Dataset: Partitioning ecosystem water fluxes into transpiration, surface evaporation, and canopy-intercepted evaporation using knowledge-guided machine learning (NEON sites)

KGML_ET_Partitioned: NEON Flux Tower Dataset for ET Partitioning Sadegh Ranjbar1, Einara Zahn2, Danielle Losos1, Sophie Hoffman1, Ojaswee Shrestha3, Elie Bou-Zeid2, Paul C. Stoy1 1Department of Biological Systems Engineering, University of Wisconsin – Madison, Madison, sWI, USA. 2Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey, USA. 3Department of Forest and Wildlife Ecology, University of Wisconsin – Madison, 1630 Linden Drive, Madison, WI 53706, USA. Corresponding author: Sadegh Ranjbar ([e-mail]) Description: High-frequency NEON flux tower dataset for partitioning total evapotranspiration (ET) into transpiration (T), surface evaporation (Es), and canopy-intercepted evaporation (Ei). Designed for use with KG-DecompNet knowledge-guided ML framework. This dataset contains half-hourly measurements from 35 NEON flux tower sites. It includes meteorological and soil variables used as predictors, as well as target ET components derived from the Flux Variance Similarity (FVS) method. The data is formatted for direct ingestion into PyTorch for knowledge-guided ML training, but can also be used in other ML or statistical workflows. File: KGML_ET_Partitioned.csvFormat: CSVNumber of sites: 35Temporal resolution: 30 minutes Columns and Units: Column Description Units Site Id Site identifier - UTC_time Timestamp (UTC) YYYY-MM-DD HH:MM:SS SoilMoisture_NEON Soil volumetric water content m³/m³ AirTemperature_NEON Air temperature °C SolarRadIn_NEON Incoming solar radiation W/m² T_CANOPY_1_1_1 Canopy temperature °C RH_NEON Relative humidity % PAR_NEON Photosynthetically active radiation µmol m⁻² s⁻¹ fluxTemp_NEON Flux tower temperature °C ustar_NEON Friction velocity m/s WindSpeed_NEON Wind speed m/s vpd_NEON Vapor pressure deficit kPa SW_IN_1_1_1 Incoming shortwave radiation W/m² SW_OUT Outgoing shortwave radiation W/m² LW_IN Incoming longwave radiation W/m² LW_OUT Outgoing longwave radiation W/m² T_soil1to4 Soil temperature layers 1–4 °C G_1to5 Ground heat flux W/m² time_since_P Time since last precipitation hours time_since_TD Time since air temperature = dew point hours Tfvs_NEON Transpiration (FVS method) W/m² Efvs_NEON Surface evaporation (FVS method) W/m² ETfvs_NEON Total ET (FVS method) W/m² Tcea_NEON Transpiration (CEA method) W/m² Ecea_NEON Surface evaporation (CEA method) W/m² ETcea_NEON Total ET (CEA method) W/m² kgml T cea KGML predicted transpiration (CEA-based) W/m² kgml Ei cea KGML predicted canopy-intercepted evaporation (CEA-based) W/m² kgml Es cea KGML predicted surface evaporation (CEA-based) W/m² kgml Dep cea KGML predicted residual flux (CEA-based) W/m² kgml T fvs KGML predicted transpiration (FVS-based) W/m² kgml Ei fvs KGML predicted canopy-intercepted evaporation (FVS-based) W/m² kgml Es fvs KGML predicted surface evaporation (FVS-based) W/m² kgml Dep fvs KGML predicted residual flux (FVS-based) W/m² Data Sources Predictor variables: NEON meteorological and soil measurements Some partitioned variables: Derived from conditional eddy accumulation (CEA) and Flux Variance Similarity (FVS) methods (Zahn et al., 2024) KGML