| （一）文章信息 |
标题：New insights into distinguishing temperate deciduous swamps from uplandforests and shrublands with SAR
期刊：《Remote Sensing of Environment》(中科院1区Top, IF=13.5)
作者：Sarah Banks et al.
doi：https://doi.org/10.1016/j.rse.2024.114377
| （二）研究背景 |
| （三）研究数据与方法 |
（四）研究结果 |
Fig. 8. Shapley values for single (HH, HV, VH or VV polarization) and grouped sets of variables (by SAR image acquisition date and by DEM, topographic wetness index (TWI) and slope (SLP) together) to demonstrate their relative importance to the independent overall accuracy (total bar height) of multiple Random Forest models (averaged across 100 runs). Shown above are all models based on single SAR images (dates) classified alone (top row) and with a DEM plus derivatives (bottom row). Shapley values are summed above each bar to indicate the independent overall accuracy of each model.
| （五）研究结论 |
本文通过对SAR影像和辅助数据的深入分析，总结出多时相影像结合DEM及其衍生数据在湿地分类中的重要性。研究表明，尽管传统观点认为较长波长的L波段更适合湿地识别，但在优化的采集时间下，C波段数据同样可以提供高精度的分类结果，特别是在生长期早期采集的HH极化数据。多时相影像的引入有效补偿了单一时相数据的局限性，尤其是在生长季节后期，影像的时效性对分类精度影响显著。
| （六）主要图表 |
Fig. 9. Example classification results (one model run per time series) based on intensity derivatives (HH, HV, VH and/or VV polarization) for a select number of single SAR images (dates), trained using all 400 samples per class. For comparison, dates for the Radarsat-2 and Sentinel-1 time series correspond to the closest acquisitions to (1) the Alos-2 image acquired on May 20th, and (2) the Alos-2 image acquired on August 3rd.
Fig. 10. Shapley values for grouped sets of Radarsat-2 variables (by SAR acquisition date and by DEM, topographic wetness (TWI) and slope (SLP) together) to demonstrate their relative importance to the independent overall accuracy (total bar height) of multiple Random Forest models (averaged across 100 runs). Shown above are all possible two-date combinations, classified alone (top row) and with a DEM and derivatives (bottom row).
Fig. 11. Shapley values for grouped sets of Sentinel-1 and Alos-2 variables (by SAR acquisition date and by DEM, topographic wetness (TWI) and slope (SLP) together) to demonstrate their relative importance to the independent overall accuracy (total bar height) of multiple Random Forest models (averaged across 100 runs). For the Sentinel-1 time series, only the three models with the highest and the three models with the lowest accuracies are shown for brevity, while all possible two-date combinations are shown for Alos-2. Models above are for those classified alone (top row) and with a DEM and derivatives (bottom row).
文章来源 ：
Banks, S., Millard, K., Dingle-Robertson, L., & Duffe, J. (2024). New insights into distinguishing temperate deciduous swamps from upland forests and shrublands with SAR. Remote Sensing of Environment, 314, 114377.
