随着全球气候变暖，季节冻土的冻结深度也发生了很大的变化。季冻区内的工程建设设施常常需要考虑季节冻结作用对结构主体的影响，其中冻胀作用最为典型。由于冻结条件直接决定着冻胀发生的范围和程度，因此防治冻胀病害发生的工程措施中多有涉及到冻结深度这个重要的冻结表征量。
\n本数据集主要依托分布在中国季节冻土区上的650个气象监测点1971-2020年逐年最大冻结深度观测数据；气温、积雪厚度、地表太阳辐射以及降水量等气候再分析数据（ ERA5-Land）；土壤数据集(GSDE)及数字高程模型(DEM)。在运用机器学习方法对中国季节冻土区年最大冻结深度变化研究的基础产生。运用机器学习模型进行最大冻结深度的预测，进而得到中国季节冻土区年代际最大冻结深度栅格数据。
\n数据为中国季节冻土区1971-2020年年代际最大冻结深度栅格数据，时空分辨率为0.1°，数据格式为tiff文件。
\n过去冻结深度的确定主要依赖有限实测资料，难以全面地反映大范围内的冻结深度全貌，本数据集基于机器学习方法显示了中国季节冻土区1971-2020年年代际最大冻结深度值的情况。
", "ds_source": "1.数据源列表参见Word文档；\n
2.年最大冻结深度监测数据来自中国气象科学数据共享服务网。
", "ds_process_way": "1.特征输入变量的优化：运用极端随机树（Extremely Randomized Trees Classifier）方法对冻结指数、融化指数、积雪深度、降水量、太阳辐射、DEM(海拔)以及土壤性质(土壤容重、有机质含量、砂土含量、黏土含量、淤泥质含量、砾石含量)12个潜在预测因子进行重要性排序，并选择排序在前面的预测因子做为特征输入变量。按照Gini要素的重要性排序，选择排序在前面的特征做为输入变量。研究最终选择冻结指数、融化指数、太阳辐射、DEM(海拔)、积雪深度、降水量、土壤容重以及砾石含量等8个预测因子做为机器学习模型的特征输入变量。\n
\n2.机器学习模型选择与对比：研究中使用的四种机器学习建模技术包括随机森林(random forests)，支持向量机回归(support vector machine regression)，K近邻(k-nearest neighbors)以及广义线性回归(generalized linear regression)。这些机器学习技术是基于Python中的scikit-learn模块实现。650个样本在若干次的模型训练过程中被随机分配。在一次学习训练中，其中90%的随机样本数据用于模型训练，剩余10%的数据则用于模型验证，且于每次训练结束后自动保存，分别生成相应的训练文件及验证文件用于保存训练成果，随训练次数的增加最终形成模型训练数据集合以及数据验证集合。通过对比各学习方法在统计学指标上的表现，最终选择学习迭代600次的支持向量机回归模型。\n
\n3.标准冻结深度栅格数据的形成：在Python训练程序中,通过统计学指标对比，选择并最终形成了研究所需的机器学习模型，通过支持向量机学习及合理训练次数下所构成的数据集合，表达了这一模型。将模型引入Python预测程序中，并结合1971-2020年中每年的冻结指数、融化指数、太阳辐射、DEM(海拔)、积雪深度、降水量、土壤容重以及砾石含量等栅格数据，对近50年中每一年的全国季节冻土区年最大冻结深度进行预测，并相应形成栅格数据。对相邻10年的栅格数据进行均值化，最终分别形成1970s、1980s、1990s、2000s、2010s中国季节冻土区年代际最大冻结深度栅格数据。
", "ds_quality": "数据质量良好。
", "ds_acq_start_time": "1971-01-01 00:00:00", "ds_acq_end_time": "2020-12-31 00:00:00", "ds_acq_place": "中国", "ds_acq_lon_east": 130.0, "ds_acq_lat_south": 50.0, "ds_acq_lon_west": 80.0, "ds_acq_lat_north": 20.0, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "login-access", "ds_total_size": 1930359, "ds_files_count": 3, "ds_format": "Geotiff", "ds_space_res": "0.1°", "ds_time_res": "", "ds_coordinate": "WGS84", "ds_projection": "", "ds_thumbnail": "ea7a6327-ad4b-460e-af5a-e67fffafae97.jpg", "ds_thumb_from": 2, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "用户在使用数据时请在正文中明确注明数据的来源，并在参考文献部分引用本元数据提供的引用方式。", "ds_from_station": null, "organization_id": "52b7b79b-860c-49a5-9083-9a70cf8bed5a", "ds_serv_man": "李红星", "ds_serv_phone": "0931-4967592", "ds_serv_mail": "lihongxing@lzb.ac.cn", "doi_value": "", "subject_codes": [ "170.4510", "170.5031" ], "quality_level": 3, "publish_time": "2024-07-27 11:32:51", "last_updated": "2026-05-11 17:21:11", "protected": false, "protected_to": null, "lang": "zh", "cstr": "11738.11.NCDC.NIEER.DB6551.2024", "i18n": { "en": { "title": "Interdecadal Maximum Freezing Depth Dataset of Seasonally Frozen ground in China", "ds_format": "Geotiff", "ds_source": "1. Please refer to the Word document for the list of data sources;\r\n
The annual maximum freezing depth monitoring data comes from the China Meteorological Science Data Sharing Service Network.
