APHRODITE: Asian Precipitation - Highly-Resolved Observational Data Integration Towards Evaluation of Water Resources

During the APHRODITE-2 project term (June 2016 – March 2019), by the request of international research community, we updated (extended) the period by using the same algorithm that we adopted at V1101. However, some data were added (see Fig.2) and quality control (QC) was improved, so that we decide to have one year of overlap (2007), and run the V1101EX-R1 algorithm thru 2015.

We create APHRODITE products in collaboration with many researchers and cooperative institutions. Using a large number of input data that are not available from publicly available data (such as Global Summary of Day, by NOAA, shown in blue dots in Fig. 1) is one of the biggest advantages of APHRODITE products. In most cases, such off-line data (shown in red dots in Fig.1) that APHRODITE (and/or Dr. Yatagai) obtained by project/personal connection have restrictions that we cannot pass the original data to a third party. Therefore, we cannot publish the original station data nor the 0.05 degree daily grid precipitation/temperature data, although we interpolate them on 0.05 degree grid interval. This is the reason that we release our data in 0.25 degree and 0.50 degree except for Japan. However, for the data providers, we pass the original 0.05 degree daily grid precipitation analysis over the country as a feedback (details are written at our FAQ http://aphrodite.st.hirosaki-u.ac.jp/faq_en.html) .

 

3. Algorithm and Improvements of precipitation products

3.1) Algorithm

Figure 3 depicts the algorithm flow of producing APHRODITE precipitation data. We create 0.05 degree climatology in Task3, and with this climatology we take daily ratio to the climatology of the relevant grid cell. Interpolation itself is an inverse-distance weighting which is similar to Shepard (1968) and Willmott et al. (1985), however we take into consideration of a ridge between the target grid and each observation. Details are written in Yatagai et al. (2012). The critical issues to create better products are (i) collect (use) as many rain-gauge observations as possible and (ii) improve climatology.  So, continuous efforts of collecting rain-gauges have been maintained. For the end of APHRODITE-2 production (January 2019), the collection of offline precipitation data (year and country) is shown in Figure 2. We made V1101EX_R1 for 2007 – 2015 with the same algorithm of V1101, however, number of input data are different at the same year (2007) of both products.

The green colored squares show the steps we added and/or modified during APHRODITE-2, namely, these changes have been applied to the products V1801R1 and V1901 from V1101. Details will be explained at next section. We released pink-colored products (0.05 degree daily climatology, 0.25/0.5 degree daily analysis) for Monsoon Asia. The blue-colored 0.05 degree daily precipitation data is only released for the Japan analysis.

3.2) Difference from the previous products

It is a trade-off to use as many data as possible or to use consistent data. By adopting the “ratio” interpolation scheme, seasonal precipitation amount can be adjusted even in data-sparse areas or years by the aid of using “climatology”, however, extreme events and their long-term trends cannot be adjusted by such a scheme (Kamiguchi et al., 2010). For APHODITE-1, our policy was to use as many data as possible, and to use the same scheme throughout the time period and regions. We used all GSOD data for which a flag indicates 24-hr accumulation in spite of unknown accumulation time (which is end of the day time). GTS-based GSOD data sometimes includes erroneous “0 mm” precipitation in spite of no observation (missing) at the day, which results in significant underestimate of precipitation.

On the contrary, for APHRODITE-2, we carefully decide whether to use GSOD data or not. At first, if we have enough other data from a country for a year, we did not use GSOD for the country and at the year. As shown in Fig. 2, if there are offline data, we did not use GSOD. In addition, we checked end of the day of each obtained data (both GSOD and offline data). Hence, the number of data we input are different in the products (V1101, V1801R1 and V1901.

After checking “end of the day” (EOD) information of each input file, we do not mix different EOD dataset at a country/year for V1801R1. While, based on EOD information and using satellite estimate precipitation, we adjusted precipitation value to the universal 24-hr (00UTC – 24 UTC) period of time, which means EOD=24 for V1901 product.

The purpose of APHRODITE-2 was to evaluate extreme events and to contribute meteorological analyses including improving forecasts. To achieve this goal, we have improved our algorithm for the three points;

i) improving quality control,

ii) Careful treatment of “end of the day (EOD)” difference, and

iii) Modifying the interpolation algorithm to represent extreme values and station value conservation (SVC)

 We explain these points as follows.

3.3) Quality Control (QC)

We are using an automatic quality control program (Hamada et al., 2011), which flags suspicious values. If a large value exceeds "country record" of the country, the data is automatically rejected. However, there is sometimes record-breaking heavy rainfall, and sometimes the "country record" information of the historical maximum is not available. In order to distinguish the real extremes with errors, we compared the large amount values with satellite estimates. For V1801R1 and V1901, we used two satellite daily precipitation products (CMORPH, TRM3B42) to check large precipitation values and isolated extreme values, and to judge whether they are really extreme or not. Then, we rescued real extremes to the white list and discarded false extremes in the blacklist.

