Data from: Robust estimation of key leaf traits from reflectance spectroscopy of herbarium specimens

Summary: Scope: 201 herbarium specimens from 13 species collected between 1876 and 2022 in the Minneapolis-St. Paul metropolitan area. Herbarium: University of Minnesota Herbarium (MIN) Publication: TBD Directory/File Descriptions: README.md: overview of dataset contents. directory: data metadata.csv: metadata spreadsheet containing required and recommended IHerbSpec Protocol metadata fields (see Part 4: https://iherbspec.github.io/protocol/part4-metadata.html), based on the IHerbSpec Metadata Spreadsheet (https://doi.org/10.5281/zenodo.15849668). MSP_spectra: unprocessed (raw) spectra data files. Filenames follow IHerbSpec Protocol filename conventions (see Table 3.1: https://iherbspec.github.io/protocol/part3-filenames.html). MSP_traits.csv: empirical trait values for leaf mass per area (LMA), foliar nitrogen content (N), and foliar carbon content (C). adaxial_spectra.normalized.csv: analysis-ready data matrix of adaxial spectra. Samples in rows, trait and reflectance data in columns. Spectra were processed as follows: co-registration/sensor matching at 1000 and 1890nm, resampled to 1nm resolution, vector-normalized, then resampled to a final resolution of 5nm. abaxial_spectra.normalized.csv: analysis-ready data matrix of abaxial spectra. Samples in rows, trait and reflectance data in columns. Spectra were processed as follows: co-registration/sensor matching at 1000 and 1890nm, resampled to 1nm resolution, vector-normalized, then resampled to a final resolution of 5nm. directory: scripts directory: data_analysis 0_extract_and_normalize.R: R script to extract reflectance spectra from .sig files, combine with metadata and trait data, and produce outputs for 1_parallel_plsr.R 1_parallel_plsr.R: R script to run PLSR analyses for complete models, models excluding NDVI (greenness) outliers, and biased models excluding samples in specific categories related to specimen age and plant function. helper scripts: parallel_spectra_utils.R: functions to summarize model statistics and setup/run PLSR models parallel_model_build.R: function to build models using k-fold or leave-one-out cross-validation parallel_optimal_number.R: function for model parameter grid search; PLSR has only one parameter, the number of components (ncomp) parallel_permutation_models.R: function for model training with the best ncomp model_assessment.R: function to summarize model outputs directory: figures 2_figures.R: R script to generate figures for the article. Figures were edited in Adobe Illustrator for publication. figures_utils.R: helper scripts to generate figures. Metadata and Trait Descriptions: Spectral data collection: Spectral data were collected from adaxial and abaxial leaf surfaces using a HR-1024i (Spectra Vista Corporation) spectrometer with a modified leaf clip to accommodate flat herbarium specimens. The leaf clip was fitted with a lens (Very Small Spot Size) to measure leaves smaller than the window of the leaf clip. The window was covered with a light-absorbing material (cardstock painted with