本数据集为大兴安岭东坡塔河地区卡马兰河流域多年冻土分布数据。结合实测钻孔和坑探数据，以地形因子、植被因子、气象因子和土壤与热状况因子数据为驱动，采用机器学习方法（随机森林）进行模型构建，总体准确率（Overall Accuracy）为0.89。可靠的精度使得此多年冻土分布数据可以作为全球变暖背景下卡马兰河流域多年冻土模拟的标定基准和历史参考。该数据格式为GeoTIFF，空间分辨率约30 m，投影为WGS_1984_Albers。", "ds_source": "
原始数据：实测钻孔和坑探数据；\n
环境变量数据：选取了地形、植被、气候及土壤四大类环境变量作为预测因子。 \n
地形因子：基于数字高程模型（DEM）获取海拔、坡度、坡向及地形起伏度。
植被因子：利用MOD13A3遥感产品提取归一化植被指数（NDVI）；基于中国森林植被碳储量数据集获取地上及地下生物量数据。 \n
气象因子：从ERA5-Land数据集提取春季降水；地表温度（GST）数据来源于HOBO监测站点的实地观测。 \n
土壤与热状况因子：整合计算融化指数、冻结指数、腐殖质厚度及土壤含水率。", "ds_process_way": "
数据预处理：\n
对上述所有多源原始栅格数据进行空间配准与标准化处理。统一投影坐标系为 WGS_1984_Albers，将空间范围裁剪至研究区边界，并采用重采样技术将所有变量的空间分辨率统一为30 m，格式统一为GeoTIFF，确保多源数据在空间上的严格匹配。 \n
提取每个样本点对应的环境变量数值，构建“样本-环境特征”高维数据集。构建的样本数据集包含目标变量（分类标签：1代表多年冻土，0代表季节冻土）及对应的特征向量。
对提取结果进行完整性检查，剔除含有缺失值（NoData）或异常值的样本，确保模型输入数据的质量。 \n
随机森林模型构建：采用200次重复5折分层交叉验证（Stratified K-Fold Cross-Validation）策略评估模型性能。为降低单次数据划分带来的随机性，设置重复次数为200，折数为5，并保持多年冻土与季节冻土的类别比例一致。基于Python 环境下的scikit-learn机器学习库构建随机森林分类模型。针对样本不平衡问题，将class_weight参数设为'balanced'，启用袋外评分（oob_score=True）作为模型泛化能力的辅助评估，模型关键参数进行合理设置，其余超参数主要采用scikit-learn 默认配置。每次重复使用不同的随机种子，以保证重复实验的独立性；最终全区预测模型使用固定随机种子训练。将环境变量作为特征输入，冻土类型（多年冻土= 1，季节冻土= 0）作为标签进行模型训练。", "ds_quality": "
本数据采用机器学习方法（随机森林）进行模型构建，总体准确率（Overall Accuracy）为0.89。", "ds_acq_start_time": "2023-08-01 00:00:00", "ds_acq_end_time": "2025-10-31 00:00:00", "ds_acq_place": "大兴安岭东坡卡马兰河流域", "ds_acq_lon_east": 123.69111111111111, "ds_acq_lat_south": 51.75694444444444, "ds_acq_lon_west": 122.48611111111111, "ds_acq_lat_north": 52.46944444444445, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "login-access", "ds_total_size": 879158, "ds_files_count": 5, "ds_format": "*.tif", "ds_space_res": "30m", "ds_time_res": "3年", "ds_coordinate": "WGS84", "ds_projection": "WGS_1984_Albers", "ds_thumbnail": "af27c5af-0c53-4d99-9968-258dabe55f03.png", "ds_thumb_from": 2, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "", "ds_from_station": null, "organization_id": "221ebf56-1b0b-4574-972b-1fb6d3cf1be7", "ds_serv_man": "敏玉芳", "ds_serv_phone": "0931-4967596", "ds_serv_mail": "ncdc@lzb.ac.cn", "doi_value": "", "subject_codes": [ "170.45" ], "quality_level": 3, "publish_time": "2026-03-31 18:21:15", "last_updated": "2026-09-28 17:23:32", "protected": false, "protected_to": null, "lang": "zh", "cstr": "11738.11.NCDC.NIEER.DB7242.2026", "i18n": { "en": { "title": "Distribution Map of 30m Permafrost in the Kamalan River Basin of Dongpo Tahe Area in Daxing'an Mountains (2023-2025)", "ds_format": "*.tif", "ds_source": "
Raw data: measured borehole and pit exploration data;\r\n
Environmental variable data: Four major environmental variables: terrain, vegetation, climate and soil were selected as prediction factors. \r\n
Terrain factors: Obtain altitude, slope, aspect and terrain relief based on digital elevation model (DEM).
