土壤表层温度（SST）是地气能量交换的重要变量，也是评估冻土状态变化的关键指标。遥感技术能够反演出大范围、长时序的陆表温度数据（LST），然而，陆表温度与土壤表层温度之间受到植被、积雪等因素影响存在偏差。本研究基于MODIS LST数据，结合植被、积雪、土壤组分、地形和太阳辐射五种环境因子，通过逐月建立这些数据与土壤表层温度之间随机森林模型的方法，生产了2003–2023年环北极（45°N以北）1 km空间分辨率的月平均SST数据集。使用站点观测数据进行精度验证发现，SST数据集均方根误差（RMSE）范围为1.14–2.09 °C，R²在0.7以上，优于现有的分季相建立多元线性回归模型方法生产的SST数据集，在低植被区域（如，稀树苔原、草地、湿地等）的冷季（9月至次年4月）和高植被区（如，森林）的暖季（5月至8月）精度优势明显。该数据集可为多年冻土分布制图、活动层厚度估算等研究提供基础支撑。", "ds_source": "
土壤表层（0-10cm）温度站点观测数据来自SoilTemp(https://doi.org/10.1111/gcb.15123)和Global Terrestrial Network for Permafrost (GTN-P)两个数据库（https://gtnp.arcticportal.org/）。\n
MODIS数据产品包括LST、归一化植被指数（NDVI）、叶面积指数（LAI）、归一化积雪指数（NDSI）、积雪覆盖度、太阳辐射，来自google earth engine(https://earthengine. google.com/)数据集。\n
土壤组分数据包括干容重、黏土含量、粉粒含量、有机碳含量、沙粒含量、含水量，来自SoilGrid250数据集（https://soilgrids. org/）。\n
地形数据来自CopernicusDEM数据集（https://registry.opendata.aws/copernicus-dem/）。\n
气候数据包括积雪厚度、积雪密度、降水，来自ERA5-Land数据集(https://earthengine. google.com/)。", "ds_process_way": "
（1）数据预处理。对所有数据取月平均值（剔除土壤表层温度站点观测数据中观测天数少于26天的月份）。\n
（2）建立逐月样本库。提取站点观测数据对应位置的MODIS LST数据和环境因子数据作为样本，将1–12月份的逐月样本分别随机选择70 %用于模型训练，30 %用于精度测试。\n
（3）对1–12月份逐月训练随机森林模型，并分别将这些模型用于对应月份的2003–2023年逐年的土壤表层温度预估，结果中存在缺失值的区域使用相邻月份的平均值进行填充。", "ds_quality": "
采用均方根误差（RMSE）和决定系数（R²）作为指标，使用测试集站点观测数据对生产的数据集进行精度验证。验证结果表明，2003–2023年环北极土壤表层温度数据集整体RMSE为1.14–2.09 ℃，R²为0.79-0.93。森林区域整体精度较高，冷季RMSE为0.85–1.36 ℃，R²为0.81–0.94；暖季RMSE为0.85–1.42 ℃、R²为0.66–0.88。草地区域在冷季精度较高，而在暖季精度有所下降，RMSE最高为1.76 ℃、R²为0.44。湿地区域在冷季精度较高，RMSE为1.22–1.83 ℃、R²为0.52–0.73；暖季精度稍低，RMSE为1.71–1.83 ℃、R²为0.68–0.74。稀树草原和灌木区域，冷季RMSE为1.3–2.5 ℃、R²为0.63–0.85；暖季RMSE升至1.7–2.5 ℃、R²为0.44–0.85。裸地区域冷季精度较低，RMSE为2.18–2.88 ℃、R²为0.56–0.60，而暖季精度有所上升，RMSE为1.24–1.34 ℃、R²为0.78。", "ds_acq_start_time": "2003-01-01 00:00:00", "ds_acq_end_time": "2023-12-31 00:00:00", "ds_acq_place": "环北极地区", "ds_acq_lon_east": 180.0, "ds_acq_lat_south": 45.0, "ds_acq_lon_west": -180.0, "ds_acq_lat_north": 90.0, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "login-access", "ds_total_size": 38030553510, "ds_files_count": 3, "ds_format": "Geotiff", "ds_space_res": "1km", "ds_time_res": "月", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "0fb47ccf-054c-4876-9476-34357dee2bb4.png", "ds_thumb_from": 2, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "", "ds_from_station": null, "organization_id": "53943799-d453-4bf2-a141-56c205c1355b", "ds_serv_man": "李红星", "ds_serv_phone": "0931-4967592", "ds_serv_mail": "ncdc@lzb.ac.cn", "doi_value": "", "subject_codes": [ "170.15", "170.4510" ], "quality_level": 3, "publish_time": "2026-05-13 15:37:22", "last_updated": "2026-05-13 15:39:16", "protected": false, "protected_to": null, "lang": "zh", "cstr": "11738.11.NCDC.ARCTIC-CHANGE.DB7158.2026", "i18n": { "en": { "title": "Monthly Soil Surface Temperature Dataset around the Arctic (2003-2023)", "ds_format": "Geotiff", "ds_source": "
Soil surface (0–10 cm) temperature observations were obtained from two databases: SoilTemp (https://doi.org/10.1111/gcb.15123) and the Global Terrestrial Network for Permafrost (GTN-P) (https://gtnp.arcticportal.org/).\r\n
MODIS data products, including land surface temperature (LST), normalized difference vegetation index (NDVI), leaf area index (LAI), normalized difference snow index (NDSI), snow cover fraction, and solar radiation, were sourced from Google Earth Engine (https://earthengine.google.com/) datasets.\r\n
Soil composition variables—bulk density, clay content, silt content, organic carbon content, sand content, and soil moisture—were obtained from the SoilGrids250 dataset (https://soilgrids.org/).\r\n
Topographic data were derived from the Copernicus DEM dataset (https://registry.opendata.aws/copernicus-dem/).\r\n
Climatic variables, including snow depth, snow density, and precipitation, were obtained from the ERA5-Land dataset (https://earthengine.google.com/).", "ds_quality": "