matte black spray paint, Krylon Flat Black Camouflage Spray Paint K04290777). One spectral measurement was taken from both the adaxial and abaxial surfaces of each leaf fragment. Only one leaf was measured per specimen. For each measurement, we targeted an area on the leaf surface that lacked primary venation and was larger than the spot. Our custom black surface was placed underneath each leaf fragment to absorb transmitted light. Spectra were measured as relative reflectance to a 99% white Spectralon panel. Metadata definitions for spectral data are available from the IHerbSpec Protocol (see Part 4: https://iherbspec.github.io/protocol/part4-metadata.html) Trait data collection: We destructively sampled ~0.2g of leaf fragments from each specimen, prioritizing fragments stored in fragment packets. From these, we measured leaf mass per area (kg/m2), and carbon and nitrogen on a mass basis (%). Carbon and nitrogen were measured by combustion using a COSTECH ESC 4010 Elemental Analyzer. Dependencies: All analyses were completed in R v.4.4.3. See README.md for detailed information on dependencies.摘要：本数据集覆盖1876年至2022年间采自明尼阿波利斯-圣保罗都会区的13个物种的207份植物标本馆标本，馆藏于明尼苏达大学植物标本馆（University of Minnesota Herbarium, MIN）。相关出版物待确定（TBD）。 目录与文件说明： README.md：数据集内容概览。 data目录： metadata.csv：元数据电子表格，包含IHerbSpec元数据规范（IHerbSpec Protocol）要求的必填与推荐元数据字段，基于IHerbSpec元数据电子表格（https://doi.org/10.5281/zenodo.15849668），详见第4部分：https://iherbspec.github.io/protocol/part4-metadata.html。 MSP_spectra：未处理的原始光谱数据文件，文件名遵循IHerbSpec文件名规范（详见第3.1表：https://iherbspec.github.io/protocol/part3-filenames.html）。 MSP_traits.csv：叶面积干重（leaf mass per area, LMA）、叶片氮含量（N）与叶片碳含量（C）的实测性状值。 adaxial_spectra.normalized.csv：可供分析使用的近轴面（adaxial）光谱标准化数据矩阵，以样本为行，性状与反射率数据为列。光谱预处理流程如下：对1000nm与1890nm波段进行配准/传感器匹配，重采样至1nm分辨率，矢量归一化后，最终重采样至5nm分辨率。 abaxial_spectra.normalized.csv：可供分析使用的远轴面（abaxial）光谱标准化数据矩阵，以样本为行，性状与反射率数据为列。光谱预处理流程如下：对1000nm与1890nm波段进行配准/传感器匹配，重采样至1nm分辨率，矢量归一化后，最终重采样至5nm分辨率。 scripts目录： data_analysis目录： 0_extract_and_normalize.R：用于从.sig格式文件中提取反射光谱、结合元数据与性状数据，并生成1_parallel_plsr.R所需输出的R脚本。 1_parallel_plsr.R：用于运行偏最小二乘回归（PLSR）分析的R脚本，包含完整模型、排除归一化差分植被指数（Normalized Difference Vegetation Index, NDVI）异常值的模型，以及排除与标本年龄、植物功能类群相关的特定类别样本的有偏模型。 辅助脚本： parallel_spectra_utils.R：用于汇总模型统计量、搭建并运行PLSR模型的函数集。 parallel_model_build.R：用于构建k折交叉验证或留一法交叉验证模型的函数。 parallel_optimal_number.R：用于模型参数网格搜索的函数；PLSR仅包含一个参数，即主成分数（ncomp）。 parallel_permutation_models.R：用于基于最优ncomp模型开展模型训练的函数。 model_assessment.R：用于汇总模型输出结果的函数。 figures目录： 2_figures.R：用于生成论文配图的R脚本，最终配图经Adobe Illustrator编辑以满足出版要求。 figures_utils.R：用于生成配图的辅助脚本。 元数据与性状采集说明： 光谱数据采集：使用搭载定制化叶夹的HR-1024i型（Spectra Vista Corporation）光谱仪采集叶片近轴面与远轴面的光谱数据，该叶夹适配扁平的植物标本馆标本。叶夹配备极小光斑尺寸镜头，可测量小于叶夹视窗的叶片。叶夹视窗覆盖有吸光材料（哑光黑色迷彩喷漆，型号为Krylon Flat Black Camouflage Spray Paint K04290777）。每份标本仅选取一片叶片，分别对其近轴面与远轴面各采集一次光谱测量值。每次测量均选取叶片表面无主叶脉且尺寸大于光斑的区域。每份叶片碎片下方均放置定制的黑色吸光垫以吸收透射光。光谱以99%反射率的Spectralon标准白板作为参考，测量相对反射率。光谱数据的元数据定义详见IHerbSpec元数据规范（https://iherbspec.github.io/protocol/part4-metadata.html，第4部分）。 性状数据采集：从每份标本中破坏性取样约0.2g的叶片碎片，优先选取标本袋中留存的碎片。据此测量叶面积干重（kg/m²）、单位质量的碳与氮含量（%）。碳、氮含量通过燃烧法使用COSTECH ESC 4010型元素分析仪测定。 依赖说明：所有分析均基于R v4.4.3完成，详细依赖信息详见README.md。