Assessment of Rainforest TECS on NSW Crown Forest Estate

Operational map for Lowland Rainforest:\r\n\r\nThe operational map for Lowland Rainforest (LORF) was constructed to resolve long-standing issues surrounding its identification, location and extent within the NSW State Forest estate covered by the coastal Integrated Forestry Operation Agreements. \r\nThe project’s Threatened Ecological Community (TEC) Reference Panel (the Panel) preceded the assessment process by reviewing the determination for LORF. The Panel found that the determination for LORF relies almost exclusively on a rainforest classification system described by Floyd (1990) where several rainforest ‘suballiances’ make up the LORF assemblage. Floyd’s suballiance classifications presented a challenge to our project as they were largely subjective and were not compatible with quantitative analysis, meaning that it was difficult to distinguish between the LORF TEC and other rainforest vegetation using statistically sound methods. \r\nTo overcome some of these problems we revisited a set of reference sites that were assigned by Floyd to the suballiances cited in the LORF determination and in other rainforest TEC determinations, and collected new floristic data using standard flora survey methods. We also targeted a range of localities on State Forest that we considered likely to include LORF and other rainforest TECs based on the suballiance descriptions, cited localities in Floyd (1990), and preliminary distribution models. Over 300 new rainforest plots were combined with a large pool of existing data covering eastern NSW to construct a provisional revised rainforest classification. We used the rainforest groups derived from this analysis to compare the species composition of Floyd’s suballiances, determination assemblage lists and recent rainforest classifications included in regional classifications. Rainforest groups (and the plots that defined them) were assigned to the Floyd suballiance with the highest degree of floristic similarity. We conferred with the Panel to resolve any inconsistencies between the results of our analyses and statements relating to the distribution and composition of individual suballiances in Floyd (1990) and the determinations.\r\nWe then used plot data and a selection of environmental and remote-sensing variables to develop a Random Forest (RF) model of the probability of occurrence of LORF. We assessed the location of plots assigned to LORF against the distribution of the RF model on and adjoining State Forests. We then completed detailed aerial photograph interpretation (API) using a prescribed set of mapping classes to delineate rainforest areas for a range of canopy cover thresholds. We constructed an operational map of LORF by assigning our API polygons as being LORF based on the modelled probabilities and plot data underlying the polygon.\r\nOur mapping identified a total of approximately 14,036 hectares of LORF, the vast majority of which was located in the north coast region. We mapped 13,209 hectares of LORF on the north coast, with the largest areas found in Ewingar and Unumgar State Forests. Only 827 hectares of LORF were mapped on the south coast, with the largest areas found in Yadboro and Currowan State Forests.\r\n\r\nOperational map for Lowland Rainforest on Floodplains:\r\n\r\nThe operational map for Lowland Rainforest on Floodplains (LRFP) was constructed to resolve long-standing issues surrounding its identification, location and extent within the NSW State Forest estate covered by the coastal Integrated Forestry Operation Agreements. \r\nThe project’s Threatened Ecological Community (TEC) Reference Panel (the Panel) preceded the assessment process by reviewing the determination for LRFP. The Panel found that the determination for LRFP relies mainly on a rainforest classification system described by Floyd (1990) where several rainforest ‘suballiances’ make up the LRFP assemblage. The determination also identifies a range of floodplain and alluvial descriptors. Floyd’s suballiance classifications presented a challenge to our project as they were largely subjective and were not compatible with quantitative analysis, meaning that it was difficult to distinguish between the LRFP TEC and other rainforest vegetation using statistically sound methods. \r\nTo overcome some of these problems we revisited a set of reference sites that were assigned by Floyd to the suballiances cited in the LRFP determination and other rainforest TEC determinations, and collected new floristic data using standard flora survey methods. We also targeted a range of localities on State Forest that we considered likely to include LRFP and other rainforest TECs based on the suballiance descriptions, cited localities in Floyd (1990), and preliminary distribution models. Over 300 new rainforest plots were combined with a large pool of existing data covering eastern NSW to construct a provisional revised rainforest classification. We used the rainforest groups derived from this analysis to compare the species composition of Floyd suballiances, determination assemblage lists and recent rainforest classifications included in regional classifications. Rainforest groups, (and the plots that defined them), were assigned to the Floyd suballiance with the highest degree of floristic similarity. We conferred with the TEC Project Reference Panel (the Panel) to resolve inconsistencies between the results of our analyses and statements relating to the distribution and composition of individual suballiances in Floyd (1990), and the determinations.\r\nWe attempted to use plot data and a selection of environmental and remote-sensing variable to develop Random Forest models of the probability of occurrence of LRFP, but we were unable to assign any of our rainforest groups to the assemblage lists or the primary suballiances cited in the LRFP determination. We overcame this problem by constructing a fine scale digital elevation model (DEM) of landscape elements that we considered were likely to be associated with the range of floodplain and alluvial descriptors identified in the determination for LRFP. We then mapped our rainforest groups onto the DEM and assigned any rainforest assemblage that overlapped with our alluvial and floodplain DEM map as LRFP TEC. Using this method we constructed an operational map of LRFP in State Forests on the NSW coast. Our mapped identified a total of 680 hectares of LRFP, all of which was located in the north coast region. \r\n\r\nOperational TEC Mapping have been derived by API at a viewing scale between 1-4000 using ADS40 50 cm pixel imagery and 1 m derived LIDAR DEM grids for floodplain EECs.