outputs from Ranjbar et al., (2025) Timeframe: Multi-year measurements, aligned to half-hourly intervals Usage Download KGML_ET_Partitioned.csv from this repository. Load in Python using Pandas: import pandas as pd df = pd.read_csv('KGML_ET_Partitioned.csv') print(df.head()) Can be used directly with KG-DecompNet, or for other ML/statistical models for ET partitioning. Citation Ranjbar, S. (2025). KGML_ET_Partitioned: High-frequency NEON flux tower dataset for partitioning ET into T, Es, and Ei using knowledge-guided machine learning. License: This dataset is shared under the CC BY 4.0 License (Attribution). Please cite the original work when using this dataset.KGML_ET_Partitioned：用于蒸散量拆分的NEON通量塔数据集 作者：Sadegh Ranjbar¹, Einara Zahn², Danielle Losos¹, Sophie Hoffman¹, Ojaswee Shrestha³, Elie Bou-Zeid², Paul C. Stoy¹ 1 美国威斯康星大学麦迪逊分校生物系统工程系，麦迪逊，威斯康星州，美国。 2 美国普林斯顿大学土木与环境工程系，普林斯顿，新泽西州，美国。 3 美国威斯康星大学麦迪逊分校森林与野生动物生态学系，1630 Linden Drive，麦迪逊，威斯康星州53706，美国。 通讯作者：Sadegh Ranjbar（邮箱：[e-mail]） ## 数据集说明 本高频NEON通量塔数据集用于将总蒸散量（Evapotranspiration, ET）拆分为蒸腾（Transpiration, T）、地表蒸发（Surface Evaporation, Es）与冠层截留蒸发（Canopy-intercepted Evaporation, Ei）三个分量，专为配合KG-DecompNet知识引导型机器学习框架设计。 本数据集包含来自35个NEON通量塔站点的半小时级观测数据，涵盖用作预测因子的气象与土壤变量，以及通过通量方差相似性（Flux Variance Similarity, FVS）方法推导得到的目标蒸散量组分。 该数据集格式可直接导入PyTorch用于知识引导型机器学习训练，同时也可应用于其他机器学习或统计分析流程。 ### 基本信息 - 数据集文件：KGML_ET_Partitioned.csv - 格式：逗号分隔值（Comma-Separated Values, CSV） - 站点数量：35个 - 时间分辨率：30分钟 ### 字段与单位 | 字段名 | 描述 | 单位 | | ---- | ---- | ---- | | Site Id | 站点标识 | 无 | | UTC_time | UTC时间戳 | YYYY-MM-DD HH:MM:SS | | SoilMoisture_NEON | 土壤体积含水量 | m³/m³ | | AirTemperature_NEON | 空气温度 | °C | | SolarRadIn_NEON | 入射太阳辐射 | W/m² | | T_CANOPY_1_1_1 | 冠层温度 | °C | | RH_NEON | 相对湿度 | % | | PAR_NEON | 光合有效辐射 | µmol m⁻² s⁻¹ | | fluxTemp_NEON | 通量塔温度 | °C | | ustar_NEON | 摩擦速度 | m/s | | WindSpeed_NEON | 风速 | m/s | | vpd_NEON | 水汽压亏缺 | kPa | | SW_IN_1_1_1 | 入射短波辐射 | W/m² | | SW_OUT | 出射短波辐射 | W/m² | | LW_IN | 入射长波辐射 | W/m² | | LW_OUT | 出射长波辐射 | W/m² | | T_soil1to4 | 1~4层土壤温度 | °C | | G_1to5 | 1~5层地面热通量 | W/m² | | time_since_P | 末次降水后时长 | hours | | time_since_TD | 空气温度达露点后时长 | hours | | Tfvs_NEON | FVS法估算蒸腾量 | W/m² | | Efvs_NEON | FVS法估算地表蒸发量 | W/m² | | ETfvs_NEON | FVS法估算总蒸散量 | W/m² | | Tcea_NEON | CEA法估算蒸腾量 | W/m² | | Ecea_NEON | CEA法估算地表蒸发量 | W/m² | | ETcea_NEON | CEA法估算总蒸散量 | W/m² | | kgml T cea | CEA基KGML预测蒸腾量 | W/m² | | kgml Ei cea | CEA基KGML预测冠层截留蒸发量 | W/m² | | kgml Es cea | CEA基KGML预测地表蒸发量 | W/m² | | kgml Dep cea | CEA基KGML预测剩余通量 | W/m² | | kgml T fvs | FVS基KGML预测蒸腾量 | W/m² | | kgml Ei fvs | FVS基KGML预测冠层截留蒸发量 | W/m² | | kgml Es fvs | FVS基KGML预测地表蒸发量 | W/m² | | kgml Dep fvs | FVS基KGML预测剩余通量 | W/m² | ### 数据来源 - 预测变量：来自NEON的气象与土壤观测数据 - 部分拆分变量：通过条件涡度积累（Conditional Eddy Accumulation, CEA）与通量方差相似性（FVS）方法推导得到（Zahn等，2024） - 知识引导型机器学习输出结果：来自Ranjbar等（2025）的研究 ### 时间范围 多年观测数据，均对齐至半小时时间间隔 ### 使用方式 1. 从本仓库下载KGML_ET_Partitioned.csv文件 2. 使用Pandas在Python中加载： python import pandas as pd df = pd.read_csv('KGML_ET_Partitioned.csv') print(df.head()) 本数据集可直接与KG-DecompNet配合使用，或用于其他用于蒸散量拆分的机器学习/统计模型。 ### 引用格式 Ranjbar, S. (2025). KGML_ET_Partitioned: High-frequency NEON flux tower dataset for partitioning ET into T, Es, and Ei using knowledge-guided machine learning. ### 许可协议 本数据集采用知识共享署名4.0（CC BY 4.0）许可协议进行共享。使用本数据集时请引用原文献。