GPCC Drought Index – a new, combined, and gridded global drought index

The Global Precipitation Climatology Centre Drought Index (GPCC-DI) provides es-timations of precipitation anomalies with respect to long term statistics. It is a combination of the Standardized Precipitation Index with adaptations from Deutscher Wetterdienst (SPI-DWD) and the Standardized Precipitation Evapotranspiration Index 5 (SPEI). Precipitation data were taken from the Global Precipitation Climatology Centre (GPCC) and temperature data from NOAA’s Climate Prediction Center (CPC). The GPCC-DI is available with several averaging periods of 1, 3, 6, 9, 12, 24 and 48 months for di ﬀ erent applications. Since spring 2013, the GPCC-DI is calculated op-erationally and available back to January 2013. Typically it is released at the 10th 10 day of the following month, depending on the availability of the input data. It is calculated on a regular grid with 1 ◦ spatial resolution. All averaging periods are integrated into one netCDF-ﬁle for each month. This dataset can be referenced by the DOI: 10.5676/DWD_GPCC/DI_M_100 and is available free of charge from the GPCC website ftp://ftp.dwd.de/pub/data/gpcc/html/gpcc_di_doi_download.html.


Introduction
Drought indices are a measure of anomalies of available water with respect to long term statistics. They can be based on station as well as on grid data like precipitation, temperature, wind speed, radiation, evaporation but also non-meteorological data like soil type, soil moisture or ground water level. It is common to classify three main con-20 ditions: drought, normal and wet. Drought conditions are refined to moderate, severe and extreme drought (Palmer, 1965).
Drought indices allow to distinguish easily between the different conditions. They provide information on the onset and end, their length and severity of droughts.
There are several types and definitions of droughts. Usually, droughts are defined 25 as a shortage of available water (Heim, 2002;Wilhite and Glantz, 1985). The differ- Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ences between aridity and drought have to be kept in mind. For example, droughts are categorized as meteorological, agricultural or hydrological droughts, with a variety of definitions (Anderson et al., 2010;Heim, 2002;McKee et al., 1993;Wilhite and Glantz, 1985) for each category. They can also differ in the time span of the water deficit.
A precipitation deficit of one or two consecutive months has usually an impact on the 5 agricultural yield whereas longer lasting deficits have an impact on ground water levels and river runoffs (e.g. Sustek and Vido, 2013).

Some existing drought indices
Depending on the available data and the target application specific drought indices were developed. Some of the most common are discussed in this section. A thorough 10 review of drought indices is given in Heim (2002). One frequently utilized drought index is the Palmer Drought Severity Index (PDSI, Palmer, 1965). It is based on several empirical relationships. The PDSI uses precipitation, evapotranspiration, soil water recharge, runoff, water loss from soil and an empirical weighting factor (Lloyd-Hughes and Saunders, 2002). Also the soil water storage 15 capacitiy is needed. The evapotranspiration can be measured or calculated using several parameterizations. A strength of the PDSI lies in its high level of standardization. On the other hand it is build on empirical relationships and comes with a high demand on input data that can be hardly addresed on the global scale (Lloyd-Hughes and Saunders, 2002). 20 Another drought index is the Reconnaissance Drought Index (RDI, Tsakiris and Vangelis, 2005). It is based on the ratio of precipitation and potential evapotranspiration (PET). Again, the potential evapotranspiration can be calculated applying several parameterizations or measured (see Sect. 5). It is possible to standardize the RDI. However, the RDI is not defined, if the PET is zero.

25
The Standardized Precipitation Index (SPI) was developed by McKee et al. (1993). It is based on the anomaly of (monthly) precipitation divided by the standard deviation of precipitation for this time span (e.g. one month). Only precipitation data are needed, which is a big advantage. On the other hand, if the temperature changes due to climate change, the SPI has not the ability to take the increased evapotranspiration into account. This leads also to misleading values in arid areas (Lloyd-Hughes and Saunders, 2002). An adjustment of the SPI by DWD, called SPI-DWD, fixed this problem 5 (Pietzsch and Bissolli, 2011). Nevertheless, the usage of the SPI was recommended by WMO (WMO, 2009). A user guide of the SPI is also given in WMO (2012). A recently developed drought index is the Standardized Precipitation Evapotranspiration Index (SPEI, Vicente-Serrano et al., 2009). It is based on the difference between precipitation and PET. As for RDI and PDSI, the calculation or measurement of the 10 PET is the main challenge. The calculation of the SPEI is similar to the calculation of the SPI. As an alternative to the SPI, the SPEI was suggested by WMO (WMO, 2009).