低地雨林（Lowland Rainforest, LORF）作业图

本研究构建低地雨林（Lowland Rainforest, LORF）作业图，旨在解决沿海综合林业作业协议覆盖的新南威尔士州（New South Wales, NSW）州立林地产权范围内，该群落的识别、分布与范围界定长期存在的争议问题。

本项目的受威胁生态群落（Threatened Ecological Community, TEC）咨询专家组（下称“专家组”）在评估流程启动前，先对LORF的划定方案进行了审查。专家组发现，LORF的划定方案几乎完全依赖Floyd（1990）提出的雨林分类系统，该系统将多个雨林“亚联盟（suballiance）”归为LORF的组成群落。但Floyd提出的亚联盟分类方法存在较大主观性，且无法适配定量分析，为本项目带来了挑战——这意味着难以通过严谨的统计学方法区分LORF受威胁生态群落与其他雨林植被。

为解决上述问题，本研究回溯了Floyd在LORF划定方案及其他雨林受威胁生态群落划定方案中指定的一批参照样地，并采用标准植物调查方法采集了新的植物区系数据。同时，本研究参照Floyd（1990）中提及的样地信息、亚联盟描述以及初步分布模型，选定了一批州立林地区域，这些区域被认为大概率存在LORF及其他雨林受威胁生态群落。本研究将300余个新增雨林样地与覆盖新南威尔士州东部的大量现有数据集整合，构建了临时性修订版雨林分类系统。

本研究利用本次分析得到的雨林群落组，对比了Floyd的亚联盟分类、划定方案的群落列表，以及区域分类体系中纳入的最新雨林分类系统的物种组成。将各雨林群落组（及其对应的样地）匹配至区系相似性最高的Floyd亚联盟。本研究与专家组沟通，解决了分析结果与Floyd（1990）中各亚联盟的分布、组成描述及划定方案之间存在的矛盾之处。

随后，本研究利用样地数据与筛选后的环境及遥感变量，构建了LORF发生概率的随机森林（Random Forest, RF）模型。本研究将划定为LORF的样地位置，与随机森林模型在州立林地区域及其周边的分布结果进行了比对。之后，本研究采用一套标准化的制图分类体系，开展了详细的航空影像解译（Aerial Photograph Interpretation, API），针对一系列冠层覆盖阈值划定了雨林区域范围。本研究基于模型预测概率与多边形对应的样地数据，将航空影像解译得到的多边形归为LORF范围，最终构建了LORF作业图。

本次制图共识别出约14036公顷的LORF区域，其中绝大多数分布于北海岸区域。北海岸区域共划定13209公顷LORF，其中面积最大的区域位于Ewingar与Unumgar州立林场。南海岸区域仅划定827公顷LORF，其中面积最大的区域位于Yadboro与Currowan州立林场。

泛滥平原低地雨林（Lowland Rainforest on Floodplains, LRFP）作业图

本研究构建泛滥平原低地雨林（Lowland Rainforest on Floodplains, LRFP）作业图，旨在解决沿海综合林业作业协议覆盖的新南威尔士州州立林地产权范围内，该群落的识别、分布与范围界定长期存在的争议问题。

本项目的受威胁生态群落（TEC）咨询专家组（下称“专家组”）在评估流程启动前，先对LRFP的划定方案进行了审查。专家组发现，LRFP的划定方案主要依赖Floyd（1990）提出的雨林分类系统，该系统将多个雨林“亚联盟”归为LRFP的组成群落，同时划定方案还纳入了一系列泛滥平原与冲积地貌描述。但Floyd提出的亚联盟分类方法存在较大主观性，且无法适配定量分析，为本项目带来了挑战——这意味着难以通过严谨的统计学方法区分LRFP受威胁生态群落与其他雨林植被。

为解决上述问题，本研究回溯了Floyd在LRFP划定方案及其他雨林受威胁生态群落划定方案中指定的一批参照样地，并采用标准植物调查方法采集了新的区系数据。同时，本研究参照Floyd（1990）中提及的样地信息、亚联盟描述以及初步分布模型，选定了一批州立林地区域，这些区域被认为大概率存在LRFP及其他雨林受威胁生态群落。本研究将300余个新增雨林样地与覆盖新南威尔士州东部的大量现有数据集整合，构建了临时性修订版雨林分类系统。

本研究利用本次分析得到的雨林群落组，对比了Floyd的亚联盟分类、划定方案的群落列表，以及区域分类体系中纳入的最新雨林分类系统的物种组成。将各雨林群落组（及其对应的样地）匹配至区系相似性最高的Floyd亚联盟。本研究与TEC项目咨询专家组（下称“专家组”）沟通，解决了分析结果与Floyd（1990）中各亚联盟的分布、组成描述及划定方案之间存在的矛盾之处。

本研究尝试利用样地数据与筛选后的环境及遥感变量，构建LRFP发生概率的随机森林模型，但未能将任何雨林群落组匹配至LRFP划定方案中提及的群落列表或核心亚联盟。为解决该问题，本研究针对划定方案中明确的泛滥平原与冲积地貌特征，构建了高精度景观元素数字高程模型（Digital Elevation Model, DEM）。随后，本研究将雨林群落组映射至该数字高程模型，并将与冲积、泛滥平原DEM图层重叠的雨林群落归为LRFP受威胁生态群落。采用该方法，本研究在新南威尔士州沿海州立林区完成了LRFP作业图的构建。本次制图共识别出680公顷的LRFP区域，全部分布于北海岸区域。

泛滥平原受威胁生态群落作业制图采用航空影像解译（API）方式生成，解译比例尺为1:4000，使用了ADS40 50厘米分辨率像素影像与面向泛滥平原EECs的1米分辨率激光雷达（Light Detection and Ranging, LIDAR）DEM栅格数据。