Coral Sea Oceanic Vegetation (NESP MaC 2.3, AIMS)

This dataset is a vector shapefile mapping the deep submerged aquatic vegetation on the bottom of the coral atoll lagoons in the Coral Sea within the Australian EEZ. This mapped vegetation predominantly corresponds to erect macroalgae, erect calcifying algae and filamentous algae (Tol, et al., 2023), with an average algae benthic cover of approximately 30 - 40%. This corresponds to only vegetation occurring on the soft sediment of the lagoons. This dataset was mapped from contrast enhanced Sentinel 2 composite imagery (Lawrey and Hammerton, 2022). Most of the mapped atoll lagoon areas were 30 - 60 m deep. Mapping at such depths from satellite imagery is difficult and ambiguous due to there only being a single colour band (Blue B2) that provides useful information about the benthic features at this depth. Additionally satellite sensor noise, cloud artefacts, water clarity changes, uncorrected sun glint, and detector brightness shifts all make distinguishing between high and low benthic cover at depth difficult. To compensate for some of these anomalies the benthic mapping was digitised manually based on visual cues. The most important element was to identify locations where there were clear transitions between sandy areas (with a high benthic reflectance) and vegetation areas (with a low reflectance). These contrast transitions can then act as a local reference for the image contrast between light and dark substrates. These transitions were often clearest around the many patch reefs in the lagoons which have a clear grazing halo of bare sand around their perimeter. These are often then further surrounded by an intensely dark halo, presumably from a high cover of algae. These concentric rings of light and dark substrate provided local references for the image brightness of low and high benthic cover. These cues also indicated where the hard coral substrate were. These were cut out from this dataset.Method:To map the vegetation in the Coral Sea, the primary data sources used were Sentinel-2 image composites optimized for the marine environment (Lawrey and Hammerton, 2022), high-resolution bathymetry data covering part of the region (Beaman, 2017), and drop camera survey results for validation (Tol et al., 2023). An additional set of Sentinel-2 images were collected for Ashmore Reef to help with the mapping of the vegetation in its lagoon (Lawrey and Hammerton, 2024). Most of the vegetation in the lagoonal floors of the atolls in the Coral Sea occur at a depth of 30 - 60 m. At these depths only the blue channel of the satellite imagery provides any useful visual information. Additionally the contrast between bright sand and dark vegetation is very small in the imagery for area at such depths. Artefacts in the imagery due to clouds, sun glint, waves, and sensor noise can easily obscure these small differences. To reduce the noise in the imagery a pixelwise statistical median composite was used, created from 4-10 of the clearest Sentinel-2 images of each scene manually selected from 2016-2021. Cloud masking and sun glint correction were applied before image composition (see Lawrey and Hammerton, 2022 for full details). To allow the deep benthic features to be seen the blue channel of the image composites was greatly contrast enhanced