Some existing drought index data sets and monitoring tools
There are a number of regional and global drought index data sets and monitoring tools. They differ in target region, input data, processing of input data (like regridding), 15 utilized drought index, timeliness and scope of provided information and data (e.g., figures, bulletins or data files). Some of them are briefly described below, but it is not a comprehensive review! For instance, the "Drought Management Centre for Southeastern Europe" (DMC-SEE) provides figures and bulletins based on SPI for Southeastern Europe (http: 20 //www.dmcsee.org/en/drought_monitor/, last access: 9 December 2013). Precipitation data are taken from the GPCC and downscaled to a higer resolution of 0.01 • . The data are available back to 1986. Another European drought data set is produced by the "Pilot Regional Climate Centre on Climate Monitoring" (RCC-CM) for the WMO-region RA VI (Europe and the Mid- There is a monitoring tool based on precipitation data from the GPCC and tempera-20 ture data from the NOAA NCEP CPC GHCN_CAMS gridded dataset (Fan and van den Dool, 2008). The estimation of the PET is based on the parameterization from Thornthwaite (1948). Additionally, an improved version for climatological applications is provided from 1901 to 2011 based on the version 3.2 of the CRU dataset (Jones and Harris, 2013) with an enhanced estimation of the PET.

Used data
The GPCC-DI is based on gridded precititation data and gridded monthly mean temperatures. Because of the necessary high timeliness only a limited number of data sets are currently available in such a timely fashion. The "First Guess Product" of the GPCC (Ziese et al., 2011, doi: 5 10.5676/DWD_GPCC/FG_M_100) is available three to five days after the end of each month and used as precipitation input data. It is based on monthly totals calculated from SYNOP-reports, interpolated to a regular grid using a modified SPHEREMAP scheme (Willmott et al., 1985) and a background climatology. A detailed desription of the data set, data base and interpolation is given in Becker et al. (2013).

10
On the other hand monthly mean temperatures were applied from the NOAA NCEP CPC GHCN_CAMS gridded data set (Fan and van den Dool, 2008), where the temperature data are taken from GHCN version 2 (Peterson and Vose, 1997) and CAMS (Ropelewski et al., 1984). Due to the higher timeliness of the CAMS data set, CAMS is more important for the drought index calculation. The station data are interpolated by 15 means of a Cressman based objective analysis scheme and a background climatology. The temperature data are delivered orginally with 0.5 • spatial resolution (see Fig. 1). To match the GPCC grid the data are regridded to 1 • spatial resolution taking land portion and area-average into account (see Fig. 2).

20
The calculation of the GPCC-DI is based on the above mentioned gridded data sets. We decided to use gridded data as input data and not station data, because both data sets have spent a lot of efforts to reach a high quality level. Also, we think an interpolation of station based drought indices is more error-prone than the calculation based on gridded data. Additionally, station data suffers from gaps in time series, relocation, The calculation of the GPCC-DI takes place in two steps. First, the SPI-DWD and SPEI are calculated independently on the grids where possible. Afterwards, the GPCC-DI is computed using the mean of both indices for each grid cell, where both indices are valid. For the other grid cells the index that is still possible to be calculated is taken.
Although the DWD-adaptation of the SPI (Pietzsch and Bissolli, 2011) was taken 5 into account, the SPI-DWD cannot be computed in very arid regions. This is because the applied gamma distribution to describe the distribution of the precipitation amounts has not a shape with a maximum above zero, which is neccessary to calculate the SPI (see also Wu et al., 2007). Examples of the SPI-DWD with averaging intervals of one, three and six months are depicted in Figs. 3, 4 and 5. Green indicates precipitation 10 around normal and blue precipitation above normal. Droughts occure in red areas, where precipitation is below normal. Grids without data are white. These areas are the oceans, Antarctica as well as areas where the SPI-DWD cannot be calculated due to the above mentioned limitations of the applied gamma function (see also Sect. 5.1). The PET is computed according to the algorithm from Thornthwaite (1948). This 15 algorithm fails in areas where the mean temperature is near or below 0 • C (see Fig. 6 and Figs. 1 and 2). Even if other parameterizations for the computation of the PET exist, they need more input data than temperature and astronomical data. To our knowledge, no data sets exist that provide for example wind speed, radiation or humidity (dew point) with the high timeliness, global coverage and spatial resolution like the applied 20 temperature data set. Also existing PET data sets are not applicable, because they are provided as figures or files unemployable for further automated usage (e.g. http: //earlywarning.usgs.gov/fews/global/index.php, last access: 9 December 2013).
To estimate the SPEI, the above described PET data and precipitation data are utilized. In areas, where the PET cannot be computed, the calculation of the SPEI is also  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | As mentioned above, the gridwise combination of SPI-DWD and SPEI yields the GPCC-DI. Examples of the GPCC-DI with averaging periods of 1, 3 and 6 months are shown in Figs. 10, 11 and 12. The combination is possible due to the comparable indicating of both indices (see for example Figs. 3, 7 and 10). A nearly global coverage is possible by the combination of both drought indices. Otherwise the one which can be 5 computed is applied. Only cold arid areas cannot be covered with this approach (e.g. Tibet).

