Monitoring blue carbon ecosystem restoration using drones and object-based classification - Data

These data were used as part of wetland restoration monitoring research, using drone imagery and object-based image classification. Drone flights were conducted over a restoration site (Fish Fry Flat, Kooragang Island, Hunter River estuary, NSW, Australia) at seven time points over a 46-month period. The raw multispectral and elevation drone data were used along with training data to classify saltmarsh and other land cover types at a saltmarsh restoration site at seven temporal points over a 46-month period. The Google Earth Engine code for this classification is available here: https://code.earthengine.google.com/f2a4ea73b8542e1e7cb4ca04ef9b8bf4. This Mendeley repository includes the classified images produced in Google Earth Engine and other associated files used to analyse the classified images in RStudio. The change tiffs were generated in QGIS by subtracting the classified images from one another. The accuracy data ("training_validation_2_alldata.xlsx") are a summary of the confusion matrix data produced in Google Earth Engine, and the variable importance data ("variable_importance_allvars.xlsx") were exported from column charts of variable importance in Google Earth Engine. The growth/loss data ("growth_loss.xlsx") refer to growth, loss and species transitions between the two major saltmarsh species at our site (see which code refers to which growth/loss transition in the "growth_loss" script on GitHub), and were generated in QGIS by reclassifying and summing classified images. The "classmetrics" and "landmetrics" csvs were produced in the "patch_metrics" R script and allow users to skip midway into the patch analysis, because the initial patch analysis takes hours to run. The data can be analysed in RStudio using the code in this GitHub repository: https://github.com/dlanceman/kooragang. The data can be used to investigate temporal and spatial changes in saltmarsh species cover over time and relationships with elevation. They can also be used to look at the importance of different variables for classification and for exploring classification accuracy between classes and over time.本数据集用于基于无人机影像与面向对象影像分类的湿地修复监测研究。 研究在澳大利亚新南威尔士州（New South Wales，NSW）亨特河河口库拉冈岛（Kooragang Island）的鱼苗滩（Fish Fry Flat）修复区开展，在46个月的周期内共完成7次无人机飞行作业。研究结合训练样本与原始多光谱、高程无人机数据，在7个时间节点上对该盐沼修复区的盐沼植被与其他土地覆被类型进行分类。本次分类所用的Google Earth Engine（谷歌地球引擎）代码可通过以下链接获取：https://code.earthengine.google.com/f2a4ea73b8542e1e7cb4ca04ef9b8bf4。 本Mendeley（曼德利）数据集仓库包含了在谷歌地球引擎中生成的分类影像，以及用于在RStudio中对分类影像开展分析的其他关联文件。变化TIFF文件通过在QGIS中对不同分类影像相减得到。精度数据文件"training_validation_2_alldata.xlsx"汇总了谷歌地球引擎中生成的混淆矩阵数据；变量重要性数据文件"variable_importance_allvars.xlsx"则从谷歌地球引擎生成的变量重要性柱状图中导出。增长/损失数据文件"growth_loss.xlsx"记录了研究区内两种优势盐沼物种之间的扩张、消退及物种转换情况（具体代码对应的增长/损失转换类型可参考GitHub上的"growth_loss"脚本），该数据通过在QGIS中对分类影像进行重分类与求和操作生成。"classmetrics"与"landmetrics"两个CSV文件由"patch_metrics"R脚本生成，可帮助用户跳过斑块分析的前置耗时步骤——初始斑块分析往往需要数小时才能完成。 用户可通过本GitHub仓库（https://github.com/dlanceman/kooragang）中的代码在RStudio中对本数据集开展分析。该数据集可用于研究盐沼物种覆盖度随时间的时空变化及其与高程的关联关系，也可用于探究分类所用不同变量的重要性，以及不同类别间、不同时段的分类精度表现。