This dataset provides monthly near-surface air temperature data for Arctic land areas (north of 60°N) from 1982 to 2015, with a spatial resolution of 10 km and projected using the Equal-Area Scalable Earth Grid (EASE-Grid 2.0). The data are generated through the fusion of multi-source observations and reanalysis data, including station observations (Fluxnet 2015), satellite-derived land surface temperature products (such as MODIS and AVHRR), and ERA5-Land reanalysis temperature data. By employing spatial interpolation, bias correction, and multi-source collaborative fusion algorithms, the dataset effectively addresses data inconsistencies caused by sparse station coverage, cloud cover interference, and sensor discrepancies in the Arctic region.In terms of quality control, a three-step validation strategy was implemented: comparison with independent meteorological station observations showed that the root mean square error (RMSE) of monthly average temperature was controlled within 1.2°C, with a mean bias error (MBE) below 0.3°C; spatial consistency tests with high-resolution reanalysis data demonstrated correlation coefficients above 0.92 in both winter extreme cold regions and summer tundra zones; cross-validation (leave-one-out method) was used to assess interpolation uncertainty, yielding standard errors below 0.8°C in station-dense areas and no more than 1.5°C in remote regions.
This dataset is based on the systematic integration of multi-source observations and reanalysis data, specifically including: the integration of monthly temperature observations from global meteorological station networks (e.g., GHCN-D), Fluxnet 2015, and Nordic national meteorological agencies, covering typical Arctic environments such as tundra, cold deserts, coastal areas, and islands; the use of MODIS land surface temperature products (MOD11C3, 0.05° resolution) and AVHRR thermal infrared datasets (CLARA-A3, 0.25° resolution), from which clear-sky land surface temperatures are extracted through radiometric calibration, cloud masking, and emissivity correction, and converted to near-surface air temperatures based on elevation and land cover types; and the use of ERA5-Land reanalysis data (0.1° resolution) as the core framework, providing an all-weather, region-wide temperature background field, with systematic biases in high-latitude complex terrain corrected through physical-statistical methods.
This dataset achieves high-quality temperature data preparation through a systematic multi-step fusion and reconstruction process. First, all source data—including station observations, satellite-retrieved products, and reanalysis datasets—undergo preprocessing: station data are homogenized and bias-corrected to ensure climate representativeness; satellite-derived land surface temperature products are physically inverted and converted to near-surface temperatures using an emissivity library and atmospheric profiles; and reanalysis data are calibrated for systematic biases relative to station observations using quantile mapping. Subsequently, a fusion model centered on geographically weighted regression and a Bayesian maximum entropy framework is constructed, integrating multi-source data and environmental covariates (elevation, distance to sea, vegetation cover, sea ice influence, etc.) for spatial interpolation. For data gaps during polar nights, a time-series imputation model based on recurrent neural networks is established. Finally, a three-fold cross-validation approach (leave-one-station-out validation, grid-level reanalysis comparison, and extreme event process verification) coupled with energy balance rationality assessment is implemented for quality control. This process yields a reliable product that includes both temperature data and an uncertainty layer.
This dataset ensures data reliability and scientific applicability through a systematic quality control system. In terms of completeness, the integration of multi-source data effectively compensates for the issues of sparse station coverage and missing satellite observations in the Arctic region. The monthly spatial coverage reaches 99.8%. Accuracy validation employs a three-tiered strategy: comparisons with independent meteorological stations (not involved in the fusion process) show a root mean square error of 1.05°C for monthly average temperature, with the mean bias controlled within ±0.25°C; spatial consistency tests with high-resolution reanalysis data reveal correlation coefficients of 0.94 and 0.91 for the winter sea ice edge zone and the summer tundra belt, respectively.
| # | number | name | type |
| 1 | 2020YFA0608502 | Impacts of Arctic Terrestrial Environmental Changes on Land–Atmosphere Energy–Water Exchanges and Their Climate Effects | National key R & D plan |
This work is licensed under
CC BY 4.0 (Creative Commons Attribution 4.0 International License).
CSTR:11738.11.NCDC.ARCTIC-CHANGE.DB7149.2026
10.12072/ncdc.arctic-change.db7149.2026
