Time filter

Source Type

Schneider U.,Global Precipitation Climatology Center | Finger P.,Global Precipitation Climatology Center | Meyer-Christoffer A.,Global Precipitation Climatology Center | Rustemeier E.,Global Precipitation Climatology Center | And 2 more authors.
Atmosphere | Year: 2017

The 2015 release of the precipitation climatology from the Global Precipitation Climatology Centre (GPCC) for 1951-2000, based on climatological normals of about 75,100 rain gauges, allows for quantification of mean land surface precipitation as part of the global water cycle. In GPCC's 2011-release, a bulk climatological correction was applied to compensate for gauge undercatch. In this paper we derive an improved correction approach based on the synoptic weather reports for the period 1982-2015. The compared results show that the climatological approach tends to overestimate the correction for Central and Eastern Europe, especially in the northern winter, and in other regions throughout the year. Applying the mean weather-dependent correction to the GPCC's uncorrected precipitation climatology for 1951-2000 gives a value of 854.7 mm of precipitation per year (excluding Antarctica) or 790 mm for the global land surface. The warming of nearly 1 K relative to pre-industrial temperatures is expected to be accompanied by a 2%-3% increase in global (land and ocean) precipitation. However, a comparison of climatology for 30-year reference periods from 1931-1960 up to 1981-2010 reveals no significant trend for land surface precipitation. This may be caused by the large variability of precipitation, the varying data coverage over time and other issues related to the sampling of rain-gauge networks. The GPCC continues to enlarge and further improve the quality of its database, and will generate precipitation analyses with homogeneous data coverage over time. Another way to reduce the sampling issues is the combination of rain gauge-based analyses with remote sensing (i.e., satellite or radar) datasets. © 2017 by the authors.

