Inundation Maps for NSW Inland Floodplain Wetlands

Under the NSW DPIE-EES Environmental Water Management Program the distribution and extent of inundation is monitored in large inland floodplain wetland assets which are targeted for environmental flow delivery and located in the NSW portion of the Murray-Darling Basin: Gwydir wetlands, Lowbidgee floodplain, Lower Lachlan wetlands, Macquarie Marshes, Barmah-Millewa Forest and Narran Lakes (since 2022-2023). Inundation maps are derived from image observations sourced from the satellite data sources of Landsat (30m pixel) and Sentinel-2 (10m pixel) for the period July 2014-June 2019. Image observations are automatically downloaded by NSW DPIE from the USGS (Unites State Geological Survey’s Earth Explorer website (http://earthexplorer.usgs.gov ) and the Copernicus Sentinel Open Access Hub (https://scihub.copernicus.eu/dhus/#/home ) as orthorectified images. NSW DPIE process these images to standardised surface reflectance (Flood et al. 2013). Image observations with high cloud coverage (>50%) are not considered because they cannot be processed. The inundation mapping procedure is a modified version of Thomas et. al (2015) which is a method to map inundation in vegetated floodplain wetlands using an integrated spectral response to water and vigorous vegetation. From each satellite image observation NSW DPIE-EES automatically generates a water index (Fisher et al. 2016) and the NDVI vegetation index. These indices are used to allocate inundated pixels to classes of open water, mixed water and vegetation, and dense vegetation cover that was inundated (Thomas et al. 2015). A process of pixel recoding is conducted to produce each inundation map. First all inundation classes are merged and allocated a value of one (1) whilst all other pixels are allocated a value of zero (0). Second, ancillary data is then used to identify irrigation infrastructure to do two things: locate inundated pixels within off-river storages (ORS) by recoding to a value of (2) and to remove cropped areas that have similar spectral properties to wetland vegetation by coding the pixels to a value of zero (0). Third, for observation dates affected by cloud shadow, which is often incorrectly detected as water, pixels are manually reclassified as cloud shadow by recoding them to a value of three (3). The final inundation classes are inundated (1), off-river storages with water (ors) (2), cloud shadow (3), and not inundated (0). Final inundation maps are clipped to the inland floodplain wetland boundaries.\r\n\r\nThe naming format of the files are:\r\nWetland_date _sensor_inundation1_ors2_cloud3.tif or Wetland_path_date _sensor_inundation1_ors2_cloud3.tif\r\n\r\nWetland:\r\nbm = Barmah Millewa floodplain\r\ngw = Gwydir floodplain\r\nlachlan = Lachlan floodplain\r\nlo = Lowbidgee floodplain\r\nmm = Macquarie Marshes floodplain\r\n\r\nPath: Specific to the Lachlan\r\nDate: Satellite image date processed\r\nSensor: Sensor type- l7 (Landsat7; l8 (Landsat 8); s2 (Sentinel2)\r\nInundation1: Inundated\r\nors2: Off-River Storage with water\r\ncloud3: Cloud shadow (in filename if present)\r\n\r\nReferences:\r\nFisher, A., Flood, N. and Danaher, T. (2016). Comparing Landsat water index methods for automated water classification in eastern Australia. Remote Sensing of Environment, 175, 167-182.\r\n\r\nFlood, N., Danaher, T., Gill, T., & Gillingham, S. (2013). An operational scheme for deriving standardised surface reflectance from Landsat TM/ETM+ and SPOT HRG imagery for eastern Australia. Remote Sensing, 5, 83–109.\r\n\r\nThomas, R. F., Kingsford, R. T., Lu, Y., Cox, S. J., Sims, N. C. and Hunter, S. J., (2015). Mapping inundation in the heterogeneous floodplain wetlands of the Macquarie Marshes, using Landsat Thematic Mapper. Journal of Hydrology 524, 194-213.\r\n