", "ds_quality": "The data quality is good.
", "ds_ref_way": "", "ds_abstract": "With global climate change, the freezing depth of seasonal permafrost has also undergone significant changes. Engineering construction facilities in seasonal freezing areas often need to consider the impact of seasonal freezing on the main structure, with frost heave being the most typical. Due to the fact that freezing conditions directly determine the scope and degree of frost heave, many engineering measures for preventing and controlling frost heave diseases involve the important freezing characteristic of freezing depth.
\r\nThis dataset mainly relies on the observation data of the maximum freezing depth from 650 meteorological monitoring points distributed in the seasonally frozen soil areas of China from 1971 to 2020; Climate reanalysis data including temperature, snow cover thickness, surface solar radiation, and precipitation (ERA5 Land); Soil dataset (GSDE) and digital elevation model (DEM). The foundation of using machine learning methods to study the annual maximum freezing depth variation in seasonal frozen soil areas in China has emerged. Using machine learning models to predict the maximum freezing depth, and then obtaining interdecadal maximum freezing depth raster data in China's seasonally frozen soil regions.
\r\nThe data is the interdecadal maximum freezing depth raster data of China's seasonally frozen soil region from 1971 to 2020, with a spatiotemporal resolution of 0.1 ° and data format in TIFF files.
\r\nIn the past, the determination of freezing depth mainly relied on limited measured data, which made it difficult to comprehensively reflect the full extent of freezing depth on a large scale. This dataset, based on machine learning methods, displays the maximum freezing depth values of seasonal frozen soil areas in China from 1971 to 2020.
", "ds_time_res": "", "ds_acq_place": "China", "ds_space_res": "", "ds_projection": "", "ds_process_way": "1. Optimization of feature input variables: The Extreme Randomized Trees Classifier method is used to rank the importance of 12 potential predictive factors, including freezing index, melting index, snow depth, precipitation, solar radiation, DEM (altitude), and soil properties (soil bulk density, organic matter content, sand content, clay content, silt content, gravel content), and select the top ranked predictive factors as feature input variables. Sort by the importance of Gini elements and select the features ranked first as input variables. The study ultimately selected 8 predictive factors, including freezing index, melting index, solar radiation, DEM (altitude), snow depth, precipitation, soil bulk density, and gravel content, as feature input variables for the machine learning model.
\r\n2. Machine learning model selection and comparison: The four machine learning modeling techniques used in the study include random forests, support vector machine regression, k-nearest neighbors, and generalized linear regression. These machine learning techniques are implemented based on the scikit learn module in Python. 650 samples were randomly assigned during several model training processes. In a learning training session, 90% of the random sample data is used for model training, and the remaining 10% of the data is used for model validation, which is automatically saved after each training session. Corresponding training files and validation files are generated to save the training results. As the number of training sessions increases, the final model training data set and data validation set are formed. By comparing the performance of various learning methods on statistical indicators, the support vector machine regression model with 600 learning iterations was ultimately selected.
\r\n3. Formation of standard frozen depth grid data: In the Python training program, through statistical index comparison, the machine learning model needed for the research was selected and finally formed, and the model was expressed through the data set formed by support vector machine learning and reasonable training times. Introduce the model into a Python prediction program and combine grid data such as freezing index, melting index, solar radiation, DEM (altitude), snow depth, precipitation, soil bulk density, and gravel content for each year from 1971 to 2020 to predict the annual maximum freezing depth of seasonal frozen soil areas in China for the past 50 years, and form corresponding grid data. Mean the grid data of adjacent 10 years, and finally form the inter decadal maximum freezing depth grid data of seasonal frozen soil regions in China in the 1970s, 1980s, 1990s, 2000s, and 2010s, respectively.
", "ds_ref_instruction": "" } }, "submit_center_id": "ncdc", "data_level": 0, "recommendation_value": 0, "license_type": "https://creativecommons.org/licenses/by/4.0/", "doi_reg_from": "reg_local", "cstr_reg_from": "reg_local", "doi_not_reg_reason": null, "cstr_not_reg_reason": null, "is_paper_in_submitting": false, "belong_to_nieer": false, "ds_topic_tags": [ "季节冻土", "年代际最大冻结深度", "机器学习方法" ], "ds_subject_tags": [ "岩土力学", "自然地理学" ], "ds_class_tags": [], "ds_locus_tags": [ "中国" ], "ds_time_tags": [ 1971, 1972, 1973, 1974, 1975, 1976, 1977, 1978, 1979, 1980, 1981, 1982, 1983, 1984, 1985, 1986, 1987, 1988, 1989, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020 ], "ds_contributors": [ { "true_name": "盛煜", "email": "sheng@lzb.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" }, { "true_name": "王硕", "email": "wangshuo@nieer.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "王硕", "email": "wangshuo@nieer.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" } ], "ds_managers": [ { "true_name": "盛煜", "email": "sheng@lzb.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" } ], "category": "冻土" }