News Article | December 21, 2016
Site: www.nature.com

We use potential evapotranspiration (PET) to represent the evaporative component of climate rather than actual evapotranspiration (AET), because PET is independent of precipitation, and thus carries more information about arid climates. Specifically, we reason that climates that are close to the MAP = PET transition are more likely to have been leached in the recent past than climates where PET greatly exceeds MAP, even if both climates have comparably small modern values of MAP minus AET. In this sense, modern PET may provide a better index of long-term leaching rates than modern AET. We represented PET using two contrasting approaches. In the first approach, we represented PET using a modified Penman–Monteith–Leuning model, which estimates evaporation as a function of net radiation (R ), air temperature, vapour-pressure deficit, and the aerodynamic and surface conductance of vegetation17. This approach is biophysically detailed, but it requires many parameters. Thus, in the second approach, we represented PET using the comparatively simple Priestley–Taylor equation, which models evaporative demand as a function of net radiation, air temperature, and a scaling parameter, α (ref. 29). We also explored the relationship between soil pH and the difference of MAP and AET, which we represented using the mean of the diagnostic data included in the LandFlux-EVAL synthesis18. By definition, MAP minus AET cannot take negative values, but approaches zero where PET > MAP10. We report values of MAP minus AET without imposing this constraint, and so some modelled AET values exceed MAP, resulting in slightly negative values. The Penman–Monteith–Leuning model17 partitions evaporation from the plant canopy (E ) and soil (E ). E is estimated using the Penman–Monteith equation30, while evaporation from soil is assumed to equal the equilibrium rate, modified by a moisture constraint. Because we were interested in obtaining an estimate of PET, we did not include a soil moisture constraint on evaporation, and then assumed that PET was equal to the sum of canopy and soil evaporation. Evaporation from wet soil can be approximated by multiplying the equilibrium evaporation rate by the Priestley–Taylor coefficient, α (ref. 31). Thus, we substituted the Priestley–Taylor model for the equilibrium model to represent soil evaporation in the Penman–Monteith–Leuning formula. The combined evaporation from canopy and soil are given by the equation: where A and A are the available energy absorbed by canopy and soil (R minus soil heat flux, in units of MJ m−2 d−1), λ is the latent heat of vaporization of water (MJ kg−1), s is the slope of the saturation vapour pressure curve ( kPa °C−1), ρ is the density of air ( kg m−3), c is the specific heat of air at constant pressure (MJ kg−1 °C−1), D is the vapour pressure deficit (kPa), γ is the psychrometric constant (kPa °C−1), G is the aerodynamic conductance ( m d−1), and G is the canopy conductance (m d−1). Radiation is partitioned between canopy and soil by two equations17: where A is equal to R (the soil heat flux is assumed to be negligible), L is the leaf area index, and k is an extinction coefficient. Canopy conductance (G ) is constrained by maximum stomatal conductance (g ), and modified by factors that represent dependence on light availability and vapour-pressure deficit17: where Q is photosynthetically active radiation at the top of the canopy (half of the incoming shortwave radiation), Q is a half-saturation constant for Q , D is a half-saturation constant for vapour-pressure deficit, and k is the extinction coefficient for short-wave radiation. Most parameters were obtained from a regional implementation of the Penman–Monteith–Leuning model32. The parameters k and k were both set equal to 0.6 m−1, while Q , and D were set equal to 2.6 MJ m−2 d−1 and 0.8 kPa (ref. 32). The maximum stomatal conductance, g , was set equal to 0.006 m s−1, which is a reasonable mean estimate for natural vegetation33, and scaled to a daily time step. The aerodynamic conductance, G , is influenced by windspeed and vegetation height. Because reliable maps of both these parameters are unavailable at a global scale, we used biome-specific parameters32, assigning forests and savannas a value of 0.033 m s−1, shrublands a value of 0.0125 m s−1, and grasslands, cropland, and barren areas 0.01 m s−1. All other parameters were calculated or obtained from the Food and Agriculture Organization (FAO) guidelines34. The Priestley–Taylor model for PET uses a single parameter, α, to account for adiabatic component of latent heat transfer29. While α may vary as a function of meteorological conditions30, a standard α value of 1.26 has been applied successfully at large scales35. Priestley–Taylor PET is given by the equation: where A is total available energy (equal to R ) and α = 1.26. Other parameters are listed above. We estimated mean annual precipitation (MAP) using a 1° gridded map created from the Global Precipitation Climatology Center Full Data Reanalysis, Version 7.016. MAP was calculated as the mean annual sum of monthly precipitation values for the years 1961–2001. We use this 40-year interval because it includes a high spatial coverage of rain-gauge stations36. We corrected for systematic rain gauge measurement error using static monthly under-catch corrections37 provided by the Global Precipitation Climatology Center. Both Penman–Monteith–Leuning and Priestley–Taylor models require monthly estimates of R and air temperature, and the Penman–Monteith–Leuning model requires monthly estimates of vapour pressure, atmospheric pressure, surface short-wave radiation, leaf-area index, and land cover type (Extended Data Table 2). For both approaches, environmental variables obtained for multi-year time series were collapsed to monthly means of daily values before calculation of PET, PET was scaled from daily to monthly values, then summed to obtain annual PET. To test the sensitivity of our results to driving data, we used two radiation data sets: mean monthly values from the NASA/CERES energy-balanced and filled surface radiation budget, version 2.8, over the years 2001–201438, and mean monthly values from the NASA/GEWEX surface radiation budget version 3.0 over the years 1984–200739. We obtained mean monthly values of air temperature and vapour pressure from the CRU TS3.13 data set, a gridded climatology at 0.5° resolution interpolated from weather station measurements, which we averaged over the period 1961–2001, the period of maximum weather station coverage40. Atmospheric pressure was obtained using mean elevations from the ETOPO1 global digital elevation model41 in each 1° cell and correcting using the ideal gas law34. Land cover classes were obtained from the NASA MODIS satellite product MOD1242 and monthly means of leaf area index for the period 2001–2012 were obtained from the MODIS-derived Global Land Surface Satellite leaf area index data set43, 44, averaged over the period 2001–2012. All data at a higher resolution than 1° were aggregated to mean values at 1° resolution before calculation of PET. We quantified rainfall seasonality by computing the coefficient of variation of under-catch corrected monthly rainfall values from the Global Precipitation Climatology Center data set. We estimated local topographic relief from the 1-arcminute resolution ETOPO1 digital elevation model41. Local relief was calculated as the difference between maximum and minimum elevations within a 10-km radius of each 1-arcminute cell centre. Local relief at 1° resolution was then calculated as the median relief within each 1° cell. We represented the extent of carbonate lithology using the Global Lithologic Map (GLiM)25. We determined which 1° grid cells contained carbonate rocks by subsetting the 0.5° raster version of GLiM for carbonate lithology, and then identifying all 1° cells that contained at least one 0.5° cell classified as carbonate rock. We combined data from eight soil profile databases (Extended Data Table 1)45, 46, 47, 48. Profiles were included if they were non-duplicated and included measurements of pH in soil-water suspension. We used pH in water rather than pH in CaCl or KCl solutions because pH in water is reported at a much higher frequency than pH in salt solutions. Data at 0.5 m and 0.1 m depth were obtained by selecting the horizon of each profile intersected by the corresponding depth. We selected absolute depths at 0.5 m and 0.1 m rather than soil horizons because horizon nomenclature varied across data sets. Although the choice of depths is somewhat arbitrary, the depths were selected to span the depths at which biological cycling typically influences cation concentrations49. Using the National Cooperative Soil Survey (NCSS) database as a reference, 0.5 m approximates the median value for the top of the B horizon (0.52 m) and 0.1 m approximates the median value for the midpoint of the A horizon (0.09 m). The total number of profiles included was 60,291 at 0.5 m depth and 67,900 at 0.1 m depth (Extended Data Table 1). The soil-to-water ratio of the slurry used to measure soil pH varied across data sets. To account for the effects of the soil-to-water ratio, data reported for a 1:5 ratio were corrected to a 1:1 ratio using linear correction factors50. We could not obtain correction factors for data measured at a ratio of 1:2.5, and so left these data uncorrected. Including uncorrected data is unlikely to drive large errors in the global pH distribution because changing the soil-to-water ratio from 1:1 to 1:5 shifts pH by about 0.5 units50, which is small relative to the global range of soil pH values. Data measured in water without a reported ratio were assumed to be measured at ratios of 1:1 or 1:2.5. Soil profile data were spatially resampled. In this approach, individual soil profiles were selected based on proximity to randomly distributed sampling nodes (n = 20,000). Sampling nodes were drawn from grid-cell centres at 1° resolution, with sampling weights based on cell area and allowing replacement. Nodes that were more than 100 km from a soil profile were not sampled to minimize edge biases. Soil profiles were selected by identifying the closest grid cell to each node that contained profiles, and then randomly drawing a profile from the total set of profiles in the cell. By design, this approach includes individual profiles multiple times in the resampled data set, with the consequence that geographically isolated profiles are included more frequently than profiles in densely sampled areas. This approach has no statistical derivation, but it produces sampling distributions that appear less-biased than the underlying data (Extended Data Fig. 1). To evaluate the relationship between MAP minus PET and soil pH, we compared observations to theoretical predictions based on calcite and gibbsite buffering systems. For all soils in grid cells where MAP minus PET < 0, the predicted pH was 8.2, and for all remaining profiles, the predicted pH was 5.1. Residuals from the model were then computed by subtracting predicted values from observed values. Because the data are bimodally distributed, residuals from this model have a heavy-tailed distribution, and measures of variation based on squared errors (for example, the coefficient of determination, R2) are inappropriate51. Instead, we estimated variation in the data using a robust measure of