在新南威尔士州第一产业与环境经济服务部（NSW DPIE-EES）的环境水管理计划框架下，针对默里-达令流域新南威尔士州段内用于环境补水的大型内陆洪泛平原湿地资产，其淹没范围与分布已被纳入监测范畴，监测对象包括吉德湿地（Gwydir wetlands）、洛比奇洪泛平原（Lowbidgee floodplain）、下拉克伦湿地（Lower Lachlan wetlands）、马夸里沼泽（Macquarie Marshes）、巴马-米勒瓦森林（Barmah-Millewa Forest）以及纳兰湖（Narran Lakes），相关监测工作自2022-2023财年起正式开展。

淹没地图基于2014年7月至2019年6月期间的卫星影像观测数据生成，数据来源为陆地卫星（Landsat，像素分辨率30米）与哨兵-2号（Sentinel-2，像素分辨率10米）。

新南威尔士州DPIE会从美国地质调查局（United States Geological Survey, USGS）的地球探索者网站（http://earthexplorer.usgs.gov）以及哥白尼哨兵开放获取枢纽（https://scihub.copernicus.eu/dhus/#/home）自动下载经过正射校正的影像作为观测数据，并将这些影像处理为标准化地表反射率数据（Flood等，2013）。云覆盖占比超过50%的影像观测数据将被排除，因其无法进行有效处理。

本次淹没制图流程基于Thomas等（2015）提出的方法改进而来，该方法通过整合水体与茂密植被的光谱响应特征，实现植被覆盖洪泛平原湿地的淹没范围制图。新南威尔士州DPIE-EES会从每幅卫星影像中自动计算水体指数（Fisher等，2016）与归一化植被指数（Normalized Difference Vegetation Index, NDVI），并利用这两类指数将淹没像元划分为敞水、水-植被混合以及被淹没的茂密植被覆盖三类（Thomas等，2015）。

为生成单幅淹没地图，需执行像元重编码流程：首先将所有淹没类别合并，赋值为1，其余像元则赋值为0；其次，利用辅助数据识别灌溉基础设施，以完成两项任务：一是将河道外储水区（off-river storages, ORS）内的淹没像元重编码为2，二是将光谱特征与湿地植被相似的耕地像元重编码为0，以剔除此类非目标区域；第三，针对常被误识别为水体的云阴影影响影像，将其中的云阴影像元手动重编码为3。最终的淹没类别包括：已淹没（1）、带水河道外储水区（2）、云阴影（3）以及未淹没（0）。最终淹没地图将裁剪至内陆洪泛平原湿地的边界范围内。

文件命名格式如下：
Wetland_date_sensor_inundation1_ors2_cloud3.tif 或 Wetland_path_date_sensor_inundation1_ors2_cloud3.tif

湿地代码说明：
bm = 巴马-米勒瓦洪泛平原
gw = 吉德洪泛平原
lachlan = 拉克伦洪泛平原
lo = 洛比奇洪泛平原
mm = 马夸里沼泽洪泛平原

字段说明：
Path：仅适用于拉克伦流域
Date：处理所用卫星影像的拍摄日期
Sensor：传感器类型，l7代表Landsat 7，l8代表Landsat 8，s2代表Sentinel-2
Inundation1：已淹没区域
ors2：带水河道外储水区
cloud3：云阴影（文件名中出现即代表包含该类别）

参考文献：
Fisher, A., Flood, N. 与 Danaher, T. (2016). 澳大利亚东部区域 Landsat 水体指数自动水体分类方法对比. 《环境遥感》(Remote Sensing of Environment), 175, 167-182.

Flood, N., Danaher, T., Gill, T. 与 Gillingham, S. (2013). 澳大利亚东部区域 Landsat TM/ETM+ 与 SPOT HRG 影像标准化地表反射率提取业务方案. 《遥感》(Remote Sensing), 5, 83–109.

Thomas, R. F., Kingsford, R. T., Lu, Y., Cox, S. J., Sims, N. C. 与 Hunter, S. J. (2015). 利用 Landsat 专题制图仪绘制马夸里沼泽异质洪泛平原湿地的淹没范围. 《水文学报》(Journal of Hydrology), 524, 194-213.