixel using two-third randomly selected samples, which yielded acceptable R-squared values of 0.92 and 0.90 for dry and wet seasons and root mean squared error (RMSE) values of 1.8 and 1.73 parts per thousand (ppt) for dry and wet seasons, respectively. The validation of the model with the remaining one-third samples also provided good statistical correlation, as the R-squared values found were 0.91 (dry) and 0.89 (wet); and the validated RMSE values obtained were 1.92 ppt (dry) and 1.84 ppt (wet), providing strong inference of the model's consistency in estimating surface water salinity. Lastly, the model was applied to six different temporal cases (i.e., dry and wet seasons of 1996-97, 2004-05, and 2016). Upon application, the model provided reasonable accuracy except for showing errors in some regions for 2004-05. The surface water salinity mapping carried out using SWOM not only provided meaningful spatial and temporal variations of surface water salinity but also highlighted the wet season hypersaline peaks within the central Florida Bay region.
- 出版日期2017-10
- 单位中国地质大学（北京）