dispersion, the median absolute difference from the median (MAD). We then gauged model fit by comparing the MAD of the residuals to the MAD of the data: the percentage variation explained was equal to 1 minus MAD /MAD . This metric is analogous to R2, but makes no assumption about the distribution of the data or residuals. We estimated the uncertainty in the percentage of variation explained by resampling the data with replacement 10,000 times52 and calculating the interquartile range of the resulting distribution of parameter estimates. We defined ‘outliers’ as soils with pH < 6.5 in strongly arid climates (driest quartile of MAP minus PET) and soils with pH > 6.5 in strongly humid climates (wettest quartile of MAP minus PET). We deliberately reduced pH to this categorical expression to emphasize large-scale deviations between pH modes, rather than small-scale deviations around each mode. To quantify the prevalence of outliers as a function of rainfall seasonality, carbonate lithology, and topographic relief, we fitted logistic regressions53, 54. Likelihood ratio tests were used to compare regressions against the null hypothesis that the proportion of outliers is uniform with respect to each predictor53. We ruled out possible collinearity between environmental predictors by checking individual correlations between predictors in both wet and dry climates. No two predictors had correlation coefficients above 0.25, and so we assume that the patterns presented are independent. We used the NCSS database to validate chemical calculations and determine the relationship between climate, calcite (CaCO ), and exchangeable aluminium (Al ). We used the NCSS database for this purpose because it contains a large number of measurements of CaCO and Al using consistent methods55, and it reports the effective cation exchange capacity, which is required for modelling the pH of gibbsite buffered soils. We used a spatially resampled subset of 20,000 data points for plotting relationships with the annual water balance, following the resampling method above. The pH of a solution exposed to calcite (CaCO ) and open to the atmosphere can be solved using an equation derived from the chemical equilibria for CaCO (ref. 56): where H is the hydrogen ion activity in moles, K is the solubility constant of CaCO (in units of mol2 l−2), K is the dissociation constant for water (in mol2 l−2), K and K are the first and second dissociation constants of carbonic acid (in mol l−1), K is Henry’s constant (in mol l−1 atm−1), and is the partial pressure of CO (in atm). We solved this equation for H+ at 25 °C and a of 3.45 × 10−4 atm using the package rootSolve57 in R and published parameters58, 59. The value of 3.45 × 10−4 atm reflects imposed by laboratory measurement conditions at standard atmospheric pressure, based on the ambient CO mole fraction in 198560, the median measurement date of the data. Older measurements made at lower atmospheric CO levels may reflect a slightly higher calcite equilibrium pH (that is, the expected pH is 8.3 before 1977). Because this difference is small and the majority of measurements were taken after this date, we report model fits for a single value. Calcite concentrations are approximate, and reported as CaCO equivalents. The NCSS database reports CaCO equivalents measured using a pressure calcimeter following acid dissolution, meaning that a range of carbonate minerals are included in the estimate55. Also, because values are reported at a precision of 1%, some soils with <1% CaCO are probably reported with zero values, even if their pH reflects buffering by CaCO . The pH of a solution exposed to gibbsite (Al(OH) ) in a soil with exchangeable aluminium (Al ) depends on the ratio of Al to other exchangeable cations (Ca ). In nature, the solubility of Al(OH) and the exchange coefficients of clays do not follow the behaviour of purified laboratory solutions, and so the relationship between Al , Ca and pH must be estimated empirically61. To derive a typical pH for Al(OH) -buffered soils, we took the mean of all measurements from the spatial sample of the NCSS database with non-zero Al (pH = 5.1). Additionally, to validate the theoretical relationship between Ca /Al and pH, we fitted a model to measurements from the NCSS database taken at 0.5 m depth with non-zero Al and effective cation exchange capacity. The Gapon exchange model can be used to develop a log-linear relationship between Ca /Al and pH (ref. 61): where b and b are fitted constants. To fit the model, we assumed that Al was equal to aluminium extractable in 1 M KCl, and Ca was equal to the effective cation exchange capacity minus Al . The data show a log-linear relationship between Ca /Al and pH (Extended Data Fig. 1, b  = 4.96, b  = 0.32, R2 = 0.36, P < 0.01), supporting control of pH by Ca /Al . However, we note that the relationship appears slightly concave-curvilinear, suggesting that the Gapon model fails to account for the total activity of Al . This issue warrants further investigation. Code used to spatially resample soil profiles, calculate PET, and perform statistical analyses are maintained on GitHub and publicly archived online at http://dx.doi.org/10.5281/zenodo.61996. Code for pre-processing of raw data sets is available from the authors upon request. All soil profile and meteorological data used in this study are publicly available from the sources listed in the text and in Extended Data Tables 1 and 2. Several of the soil profile databases are only available by direct request from the providing institutions (see Extended Data Table 1). As such, the combined soil profile data set used in this study is available from the authors upon request, given permission from providing institutions.

