Producing a long-term gridded data set in Finland - uncertainty and spatio-temporal trends Juha Aalto, Pentti Pirinen, Kirsti Jylhä 10th EUMETNET Data Management Workshop, St. Gallen, Switzerland, 30.10.2015
1. Introduction and aims • PLUMES consortium • Task: create a high-quality daily gridded climate data set of the key variables across 1961- 2010 (”FMI_ClimGrid_1.0”) • Focus on interpolation uncertainty • Use gridded data to investigate temporal trends in climate • Compare the results with existing data (E-OBS) 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 2
1. Introduction – gridded data • Spatially continous data based on a set of observations • Most often based on a statistical model • Important applications: climate change studies, forest management, agriculture, biosphere modelling, permafrost …. Observations Spatial grid Gridded data 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 3
2. Data – observations • Seven climate variables: • Data sources: • mean temperature ( Tday ) • FMI database • maximum temperature ( Tmax ) • ECA&D pan-European database • minimum temperature ( Tmin ) • Sweden, Norway, Russian and Estonia • precipitation sum ( Prec ) • mean relative humidity ( RH ) • air pressure ( P ) • snow depth ( Sn ) 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 4
2. Data – observations 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 5
2. Data – quality control 150km • National operational QC • ” Non-blended ” ECA&D series • Misscodings, duplicates • Local outlier detection protocol: 1. compare each value to local average and stdev (station in turn excluded) 2. Compare the local stdev to long-term monthly stdev (1961-2010) 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 6
2. Data – grid specifications • Spatial resolution = 10 km x 10 km • Euref-FIN TM35 (epsg: 3067) • 5224 points (3364 inside, 1860 outside Finland) 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 7
2. Methods – kriging interpolation • Kriging interpolates the value at given point using a weighted average of the know values inside a neighborhood • Weights are assigned by ( decreasing) function of the distance, based on the spatial covariance structure • Variogram is used to quantify the spatial dependency in the data 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 8
2. Data – background data • Used as covariates in the interpolation model (i.e. trend model) • Latitude and longitude 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 9
2. Methods – details • Separate trend model was estimated for each day • ” Semi ” climatological variogram models: • Range = monthly means of daily ranges (1961-2010) • Separate sill for each day • Nugget = 30 % of the measurement precision (e.g. 0.03 for Temp) • Exponential variogrammodels • Global kriging 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 10
2. Methods – interpolating precipitation and snow • High and potentially discrete variation vs. sparse observation network • Satellite and radar data might improve • Solution: interpolate the probability of precipitation / snow depth and combine with interpolated amounts 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 11
2. Methods - evaluation • 20 independent evaluation stations • Compare the observed and interpolated values 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 12
3. Results – interpolation accuracy 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 13
3. Results – seasonal variation in accuracy 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 14
3. Results – interpolation accuracy • Some meteorological conditions are more challenging to interpolate than others … Temperature inversion? 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 15
3. Results - uncertainty Sources of uncertainty: • Observations (measurements, network, inhomogeneities …) • Background variables (georeferencing, averaging…) • Interpolation method • 50 random permutation / day -> 50 different interpolations • Daily uncertainty estimate for each variable 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 16
3. Results – a comparison with E-OBS 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 17
4. Trends in past climate • Daily grid averages -> seasonal / annual aggregates • Most uncertain areas excluded from the trend analysis 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 18
4. Trends in past climate 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 19
5. Conclusions • Long-term gridded dataset were succesfully produced • Daily permutation-based uncertainty estimates • Clear, but locally varying signal of past climate change • Wind and solar radiation in the future • The dataset will be made freely available with regular updates • Manuscript in progress … 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 20
Computing environment • R in linux server (FMI supercomputer ”Voima”) • Required R-packages: gstat, sp , rgdal, raster, maptools, PresenceAbsence, Roracle • Total time of calculations ~ 6 days / per variable 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 21
More information: juha.aalto@fmi.fi pentti.pirinen@fmi.fi www.fmi.fi 10/30/2015 Juha Aalto | juha.aalto@fmi.fi 22
Recommend
More recommend