3.4) End of the Day (EOD)

In some countries, precipitation is observed hourly or every 10 minutes, and in other countries, precipitation is measured in the morning or at a regulated time manually. For example, in India, "daily precipitation" is observed in the morning (8:30 local time, LT), and stamped on that day. 8:30LT corresponds to 03 UTC (universal time). Hence, if station data shows 10 mm/day on August 31st, then, 10 mm rainfall is observed from 03 UTC of August 30th to 03 UTC of August 31st. We call this record as "EOD = 03 UTC". In V1101 and V1801 products, daily grid precipitation data over India ends at 03 UTC of recorded time.

In Japan, the AMeDAS network is based on automatic measurements at 10-minute intervals, and we can accumulate hourly and/or daily values at any interval from the 10-minute observations. In APHRO_JP, we made gridded precipitation according to Japan standard time (JST), which is from 15 UTC of previous day to 15 UTC of the stamped date. In this case, EOD=15. Further, Japan Meteorological Agency (JMA) sends either 6-hrly, 12-hrly or 24-hrly precipitation to GTS network according to the WMO's regulation. (e.g. Synoptic stations (relatively big stations) must send previous 24-hr rainfall at 00 UTC). This leads to different EOD data archived at GSOD database, although it contains flags of their processing. We use only AMeDAS for APHRO_JP, but we used both AMeDAS and GTS based data for APHRO_MA_V1101. While in V1801R1, we checked EOD of each data file and we carefully select data to use. We decide not to use different EOD data in a domain for V1801R1. So, the number of data we used in V1801R1 is sometimes less than that of V1101. The detailed procedure to estimate EOD using a satellite (CMORPH) hourly precipitation product is described at Yatagai et al. (2020, revised, to Atmosphere.)

3.5) Station Value Conservation (SVC)

When we interpolate station values (or the ratio of the station value to climatology), we take several to several decades worth of station data around the target grid into consideration. Hence, if extremely heavy and localized precipitation occurs at or very near to the grid box, using many surrounding station data may cause underestimate/overestimate of the target grid-box precipitation. In V1801, after interpolation of ratio to climatology and after getting grid precipitation values by multiplying ratio to climatology, we put the original station value to the 0.05 degree grid box, if the grid box has a station. That means, the station value is conserved (SVC) at each 0.05 grid box. If there are two or more stations in a grid box, we average them and put the averaged value into the grid box. In the same manner, if the grid box has no rainfall, but if surrounding stations have rainfall, V1101 showed some precipitation. But V1801 and V1901 conserve the station value at 0.05 degree (intermediate) file, we can get statistics of "0 mm" at the grid box. For the publicly released data, we regrid 0.05 grid to 0.25/0.5 degree, and the station value is not conserved (see Figures 1 and 2 of V1801R1 readme, http://aphrodite.st.hirosaki-u.ac.jp/product_readme/V1801R1_readme.pdf).

3.6) New climatology

The APHRODITE product is calculated by ratio (station value/grid climatology) interpolation. Climatology of previous products are based on monthly analysis (Yatagai et al., 2012). Now that our project has collected dense observation data, new daily climatology is developed from the APHRODITE daily precipitation product.

First, 30-year (1981-2010) daily precipitation product is calculated using previous climatology (WorldClim + Monthly station observation), without SVC. EOD adjustment is not applied, but an unmixed EOD source is used for each country. The 21-day running mean of daily average for 30 years is defined as new climatology. Data onFebruary 29th of non-leap year is prepared by the average of February 28th and March 1st and the 30-year average is carried out for that day.

We released this climatology in V1901.

3.7) Rain-gauge availability information (RSTN)

It is very important to know whether each grid-box value is just interpolation or reflects an observed value within the grid box or nearby. Hence, we put “number of observations” information for each day on each grid at 0.05 degree product such as APHRO_JP and the feedback versions given to data providers. When we apply regridding at Task 5 (see, Fig.3), we count the grids with at least one rain-gauge observation. When we create the 0.25 (0.50) degree products, each grid box has 25 (100) grids of 0.05 degree. If there is one grid with 25 grids, we count “ration of station grids (RSTN)”, and put this information with the product day by day. For 0.25 degree gridded products, RSTN=4(%) means there is one grid with a station, and RSTN=8(%) means there are two grids with a station.

Hence, users must check the value of RSTN, and use only grids where RSTN is not 0.0, especially when APHRODITE ais used as a reference to validate satellite estimates and/or model simulate precipitation. If they want to evaluate trends and/or long-term change of precipitation amount, they must check the time series of RSTN in their target area.

