TL;DR
This paper introduces SMHIGridClim, a new high-resolution (2.5 km) gridded climate dataset for Fennoscandia covering 1961-2018. It provides essential climate variables at sub-daily to daily temporal resolutions, addressing the need for detailed regional climate information for research and decision-making. The dataset demonstrates comparable quality to existing Nordic climate datasets.
What Problem Does This Paper Address?
Accurate, high-resolution climate data is crucial for understanding climate change impacts, especially in rapidly warming regions like Fennoscandia. While global climate models offer broad insights, they often lack the spatial detail needed for regional assessments. Existing gridded datasets for the Nordic region, like the Nordic Gridded Climate Dataset (NGCD), have limitations such as lacking sub-daily data or having sparse station coverage in remote areas, leading to uncertainties. This paper aims to fill this gap by providing a comprehensive, high-quality, long-term climate dataset for Fennoscandia that captures regional variations and microscale climate features.
Methods
The SMHIGridClim dataset is generated using an optimal interpolation (OI) technique implemented with the open-source software gridpp. This method combines in-situ observations with gridded forecasts (the "first guess") to produce the final analysis.
Key methodological steps include:
- Downscaling: The UERRA-HARMONIE reanalysis (11 km resolution) is statistically downscaled to the 2.5 km SMHIGridClim domain using linear least squares regression. This process accounts for diurnal and seasonal variations.
- Quality Control: Observations undergo rigorous quality control using the TITAN software, which includes checks for digital elevation, plausibility range, buddy checks, and spatial consistency. A gross error check is also applied using the first guess.
- Optimal Interpolation: The
gridpplibrary is used to perform OI, combining the downscaled UERRA fields (as the first guess) with quality-controlled observations. The analysis accounts for spatial covariances between the first guess and observations. - Variable Derivation: Relative humidity is diagnosed from dew point temperature and 2m temperature. Snow depth is derived from snow water equivalent and snow density.
Datasets
The SMHIGridClim dataset utilizes the following data sources:
- In-situ Observations:
- Swedish Meteorological and Hydrological Institute (SMHI) MORA database.
- Norwegian Meteorological Institute's Frost archive.
- Finnish Meteorological Institute's internal database.
- ECMWF's MARS archive (SYNOP station observations).
- Gridded Forecasts (First Guess):
- UERRA-HARMONIE reanalysis (11 km resolution).
- Nordic operational NWP system MEPS (2.5 km resolution) for an overlapping period.
Key Insights and Ideas
- Bridging the Resolution Gap: The downscaling of UERRA-HARMONIE to 2.5 km significantly improves the spatial detail, particularly in mountainous regions, offering a finer representation of local climate features.
- Importance of Sub-daily Data: The inclusion of 1-, 3-, and 6-hourly temperature and humidity data addresses a critical gap in many existing datasets, enabling more detailed climate analysis.
- Data Quality is Key: The rigorous quality control of observations using TITAN and the subsequent OI analysis are crucial for producing a reliable dataset, especially in regions with varying observation densities.
- Temporal Variability in Parameters: The study acknowledges that optimal OI parameters vary over time due to changing observation density, and their derivation accounts for this temporal variability.
Key Results
- SMHIGridClim provides a high-resolution (2.5 km) gridded climatology for Fennoscandia for the period 1961-2018.
- The dataset includes 2m temperature, 2m relative humidity (at 1-, 3-, or 6-hour resolution), daily min/max temperatures, daily precipitation, and daily snow depth.
- Cross-validation results show that SMHIGridClim's quality is comparable to other Nordic gridded datasets like seNorge and FMI_ClimGrid, with some variables showing slightly better performance.
- The downscaling process generally improves upon the UERRA forecasts, reducing bias and standard deviation for most parameters.
- Analysis increments reveal that the first guess often has biases (e.g., too cold temperatures in UERRA, too much precipitation in some regions), which the analysis partially corrects.
Limitations and Future Work
- Effective Resolution: While the grid resolution is 2.5 km, the effective resolution (information content) is likely larger due to the nature of numerical solutions and parametrizations in the underlying model fields.
- Inhomogeneity: The dataset is based on UERRA reanalysis (a frozen model system) combined with observations that vary in frequency, density, and quality over time. This can introduce artificial trends. The authors acknowledge that this makes it "not appropriate to use for detailed studies of trends."
- Extreme Events: The dataset may not accurately represent extreme precipitation events due to the OI method's assumption of a normal distribution and the coarse resolution of the underlying UERRA data.
- Coastal Areas: Differences in coastal regions are noted, where the land-sea contrast is not well represented by the grid scale, leading to less realistic values.
- Future Work:
- Ongoing work aims to extend the dataset to the present day.
- Development includes switching to
titanlibinstead of TITAN, adding wind speed, and using CERRA as the first guess. - Plans to incorporate observations from Denmark, Estonia, Latvia, and Lithuania are underway.
- A more thorough comparison with other gridded datasets is suggested.
- Handling of extreme precipitation events and other occasions like heavy rain/snowfall needs to be addressed separately in future analyses.
Summary updated