Global crop calendars of maize and wheat in the framework of the WorldCereal project

Crop calendars provide valuable information on the timing of important stages of crop development such as the planting (SOS) and harvesting (EOS) dates. This information is critical for many crop monitoring applications such as crop type mapping, crop condition monitoring, and crop yield estimation and forecasting. Spatially detailed information on the crop calendars provides an important asset in this respect, as it allows the algorithms to account for specific local circumstances while also maximizing their robustness and global applicability. Existing global crop calendar products, as produced by the Group on Earth Observations' Global Agricultural Monitoring (GEOGLAM) Crop Monitor, the United States Department of Agriculture Foreign Agricultural Service (USDA-FAS), the Food and Agriculture Organization (FAO), and the European Commission Joint Research Centre's Anomaly hot Spots of Agricultural Production (ASAP), generally provide this information only at national or subnational level. In this work, we present gridded SOS and EOS maps for wheat and maize that represent the crop calendars' spatial variability at 0.5° spatial resolution. These maps are generated in the framework of WorldCereal, which is a European Space Agency (ESA) funded project whose cropland and crop type wheat and maize algorithms at global scale and at 10 m spatial resolution require this information. The proposed maps are built leveraging the above noted global products (Crop Monitor, USDA-FAS, FAO, ASAP) whose datasets are combined into a baseline map and sampled to train a Random Forest algorithm based on climatic and geographic data. Their evaluation against test data from the baseline maps set aside for validation purposes show a good performance with SOS (EOS) R2 of 0.87 (0.92) and a Root Mean Square Error (RMSE) of 27 (26) days for wheat, showing the lowest errors (RMSE < 15 days) in North America, Central Europe, South Africa, and Australia, all critical areas for global wheat production and trade. Meanwhile, the largest errors (RMSE between 40 and 60 days) occurred in regions of South America close to the Amazon forest and in Africa close to the Congo Basin. In the case of maize, the SOS (EOS) evaluation shows a R2 of 0.88 (0.79) and a RMSE of 24 (28) days for maize, with the best performing regions (RMSE < 15 days) located in the Northern Hemisphere, South Africa, and Australia, important areas for global maize production and trade. Meanwhile, the worst performing regions were in Brazil, Saudi Arabia and India. Additionally, the crop calendars were evaluated using a simple Land Surface Phenology (LSP) model based on Sentinel-2 and Landsat 8 Earth Observation data from Sentinel-2 and Landsat 8 over known wheat and maize fields. The results show a SOS (EOS) R2 of 0.75 (0.88) and a RMSE of 25 (18) days for wheat and SOS (EOS) R2 of 0.80 (0.88) and a RMSE of 35 (24) days for maize. Therefore, the presented calendars present an advancement over the existing crop calendar products in terms of capturing spatial coverage and variability and reporting their accuracy.File description:HTML_Maize_FINAL: Land Surface Phenology model (LSP) for maizeHTML_WinterWheat_FINAL: Land Surface Phenology model (LSP) for Winter-wheatM1_EOS_WGS84: Gridded map for end of season of the first season of maizeM1_SOS_WGS84: Gridded map for start of season of the first season of maizeWW_EOS_WGS84: Gridded map for end of season of the first season of winter wheatWW_SOS_WGS84: Gridded map for start of season of the first season of winter wheat

作物物候历（crop calendars）可提供作物发育关键阶段的重要时序信息，例如播种始期（SOS，Start of Season）与收获终期（EOS，End of Season）。此类信息对诸多作物监测应用至关重要，涵盖作物类型制图、作物状况监测以及作物产量估算与预测。具备空间细节的作物物候历信息在此方面具有重要价值，可使算法兼顾局地特殊情境，同时最大限度提升算法的鲁棒性与全球适用性。
当前已有的全球作物物候历产品，多由地球观测组织全球农业监测小组（Group on Earth Observations' Global Agricultural Monitoring, GEOGLAM）作物监测系统、美国农业部海外农业服务局（United States Department of Agriculture Foreign Agricultural Service, USDA-FAS）、粮食及农业组织（Food and Agriculture Organization, FAO）以及欧盟委员会联合研究中心农业生产异常热点（European Commission Joint Research Centre's Anomaly hot Spots of Agricultural Production, ASAP）产出，通常仅提供国家级或次国家级尺度的信息。
本研究构建了适用于小麦与玉米的网格化播种始期（SOS）与收获终期（EOS）分布图，以0.5°的空间分辨率呈现作物物候历的空间变异特征。此类分布图依托欧洲空间局（European Space Agency, ESA）资助的WorldCereal项目框架生成——该项目的全球尺度10m分辨率耕地、作物类型（小麦与玉米）算法亟需此类信息支撑。
本研究提出的分布图，通过整合上述全球产品（作物监测系统、USDA-FAS、FAO、ASAP）的数据集以构建基准分布图，并基于气候与地理数据采样训练随机森林（Random Forest）算法生成。
针对预留作验证用途的基准图集测试数据开展评估，结果显示小麦的播种始期（SOS）与收获终期（EOS）决定系数R²分别为0.87与0.92，均方根误差（Root Mean Square Error, RMSE）分别为27天与26天。在北美、中欧、南非与澳大利亚这些全球小麦生产与贸易的关键区域，模型误差最低（RMSE<15天）；而南美亚马逊雨林周边区域与非洲刚果盆地周边区域的误差最大，均方根误差介于40至60天之间。
就玉米而言，其播种始期（SOS）与收获终期（EOS）的评估结果分别为R²=0.88与0.79，均方根误差分别为24天与28天。模型表现最优的区域（RMSE<15天）位于北半球、南非与澳大利亚——这些均为全球玉米生产与贸易的重要区域；而表现最差的区域则为巴西、沙特阿拉伯与印度。
本研究还基于哨兵-2号（Sentinel-2）与陆地卫星8号（Landsat 8）的地球观测数据，依托简单地表物候（Land Surface Phenology, LSP）模型，在已知的小麦与玉米田块上对作物物候历开展验证。结果显示，小麦的播种始期（SOS）与收获终期（EOS）决定系数分别为0.75与0.88，均方根误差分别为25天与18天；玉米的播种始期（SOS）与收获终期（EOS）决定系数分别为0.80与0.88，均方根误差分别为35天与24天。
综上，本研究提出的作物物候历在空间覆盖范围、空间变异捕捉能力以及精度报告维度上，均实现了对现有作物物候历产品的优化升级。
文件说明：
HTML_Maize_FINAL：玉米地表物候（LSP）模型
HTML_WinterWheat_FINAL：冬小麦地表物候（LSP）模型
M1_EOS_WGS84：玉米第一季收获终期（EOS）网格化分布图（WGS84坐标系）
M1_SOS_WGS84：玉米第一季播种始期（SOS）网格化分布图（WGS84坐标系）
WW_EOS_WGS84：冬小麦第一季收获终期（EOS）网格化分布图（WGS84坐标系）
WW_SOS_WGS84：冬小麦第一季播种始期（SOS）网格化分布图（WGS84坐标系）

搜集汇总

背景与挑战

背景概述

该数据集提供了全球玉米和小麦的作物日历信息，包括种植和收获日期，空间分辨率为0.5°。通过结合多种全球产品和随机森林算法，数据集在空间覆盖和变异性方面表现出色，并提供了详细的准确性评估。