to show the very faint differences in brightness due to changes in the benthos. The amount of contrast enhancement, and thus the maximum depth that could be analysed was limited by the visual anomalies in the imagery and the magnified variations in brightness across the images.The atoll lagoonal areas were classified manually and hand digitised as bare, vegetation or reef based on the estimated benthic reflectance. Lighter benthic regions were assumed to be bare sand, while darker regions assumed to be vegetation or reef features. Determining the benthic reflectance at such depth from satellite imagery is potentially ambiguous as areas might appear dark because they are deep, covered in vegetation or reef, affected by coloured dissolved organic matter in the water column absorbing light, or there is a tonal shift from different satellite sensors across the image swath. These factors make image interpretation challenging. To resolve some of these confounding factors the mapping was done using visual cues to identify reference points across the scene to help compensate for tonal and contrast shifts due to depth, water clarity changes and the satellite sensor. These visual cues identify features where there is a high confidence in the benthic cover (sand or vegetation) and these act as local references for classifying the rest of the area between these reference locations. As most areas of the coral atoll lagoons are gently sloping, rapid changes in visual brightness are typically caused by changes in benthic reflectance, rather than changes in depth. We use this to find the edges of vegetation regions. We employed the following multi-step process to map and verify the oceanic vegetation:1. Identifying Visual Cues: We identified a set of potential visual cues to detect likely vegetated areas. These cues relied on distinguishing probable patches of sand to estimate local depth and water conditions and observing transitions between light and dark regions to identify vegetation boundaries.2. Manually mapping: The vegetation boundaries were manually hand digitised based on visual cues in the satellite imagery for Flinders and Holmes Reefs.3. Benthic Reflectance Estimation: We developed benthic reflectance estimates for the North Flinders and Holmes Reefs regions using both high-resolution and accuracy bathymetry (Beaman, 2017) and satellite imagery (see Lawrey, 2024a for details).4. Compared Analysis: The initial vegetation mapping from step 1 was compared against the benthic reflectance from step 3 to identify the most reliable visual cue techniques. This identified which were most robust against changes in depth and tonal shifts. 5. Coral Sea Mapping: Using the insights from the previous steps, we manually mapped the remaining Coral Sea region using only satellite imagery.6. Validation: The final map was validated against the available drop camera survey data on Lihou and Tregrosse Reefs.We previously, separately mapped reefs (Lawrey, 2024c). This mapping was used ensure that reef areas were not interpreted as vegetation. Reef areas were determined by their granular visual texture, their elevated central region, and by the grazing halos around their base. Visual Cues for Benthic Cover Identification:The following is a summary of the key visual cues used to classify the areas as