How to calculate the parameters for the SPI-DWD and SPEI
To estimate the parameters of the distribution function of the SPI-DWD and SPEI, data from the reference period 1961 to 1990 were used. Temperature data were 10 taken from the above mentioned temperature data set, regridded to the GPCC grid with 1 • spatial resolution. Due to its higher data coverage and the more rigorous quality control applied, the Full Data Reanalysis Version 6 with 1 • spatial resolution was utilized for the precipitation parameterization (Schneider et al., 2011, doi: 10.5676/DWD_GPCC/FD_M_V6_100). The parameters for precipitation and PET were 15 calculated for each grid box and averaging intervall seperately.
Depending on the parameters of the gamma distribution used, the distribution has a maximum at zero or above zero. Parameter sets leading to a maximum of the distribution at zero are not applicable to compute SPI-DWD values. This occures in areas with monthly mean precipitation at about zero millimeters. Due to the averaging of several 20 months for the longer averaging periods, the mean precipitation total for this period can result in a maximum of the distribution above zero, even if some months of this period have monthly totals of zero millimeter. Therefore, the SPI-DWD could be calculated for some grid cells only for the longer sampling intervals, for instance in arid regions in India, China, Mongolia, Southern and Central Africa or Southern and Central America  The provided drought index is a standardized precipitation anomaly. The value of the index corresponds with the σ-value of a standardized normal distribution and can be interpreted as the SPI (see Table 1, Lloyd-Hughes and Saunders, 2002). Negative values correspond with precipitation totals less than normal − drought -whereas positive 5 values conform with precipitation totals wetter than normal. Values between −1 and 1 match to the 1σ-environment and are defined as normal conditions or mild drought/wet.

Access to the GPCC-DI
The GPCC-DI can be downloaded as netCDF-files (net, 2014) from the DOI-referenced website ftp://ftp.dwd.de/pub/data/gpcc/html/gpcc_di_doi_download.html. No registra-10 tion is required to download the data. The file for each month contains seven sets of GPCC-DI data for the different averaging intervals. These averaging intervals are 1, 3, 6, 9, 12, 24 and 48 months. The GPCC-DI is provided at a regular global grid ( tre (CDC) of Deutscher Wetterdienst (DWD), which disseminates ISO 19139 compliant metadata on its data sets through the Geo-Network software application. The data set is regulary updated at the tenth day of each month. It can be delayed, if input data are not available in time. In this case, it will be delivered upon availability of the missing data.

Conclusions
The Global Precipitation Climatology Centre Drought Index (GPCC-DI) is a new gridded drought index with nearly global coverage. It is a combination of the SPI-DWD and SPEI and based on precipitation analyses from the Global Precipitation Climatology Centre (GPCC) and temperature data from the NOAA NCEP CPC. The spatial resolu-10 tion is 1 • latitude by longitude. Seven averaging intervals are provided: 1, 3, 6, 9, 12, 24 and 48 months to cover several applications from meteorological droughts to hydrological droughts. All averaging intervals are summarized to one downloadable netCDFfile (ftp://ftp.dwd.de/pub/data/gpcc/html/gpcc_di_doi_download.html). The download is free of charge and no registration is required. Due to limitations in the validity range 15 of the underlying drought indices, the GPCC-DI cannot be provided for cold arid areas like the Southern Andes and the Himalayan, so in areas where it would not be of use anyway given the hostile conditions there.
The GPCC-DI is available back to January 2013. If the input data are available, the data set is updated regulary at the tenth of each month. It should only be used for 20 monitoring purposes because the input data are not homogenized. Therefore we don't recalculate the data set for earlier years.
Acknowledgements. First of all we are most appreciative to the data suppliers who are to the largest extent the worldwide spread National Meteorological and/or Hydrological Services, but also other institutes. These data contributions have put GPCC into the position to provide the 25 underlying global precipitation analyses applied for the described data set, and we are look- Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ing forward to their further contributions, which are crucial in order to maintain and enhance GPCC's level of products in terms of scope and quality.
We would like to thank also NOAA NCEP CPC for providing the NOAA NCEP CPC GHCN_CAMS gridded data set and their permission to use these data.  Meteorol. Soc., 83, 1181Soc., 83, -1190Soc., 83, , 2002. 247 Thornthwaite, C.: An approach towards a rational classification of climate, Geogr. Rev., 38, 55-94, 1948. 247, 249, 264 Tsakiris, G. and , 10, 111-120, doi:10.1080/02508068508686328, 1985. 244, 245 5 Willmott, C., Rowe, C., and Philpot, W.: Small-scale climate maps: a sensitivity analysis of some common assumptions associated with grid-point interpolation and contouring, Am. Carthographer, 12, [5][6][7][8][9][10][11][12][13][14][15][16]1985       The calculation of the GPCC tioned gridded data sets. We input data and not station d spent a lot of efforts to re we think an interpolation of more error-prone than the ca Additionally, station data s relocation, opening and clos The calculation of the GP First, the SPI-DWD and SP on the grids where possibl computed using the mean o where both indices are valid. that is still possible to be cal Although the DWD-adap Bissolli, 2011) was taken int be computed in very arid reg gamma distribution to descri        and 2). Even if other parameterizations for the computation       5.1 How to calculate the parameters for the SPI-DWD