Becker A.,Global Precipitation Climatology Center | Finger P.,Global Precipitation Climatology Center | Meyer-Christoffer A.,Global Precipitation Climatology Center | Rudolf B.,Global Precipitation Climatology Center | And 3 more authors.
Earth System Science Data | Year: 2013

The availability of highly accessible and reliable monthly gridded data sets of global land-surface precipitation is a need that was already identified in the mid-1980s when there was a complete lack of globally homogeneous gauge-based precipitation analyses. Since 1989, the Global Precipitation Climatology Centre (GPCC) has built up its unique capacity to assemble, quality assure, and analyse rain gauge data gathered from all over the world. The resulting database has exceeded 200 yr in temporal coverage and has acquired data from more than 85 000 stations worldwide. Based on this database, this paper provides the reference publication for the four globally gridded monthly precipitation products of the GPCC, covering a 111-yr analysis period from 1901–present. As required for a reference publication, the content of the product portfolio, as well as the underlying methodologies to process and interpolate are detailed. Moreover, we provide information on the systematic and statistical errors associated with the data products. Finally, sample applications provide potential users of GPCC data products with suitable advice on capabilities and constraints of the gridded data sets. In doing so, the capabilities to access El Niño–Southern Oscillation (ENSO) and North Atlantic Oscillation (NAO) sensitive precipitation regions and to perform trend analyses across the past 110 yr are demonstrated. The four gridded products, i.e. the Climatology (CLIM) V2011, the Full Data Reanalysis (FD) V6, the Monitoring Product (MP) V4, and the First Guess Product (FG), are publicly available on easily accessible latitude/longitude grids encoded in zipped clear text ASCII files for subsequent visualization and download through the GPCC download gate hosted on ftp://ftp.dwd.de/pub/data/gpcc/html/download-gate.html by the Deutscher Wetterdienst (DWD), Offenbach, Germany. Depending on the product, four (0.25°, 0.5°, 1.0°, 2.5° for CLIM), three (0.5°, 1.0°, 2.5°, for FD), two (1.0°, 2.5° for MP) or one (1.0° for FG) resolution is provided, and for each product a DOI reference is provided allowing for public user access to the products. A preliminary description of the scope of a fifth product – the Homogenized Precipitation Analysis (HOMPRA) – is also provided. Its comprehensive description will be submitted later in an extra paper upon completion of this data product. DOIs of the gridded data sets examined are as follows: doi:10.5676/DWD-GPCC/CLIM-M-V2011 025, doi:10.5676/DWD-GPCC/CLIM-M-V2011-050, doi:10.5676/DWD-GPCC/CLIM-M-V2011-100, doi:10.5676/DWD-GPCC/CLIM-M-V2011-250, doi:10.5676/DWD-GPCC/FD-M-V6-050, doi:10.5676/DWD-GPCC/FD-M-V6-100, doi:10.5676/DWD-GPCC/FD-M-V6-250, doi:10.5676/DWD-GPCC/MP-M-V4-100, doi:10.5676/DWD-GPCC/MP-M-V4-250, doi:10.5676/DWD-GPCC/FG-M-100.©Author(s) 2013.