The accuracy of the generated dataset was evaluated using in-situ observations from the test set, with the root mean square error (RMSE) and coefficient of determination (R²) as assessment metrics.Validation results show that for the period 2003–2023, the overall RMSE of the circum-Arctic soil surface temperature dataset ranges from 1.14 °C to 2.09 °C, with R² values between 0.79 and 0.93.In forest regions, overall accuracy is high: during the cold season, RMSE ranges from 0.85 °C to 1.36 °C with R² between 0.81 and 0.94; during the warm season, RMSE ranges from 0.85 °C to 1.42 °C with R² between 0.66 and 0.88.In grassland regions, accuracy is higher in the cold season but decreases in the warm season, with RMSE reaching up to 1.76 °C and R² as low as 0.44.In wetland regions, accuracy is also higher during the cold season (RMSE = 1.22–1.83 °C; R² = 0.52–0.73), while slightly lower in the warm season (RMSE = 1.71–1.83 °C; R² = 0.68–0.74).For savanna and shrubland regions, RMSE ranges from 1.3 °C to 2.5 °C and R² from 0.63 to 0.85 in the cold season; in the warm season, RMSE increases to 1.7–2.5 °C and R² ranges from 0.44 to 0.85.In bare land regions, accuracy is relatively low during the cold season (RMSE = 2.18–2.88 °C; R² = 0.56–0.60) but improves in the warm season (RMSE = 1.24–1.34 °C; R² = 0.78).", "ds_ref_way": "", "ds_abstract": "
Soil surface temperature (SST) is an important variable for energy exchange between the earth and atmosphere, and a key indicator for evaluating changes in permafrost conditions. Remote sensing technology can invert large-scale, long-term land surface temperature data (LST). However, there are biases between land surface temperature and soil surface temperature due to factors such as vegetation and snow cover. This study is based on MODIS LST data, combined with five environmental factors including vegetation, snow cover, soil composition, terrain, and solar radiation. By establishing a random forest model between these data and soil surface temperature on a monthly basis, a monthly average SST dataset with a spatial resolution of 1 km around the Arctic (north of 45 ° N) from 2003 to 2023 was produced. Using station observation data for accuracy verification, it was found that the root mean square error (RMSE) range of the SST dataset is 1.14-2.09 ° C, with an R ² of 0.7 or above, which is superior to the SST dataset produced by existing seasonal multiple linear regression modeling methods. The accuracy advantage is significant in low vegetation areas (such as sparse tundra, grassland, wetlands, etc.) during the cold season (September to April of the following year) and high vegetation areas (such as forests) during the warm season (May to August). This dataset can provide fundamental support for the mapping of permafrost distribution and estimation of active layer thickness.", "ds_time_res": "", "ds_acq_place": "Circumpolar region", "ds_space_res": "", "ds_projection": "", "ds_process_way": "
（1）Data preprocessing. Compute monthly means for all datasets (excluding months in the soil surface temperature station records with fewer than 26 observation days).\r\n
（2）Build month-wise sample sets. At station locations, extract the corresponding MODIS LST and environmental factor data as samples. For each month (January–December), randomly select 70% of samples for model training and 30% for accuracy testing.\r\n
（3）Modeling and prediction. Train a separate random forest model for each month (January–December), and use the corresponding model to estimate soil surface temperature for each year from 2003 to 2023 for that month. For areas with missing values in the results, impute using the mean of the adjacent months.", "ds_ref_instruction": "" } }, "submit_center_id": "ncdc", "data_level": 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, "ds_topic_tags": [ "土壤表层温度", "环北极", "多种环境因子", "逐月建模", "MODIS陆表温度" ], "ds_subject_tags": [ "大气科学", "自然地理学" ], "ds_class_tags": [], "ds_locus_tags": [ "环北极地区" ], "ds_time_tags": [ 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022, 2023 ], "ds_contributors": [ { "true_name": "朱文泉", "email": "zhuwq75@bnu.edu.cn", "work_for": "北京师范大学", "country": "中国" }, { "true_name": "郭红翔", "email": "202131051035@mail.bnu.edu.cn", "work_for": "北京师范大学", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "朱文泉", "email": "zhuwq75@bnu.edu.cn", "work_for": "北京师范大学", "country": "中国" }, { "true_name": "郭红翔", "email": "202131051035@mail.bnu.edu.cn", "work_for": "北京师范大学", "country": "中国" } ], "ds_managers": [ { "true_name": "朱文泉", "email": "zhuwq75@bnu.edu.cn", "work_for": "北京师范大学", "country": "中国" } ], "category": "极地" }