We are frequently asked “Can we use APHRODITE for studying extreme events?”. The answer is, it depends on your purpose. Although we applied SVC for V1801R1 and V1901 so that the APHRODITE team can evaluate changes of extreme value with 0.05 degree products, \ we only release regridded (0.25/0.5) products so that general users cannot discuss the precise change based on the value which has the same accuracy with the station values after quality control. In any case, the user must check the change of RSTN. Please also check our FAQ page (http://aphrodite.st.hirosaki-u.ac.jp/faq_en.html).

As for V1901, we adjusted EOD before interpolation. At some areas where the EOD of original data are either 0 or 24, we have no change at each value, however, the case of original data EOD is not 0 or 24, users cannot treat our gridded value as equivalent to observed precipitation.     

 

4. Algorithm and Improvements of temperature products

AphroTemp_V1808 is a daily mean temperature product at 0.50 and 0.25 degrees. More data is collected and analyzed than AphroTemp_V1204R1 and the data period is also extended to 2015. A topographical effect scheme is introduced to interpolation algorithm.

Surface observations in India, Bhutan, Thailand, Myanmar are added to the grid analysis after AphroTemp_V1204R1. Due to the increased number of observations of and improvement of interpolation algorithm, climatology and estimated values in South Asia are well represented in the new product.

4.1) Algorithm and important information on RSTN

For temperature, the products we release are 0.5x0.5-degree and 0.25x0.25-degree gridded daily mean temperature data over Monsoon Asia (APHRO_TAVE_MA_V1808).

The gridded fields of daily mean temperature are defined by interpolating gauge observations obtained from meteorological stations throughout the region. The algorithm for the new product is made with a few modifications on the precipitation algorithm V1101 (Yatagai et al., 2012). Figure 4 shows the flowchart for the temperature algorithm. Instead of ratio interpolation as for precipitation, we interpolate anomalies at the target grid.  We interpolate the subtraction of the daily mean temperature from the climatology at a resolution of 0.05 degrees, then add each gridded anomaly to each gridded climatology value day-by-day. We then re-gridded the 0.05-degree analysis to both 0.5-degree and 0.25-degree grids. Details are given in Yasutomi et al. (2011).

An indicator is introduced to represent the reliability of the interpolated daily temperature fields.  This indicator, named RSTN, is also calculated for each re-gridded 0.50-degree (0.25-degree) cell as the proportion of 0.05-degree cell(s) containing station(s).   The number of input data differs from year to year. Users interested in long-term changes should consider this indicator.

4.2) Points of concern

i) Lack of observation data (in Indonesia and Papua New Guinea) 

Daily observations over Indonesia and Papua New Guinea are sparse. A considerable number of temperature grids over New Guinea Island are missing until the 1980's. The distribution of valid station observations is provided (RSTN) for each day. Users should pay attention to the density of observations.

ii) Homogenization of observations

We do not homogenize the observed time series of temperature data. Changes in gauges, location of the stations, and many other factors might cause discontinuity of observational data.

iii) Improvement in calculation after Yasutomi et al.(2011)

The anomaly interpolation method is introduced in Yasutomi et al. (2011). The climatological mean temperature is based on AphroClim_V1808. First, we derive the difference from climatology, next, anomalies are interpolated into 0.05 degree grid. Then anomalies are added to the climatology and regridded into 0.25 degree and 0.50 degree grid.  We released the 0.05 degree climatology.

 

5. Current Developments (APHRODITE-3) and Future Plans

The APHRODITE team at Hirosaki University (led by Prof. Yatagai) received support from the university for 2019-2022 to continue APHRODITE activity and improve its algorithm for cold regions. As a part of APHRODITE-3 activity, we had a workshop by inviting more than 20 researchers from Russia, the Central Asia and United States. Based on the data we obtained recently, we will make update versions of APHRO_RU and APHRO_ME.

However, for EOD adjustment, satellite precipitation estimates over high latitudes/altitudes are challenging, especially for the snow-covered land surface. We also try to use meteorological reanalysis products to compensate for the problem. As a university located in northern Japan, where we have plenty of snow in winter, we are applying some adjustment method to get quantitative solid precipitation. A test product to apply wind-induced undercatch (Masuda et al., 2019) is also released at our web-site.

There are many trade-offs between keeping original values to reserve extreme values and adjusting original data to quantified value. As shown in Table 1, we already have various versions of the products. We are planning to make a tool for subsetting and downloading the data that users choose. In order to contact every user when the APHRODITE team makes updates or discovers errors, we ask every user to register at our site. Users are also strongly requested to pay attention to the distribution and change of the parameter RSTN. ##