Schneider U.,Global Precipitation Climatology Center | Becker A.,Global Precipitation Climatology Center | Finger P.,Global Precipitation Climatology Center | Meyer-Christoffer A.,Global Precipitation Climatology Center | And 2 more authors.
Theoretical and Applied Climatology | Year: 2014

In 1989, the need for reliable gridded land surface precipitation data sets, in view of the large uncertainties in the assessment of the global energy and water cycle, has led to the establishment of the Global Precipitation Climatology Centre (GPCC) at Deutscher Wetterdienst on invitation of the WMO. The GPCC has calculated a precipitation climatology for the global land areas for the target period 1951-2000 by objective analysis of climatological normals of about 67,200 rain gauge stations from its data base. GPCC's new precipitation climatology is compared to several other station-based precipitation climatologies as well as to precipitation climatologies derived from the GPCP V2.2 data set and from ECMWF's model reanalyses ERA-40 and ERA-Interim. Finally, how GPCC's best estimate for terrestrial mean precipitation derived from the precipitation climatology of 786 mm per year (equivalent to a water transport of 117,000 km3) is fitting into the global water cycle context is discussed. © 2013 The Author(s).

Walsh J.E.,University of Alaska Fairbanks | Overland J.E.,National Oceanic and Atmospheric Administration | Groisman P.Y.,National Oceanic and Atmospheric Administration | Rudolf B.,Global Precipitation Climatology Center
Ambio | Year: 2011

During the past decade, the Arctic has experienced its highest temperatures of the instrumental record, even exceeding the warmth of the 1930s and 1940s. Recent paleo-reconstructions also show that recent Arctic summer temperatures are higher than at any time in the past 2000 years. The geographical distribution of the recent warming points strongly to an influence of sea ice reduction. The spatial pattern of the near-surface warming also shows the signature of the Pacific Decadal Oscillation in the Pacific sector as well as the influence of a dipole-like circulation pattern in the Atlantic sector. Areally averaged Arctic precipitation over the land areas north of 55°N shows large year-to-year variability, superimposed on an increase of about 5% since 1950. The years since 2000 have been wetter than average according to both precipitation and river discharge data. There are indications of increased cloudiness over the Arctic, especially low clouds during the warm season, consistent with a longer summer and a reduction of summer sea ice. Storm events and extreme high temperature show signs of increases. The Arctic Ocean has experienced enhanced oceanic heat inflows from both the North Atlantic and the North Pacific. The Pacific inflows evidently played a role in the retreat of sea ice in the Pacific sector of the Arctic Ocean, while the Atlantic water heat influx has been characterized by increasingly warm pulses. Recent shipboard observations show increased ocean heat storage in newly sea-ice-free ocean areas, with increased influence on autumn atmospheric temperature and wind fields. © Royal Swedish Academy of Sciences 2012.

Herold N.,University of New South Wales | Alexander L.V.,University of New South Wales | Donat M.G.,University of New South Wales | Contractor S.,University of New South Wales | Becker A.,Global Precipitation Climatology Center
Geophysical Research Letters | Year: 2016

Despite the availability of several observationally constrained data sets of daily precipitation based on rain gauge measurements, remote sensing, and/or reanalyses, we demonstrate a large disparity in the quasi-global land mean of daily precipitation intensity. Surprisingly, the magnitude of this spread is similar to that found in the Coupled Model Intercomparison Project Phase 5 (CMIP5). A weakness of reanalyses and CMIP5 models is their tendency to over simulate wet days, consistent with previous studies. However, there is no clear agreement within and between rain gauge and remotely sensed data sets either. This large discrepancy highlights a shortcoming in our ability to characterize not only modeled daily precipitation intensities but even observed precipitation intensities. This shortcoming is partially reconciled by an appreciation of the different spatial scales represented in gridded data sets of in situ precipitation intensities and intensities calculated from gridded precipitation. Unfortunately, the spread in intensities remains large enough to prevent us from satisfactorily determining how much it rains over land. © 2015. American Geophysical Union. All Rights Reserved.

Loading Global Precipitation Climatology Center collaborators
Loading Global Precipitation Climatology Center collaborators