either vegetated or unvegetated.1. Grazing Halos Around Patch Reefs: Grazing halos appear as pale rings of bare sand surrounding a textured dark, rounded feature (patch reef), see Figure 40 for examples. These occur because herbivorous fish forage and clear the surrounding sand of any algae. While grazing halos are well studied in shallow reef systems, (DiFiore et al., 2019) they are not well studied at the depths seen in the Coral Sea atoll lagoons. In this mapping we, however, assume the grazing halos in the Coral Sea are caused by a similar mechanism and thus where we see them, they indicate a central reef structure surrounded by a sandy area that is largely devoid of algae. Based on a review of bathymetry transects of patch reefs in North Flinders reefs, the depths of these grazing halos tend to be very close in depth to the surrounding lagoon. This allows them to act as an excellent reference for the brightness of sand at the depth of the lagoon in the area near the reef. Frequently, dark halos of dense vegetation surround these grazing halos, serving as a brightness reference for high-density vegetation. 2. Atoll Plains: On the atoll plains, particularly on the western side of Tregrosse Reefs platform there are large patches of dark substrate that have pale patches, unrelated to the presence of reefs. In this case, local tonal references were identified at locations where there was a clear step change in brightness and the shape and texture of the dark areas matched typical patterns of algae seen in other regions. Validation: Since this dataset was manually mapped from noisy and ambiguous imagery, validating this visual mapping approach was essential. The first form of validation was comparing the mapped vegetation boundaries with the benthic reflectance of Flinders reef. This comparison showed a very strong alignment between the manual visual mapping and benthic reflectance (Lawrey, 2024a), with the main deviations occurring around reef edges where the digitised vegetation did not capture all the details. It also deviated in areas where the benthic features were harder to see due to lower water clarity caused by coloured dissolved organic matter increasing the water column light absorption. No significant adjustments were needed to the visual cue approach following this comparison. However, it highlighted the importance of identifying the local visual cues to compensate for varying depths, satellite sensor brightness shifts and changes in the water clarity.The final validation involved comparing vegetation maps of Holmes, Tregrosse, and Lihou Reefs against the results of a drop camera survey conducted by JCU in December 2022 (Tol et al., 2023). The locations of the validation sites are shown in Figure 42. From this survey, 237 locations overlap the atoll lagoons. Figure 42 compares the vegetation density estimated from satellite mapping with the benthic cover assessed through the drop camera survey. This demonstrates a strong relationship between the mapped vegetation density and the benthic cover measured by drop cameras. The data show considerable variability, possibly due to fine-scale vegetation patchiness not captured by the satellite-based mapping. The drop camera results represent very small survey patches (less than 1 m across), while the satellite mapping represents