Vegetation factors: Use MOD13A3 remote sensing products to extract normalized vegetation index (NDVI); obtain above-ground and underground biomass data based on the China forest vegetation carbon storage data set. \r\n
Meteorological factors: Spring precipitation was extracted from the ERA5-Land dataset; surface temperature (GST) data was derived from field observations at the HOBO monitoring station. \r\n
Soil and heat condition factors: Integrate the melting index, freezing index, humus thickness and soil moisture content.", "ds_quality": "
This data is modeled using machine learning methods (random forest), and the overall accuracy is 0.89.", "ds_ref_way": "", "ds_abstract": "
This dataset is the permafrost distribution data of the Kamalan River Basin in the Tahe area on the east slope of the Greater Xing 'an Mountains. Combined with measured borehole and pit exploration data, driven by terrain factors, vegetation factors, meteorological factors, and soil and thermal condition factor data, the model is constructed using machine learning method (random forest). The Overall Accuracy is 0.89. The reliable accuracy allows this permafrost distribution data to be used as a calibration benchmark and historical reference for permafrost simulation in the Kamaran River Basin under the background of global warming. The data format is GeoTIFF, with a spatial resolution of approximately 30 m, and the projection is WGS_1984_Albers.", "ds_time_res": "3 years", "ds_acq_place": "Kamaran River Basin on the East Slope of Daxing'an Mountains", "ds_space_res": "30m", "ds_projection": "WGS_1984_Albers", "ds_process_way": "
Data pretreatment:\r\n
All the above-mentioned multi-source raw raster data are subjected to spatial registration and standardization. The unified projection coordinate system is WGS_1984_Albers, which clips the spatial range to the boundary of the study area. Resampling technology is used to unify the spatial resolution of all variables to 30 m, and the format is unified to GeoTIFF to ensure strict spatial matching of multi-source data. \r\n
Extract the environmental variable value corresponding to each sample point and build a high-dimensional data set of \"sample-environment characteristics\". The constructed sample dataset contains the target variable (classification label: 1 represents permafrost, 0 represents seasonal permafrost) and the corresponding feature vector.
The extraction results are checked for integrity, and samples containing missing values (NoData) or abnormal values are eliminated to ensure the quality of model input data. \r\n
Random forest model construction: A 200-repeated 5-fold Stratified K-Fold Cross-Validation strategy was used to evaluate model performance. In order to reduce the randomness caused by single data division, the number of repetitions is set to 200 and the number of folds is 5, and the proportion of permafrost and seasonal permafrost categories is maintained consistent. Build a random forest classification model based on the scikit-learn machine learning library in Python environment. For the sample imbalance problem, set the class_weight parameter to 'balanced', enable out-of-bag scoring (oob_score=True) as an auxiliary evaluation of the model's generalization ability, set the key parameters of the model reasonably, and the other super parameters mainly adopt the scikit-learn default configuration. Different random seeds are used each time to ensure the independence of repeated experiments; finally, the whole-area prediction model is trained using fixed random seeds. Environmental variables are used as feature inputs and frozen soil types (permafrost = 1, seasonal frozen soil = 0) are used as labels for model training.", "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, "allow_update_data": false, "created_from": "fair", "ds_topic_tags": [ "多年冻土分布", "大兴安岭", "卡马兰流域", "中国东北" ], "ds_subject_tags": [ "地理学" ], "ds_class_tags": [], "ds_locus_tags": [ "大兴安岭东坡卡马兰河流域" ], "ds_time_tags": [ 2023, 2024, 2025 ], "ds_contributors": [ { "true_name": "臧淑英", "email": "zsy6311@163.com", "work_for": "哈尔滨师范大学", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "郭殿繁", "email": "dfguo@hrbnu.edu.cn", "work_for": "哈尔滨师范大学", "country": "中国" }, { "true_name": "陈梦瑶", "email": "cmy_543@163.con", "work_for": "哈尔滨师范大学", "country": "中国" }, { "true_name": "余江涛", "email": "yujiangtao23@163.com", "work_for": "哈尔滨师范大学", "country": "中国" } ], "ds_managers": [ { "true_name": "孙丽", "email": "sunli_wabb@163.com", "work_for": "哈尔滨师范大学", "country": "中国" } ], "category": "冻土" }