patches around 400 m across.Areas identified as having high benthic vegetation in satellite mapping showed 15-70% (average 40%) algal benthic cover. In contrast, lagoonal regions mapped as sand (outside the identified vegetation but not on reefs) had significantly lower algal benthic cover, ranging from 0-20% with an average of 4%.Limitations:This dataset was mapped at a scale of 1:400k, with our goal being to limit the maximum boundary error to 400 m. Where the imagery was clear the mapped boundary accuracy is likely to be significantly better than this threshold. The spacing of the digitised polygon vertices was adjusted to reflect the level of uncertainty in the boundary. Where visibility was good the digitisation spacing was 100 - 200 m. In high uncertainty areas the digitised distance was increased to 500 - 1000 m. The likely boundary error is approximately equal to the vertex spacing. Many of the large areas of vegetation were littered with hundreds of small patches of lower or no vegetation. These areas were cut out as holes in the digitisation where the holes were a feature larger than 200 - 300 m in size.The vegetation areas were categorised into three levels of vegetation density (Low, Medium and High) based on how dark the substrate appeared, relative to the nearby reference indicators (dark halos around reefs, and clear patches of bare sand). In practice the accuracy of this categorisation is probably quite low, as areas were only cut into these different categories at a large scale. It was very difficult to determine the extent of the vegetation in the lagoon of Ashmore Reef. The lagoon appears to have a low flushing rate and a high level of CDOM accumulates in the lagoon, reducing the visibility to the point were most of the benthos of most of the lagoon is not visible. To help map this reef the full series of Sentinel 2 images was carefully reviewed for tonal differences that indicate the areas of sand and vegetation. Even still only 20% of the boundary of the vegetation could be accurately determined, the rest of the mapped boundary is speculative. Change Log:- 2025-09-22 Eric LawreyRAID for the NESP MaC 2.3 project was added to the metadata https://raid.org/10.71676/a017e4a2- 2024-05-21 Eric LawreyThe digitisation of the boundaries were refined by an additional 15%. This included refining the positioning of the boundaries and cutting out holes corresponding to small patch reefs and small sandy areas within larger vegetation areas. This digitisation was done as part of the preparation for the diagram showing the example light and dark local substrate references. These changes resulted in minimal change to the validation results (本数据集为矢量形状文件，用于绘制澳大利亚专属经济区（Exclusive Economic Zone, EEZ）内珊瑚海区域珊瑚环礁潟湖湖床的深水沉水水生植被分布。本次绘制的植被主要为直立大型藻类、直立钙化藻类及丝状藻类（Tol等，2023），藻类底栖覆盖度平均约为30%~40%。该数据集仅对应潟湖软沉积物上生长的植被。 本数据集通过对比度增强的哨兵二号（Sentinel-2）合成影像完成绘制（Lawrey与Hammerton，2022）。多数已绘制的环礁潟湖区域水深为30~60 m。在该水深下，仅能通过单个蓝色波段（B2波段）获取底栖特征的有效信息，因此从卫星影像中开展此类深度的底栖制图极具难度且存在歧义。此外，卫星传感器噪声、云伪影、水体透明度变化、未校正的太阳反光以及探测器亮度偏移等问题，均会加大深水区底栖覆盖度高低区分的难度。 为补偿部分此类异常，底栖制图工作基于视觉线索手动数字化完成。其中最关键的要素是识别沙地（底栖反射率较高）与植被区（反射率较低）之间的清晰边界。此类对比度边界可作为图像中明暗底物间对比度的局部参考基准。这类边界在潟湖内众多斑礁周边最为清晰，斑礁外围通常环绕着裸露沙地形成的清晰牧食晕圈（grazing halo），而该晕圈外侧往往进一步被颜色极深的光晕环绕，推测对应高藻类覆盖度区域。这种明暗底物的同心环结构可为低、高底栖覆盖度的图像亮度提供局部参考基准，同时也可用于识别硬珊瑚基质分布区域，此类区域已从本数据集中剔除。 ### 制图方法 为绘制珊瑚海区域的植被分布，本次研究采用的主要数据源包括：针对海洋环境优化的哨兵二号影像合成数据（Lawrey与Hammerton，2022）、覆盖部分研究区域的高分辨率水深地形数据（Beaman，2017），以及用于验证的水下投放式相机调查（drop camera survey）结果（Tol等，2023）。此外，研究人员还额外收集了阿什莫尔礁的哨兵二号影像，以辅助其潟湖区域的植被制图工作（Lawrey与Hammerton，2024）。 珊瑚海环礁潟湖湖床的多数植被分布于30~60 m水深区域。在该水深下，卫星影像仅蓝色波段可提供有效的视觉信息，且沙地与深色植被间的对比度极低。云层、太阳反光、海浪以及传感器噪声等带来的影像伪影极易掩盖此类细微差异。 为降低影像噪声，研究人员采用逐像素统计中值合成法，从2016-2021年手动选取的每个场景的4~10幅最清晰的哨兵二号影像中生成合成影像。影像合成前已完成云掩膜与太阳反光校正（详细方法参见Lawrey与Hammerton，2022）。为凸显深水底栖特征，研究人员对影像合成数据的蓝色波段进行了大幅对比度增强，以显示底栖生物变化带来的极微弱亮度差异。对比度增强程度以及可分析的最大水深，均受影像中的视觉异常以及影像整体亮度变化幅度的限制。 环礁潟湖区域基于估算的底栖反射率，被手动分类、手工数字化为裸地、植被或礁体区域：亮度较高的底栖区域被认定为裸露沙地，而较暗区域则被认定为植被或礁体特征。从卫星影像中判断该深度下的底栖反射率存在潜在歧义，因为区域变暗可能源于水深过大、被植被或礁体覆盖、受水体中有色溶解有机物（Colored Dissolved Organic Matter, CDOM）吸光影响，或是影像条带内不同卫星传感器产生的色调偏移。上述因素均加大了影像解译难度。 为缓解部分此类混淆因素，制图工作采用视觉线索识别场景内的参考点，以补偿因水深、水体透明度变化以及卫星传感器带来的色调与对比度偏移。这些视觉线索可用于识别底栖覆盖度（沙地或植被）可信度较高的区域，并以此作为参考基准，对参考点之间的其余区域进行分类。由于多数珊瑚环礁潟湖区域坡度平缓，视觉亮度的快速变化通常由底栖反射率变化而非水深变化引起，研究人员据此识别植被区域的边界。 本次研究采用以下多步骤流程完成大洋植被的制图与验证： 1. 视觉线索识别：识别一系列可用于检测潜在植被区域的视觉线索，该类线索依赖于区分疑似沙地斑块以估算局部水深与水体状况，并通过观察明暗区域间的边界识别植被分界。 2. 手动制图：基于哨兵二号影像中的视觉线索，手动数字化绘制弗林德斯礁与霍姆斯礁的植被边界。 3. 底栖反射率估算：结合高分辨率水深地形数据（Beaman，2017）与卫星影像，对北弗林德斯礁与霍姆斯礁区域开展底栖反射率估算（详细方法参见Lawrey，2024a）。 4. 对比分析：将步骤1得到的初始植被制图结果与步骤3得到的底栖反射率进行对比，以识别最可靠的视觉线索技术，筛选出对水深变化与色调偏移鲁棒性最强的方法。 5. 珊瑚海区域制图：基于前述步骤得到的结论，仅利用卫星影像手动完成珊瑚海其余区域的植被制图。 6. 验证：将最终制图结果与李侯礁与特雷格罗斯礁的现有水下投放式相机调查数据进行验证。 研究团队此前已单独完成礁体区域的制图（Lawrey，2024c），本次制图工作以此为依据，确保礁体区域未被误判为植被。礁体区域可通过其颗粒状视觉纹理、凸起的中心区域以及外围的牧食晕圈进行识别。 ### 底栖覆盖度识别视觉线索 以下为用于将区域分类为植被覆盖或未植被覆盖的核心视觉线索总结： 1. 斑礁周边牧食晕圈：牧食晕圈表现为环绕深色纹理圆形特征（斑礁）的浅色裸露沙地环带，示例参见图40。该现象源于植食性鱼类觅食并清除了周边沙地中的藻类。虽然牧食晕圈在浅礁系统中已得到充分研究（DiFiore等，2019），但在珊瑚海环礁潟湖的此类深度下，相关研究仍较为匮乏。本次制图仍假设珊瑚海的牧食晕圈由相同机制形成，即凡出现此类晕圈的区域，其中心为礁体结构，外围环绕着基本无藻类覆盖的沙地。通过对北弗林德斯礁斑礁的水深断面分析可知，此类牧食晕圈的水深与周边潟湖基本一致，因此可作为礁体附近区域潟湖深度下沙地亮度的优质参考基准。此类牧食晕圈外围通常环绕着高密度植被形成的深色光晕，可作为高植被密度区域的亮度参考基准。 2. 环礁平原：在环礁平原区域，尤其是特雷格罗斯礁平台西侧，存在大片深色底物区域且带有浅色斑块，此类斑块与礁体分布无关。在此类场景中，研究人员在亮度存在明显阶跃变化的区域识别局部色调参考基准，且深色区域的形状与纹理符合其他区域观测到的典型藻类分布模式。 ### 验证 由于本数据集是从存在噪声与歧义的影像中手动绘制得到，验证该视觉制图方法的有效性至关重要。第一类验证方式是将绘制的植被边界与弗林德斯礁的底栖反射率进行对比，结果显示手动视觉制图与底栖反射率具有极强的一致性（Lawrey，2024a），主要偏差出现在礁体边缘区域，此时数字化植被边界未能捕捉全部细节；此外，在因有色溶解有机物增加水体吸光能力导致水体透明度较低的区域，底栖特征较难识别，同样会出现偏差。经该对比后，视觉线索方法无需进行重大调整，但该验证凸显了识别局部视觉线索的重要性，以补偿水深变化、卫星传感器亮度偏移以及水体透明度变化带来的影响。 最终验证环节是将霍姆斯礁、特雷格罗斯礁与李侯礁的植被制图结果，与詹姆斯库克大学（James Cook University, JCU）2022年12月开展的水下投放式相机调查结果进行对比（Tol等，2023）。验证点位分布如图42所示，其中237个点位位于环礁潟湖范围内。图42对比了卫星制图估算的植被密度与水下投放式相机调查评估的底栖覆盖度，结果显示制图得到的植被密度与相机实测的底栖覆盖度之间存在显著相关性。数据存在一定离散性，可能源于卫星制图未捕捉到的细尺度植被斑块性：水下投放式相机调查的采样斑块尺寸极小（直径小于1 m），而卫星制图的采样斑块直径约为400 m。 卫星制图中被认定为高植被底栖覆盖度的区域，其藻类底栖覆盖度为15%~70%（平均40%）；而被划为沙地的潟湖区域（已识别植被区域之外且非礁体区域），藻类底栖覆盖度则显著更低，范围为0%~20%，平均仅为4%。 ### 局限性 本数据集的制图比例尺为1:400000，研究目标是将最大边界误差控制在400 m以内。在影像清晰度较高的区域，绘制的边界精度大概率显著优于该阈值。数字化多边形顶点的间距根据边界不确定性程度进行调整：在可见度良好的区域，数字化间距为100~200 m；在不确定性较高的区域，数字化间距则增大至500~1000 m，边界误差大致与顶点间距相当。 多数大面积植被区域散布着数百个小型低覆盖度或无植被斑块，当此类斑块尺寸大于200~300 m时，将其作为孔洞从数字化区域中剔除。 植被区域根据底物相对于邻近参考指标（礁体周边深色光晕、裸露沙地的清晰斑块）的暗度，被划分为三个植被密度等级（低、中、高）。实际应用中该分类的精度可能较低，因为仅在大尺度下对区域进行了等级划分。 阿什莫尔礁潟湖的植被分布范围极难确定：该潟湖水体交换率较低，有色溶解有机物大量累积，导致能见度大幅下降，潟湖多数区域的底栖生物无法通过影像观测。为辅助该礁体的制图工作，研究人员仔细审阅了全部哨兵二号影像，以识别指示沙地与植被分布的色调差异。即便如此，仅能准确确定20%的植被边界，其余绘制边界均为推测性结果。 ### 更新日志 - 2025-09-22 Eric Lawrey：为元数据添加了NESP MaC 2.3项目的RAID标识（https://raid.org/10.71676/a017e4a2） - 2024-05-21 Eric Lawrey：对边界数字化工作进行了15%的优化调整，包括修正边界位置，并剔除大型植被区域内对应小型斑礁与小型沙地的孔洞。本次数字化工作是为绘制示例明暗局部底物参考基准示意图的准备环节。本次调整对验证结果的影响极小。