Accurate, timely, unbiased, consistent gauge-enhanced satellite precipitation can help understand and manage agricultural and hydrologic risks. Unfortunately, the number of global gauge observations is very low, covering an area of less than a single football field or soccer pitch (Kidd et al. 2017). Globally, the number of gauges1 has declined between 1981 and 2021, from about 32,000 to 14,000, while in Africa, the corresponding figures are just 3,400 and 600. These declines are caused by a combination of data policy and decaying observation and transmission systems. Precipitation thus remains a leading source of uncertainty for agroclimatic risk assessments and related agricultural modeling applications (Ruane et al. 2021). Gridded satellite rainfall datasets, therefore, provide a key resource for understanding weather, providing effective early warning, and assessing changes in extremes in a warming world (Kirschbaum et al. 2017). There are three primary categories of satellites used to study and quantify precipitation from space. The oldest category dates back to the Television and Infrared Observation Satellite (TIROS), which launched in 1960 and hosted imagers capable of capturing visible and infrared wavelength Earth observations (EOs). Modern descendants of TIROS occupy geostationary orbits over the equator. Circling in step with Earth’s rotation, they provide continuous information about cloud-top temperatures (Janowiak et al. 2001; Knapp et al. 2011). These geostationary infrared (GEO-IR) observations, however, do not directly observe precipitation, but rather, just the highest layer of clouds and/or the land surface. Passive microwave observations (PMW), on the other hand, can observe hydrometeors, supporting the retrieval of rainfall intensities using techniques like the Goddard profiling algorithm(s) (GPROF) (Kummerow et al. 1996, 2001, 2015). Seminal combined estimation approaches merge PMW and GEO-IR estimates using weights determined by standard errors (Huffman et al. 1997; Xie and Arkin 1997), use sophisticated artificial neural nets to translate GEO-IR observations into precipitation (Hsu et al. 1997), or use collocated PMW and GEO-IR observations to tune local GEO-IR estimates based upon PMW observations (Huffman et al. 2007; Turk et al. 2003, 2010). The neural net approach can use cloud classifications, as in the Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks Cloud Classification System (PERSIANN-CCS) (Hong et al. 2004).
A third category of satellite—characterized by the 1997–2015 Tropical Rainfall Measuring Mission (TRMM) (Kummerow et al. 1998, 2000) and the 2014–present Global Precipitation Measurement mission Core Observatory (GPM-CO) (Hou et al. 2014; Skofronick-Jackson et al. 2017)—provided single platforms that contained multiple sensors explicitly designed to monitor precipitation. TRMM had a Ku-band radar and a passive microwave multifrequency imaging radiometer; these instruments were capable of producing three-dimensional views of precipitation events (Kozu 2001). The GPM-CO is even more sophisticated, with additional channels on the microwave imager and a dual-frequency precipitation radar (Hou et al. 2014; Skofronick-Jackson et al. 2017).
Building on techniques used to develop the experimental multisatellite TMPA and TMPA-RT products (Huffman et al. 2018, 2020) (Huffman et al. 2007), the U.S. GPM team now combines information from the GPM-CO, PMW constellation, and GEO-IR-based PERSIANN-CCS to produce the IMERG Early Run and IMERG Late Run (IMERGlate) datasets with latencies of about four and fourteen hours (Huffman et al. 2018, 2020). The PMW constellation2 includes several types of radiometers: the Advanced Microwave Scanning Radiometer-2 (AMSR-2) on JAXA’s Global Change Observation Mission–Water (GCOM-W1), the Advanced Technology Microwave Sounder (ATMS) on the Suomi NPP and NOAA-20 polar orbiters, the multichannel Sondeur Atmospherique du Profil d’Humidite Intertropicale par Radiometrie (SAPHIR) on the Megha-Tropiques satellite, the Microwave Humidity Sounders (MHS) on the U.S. NOAA-19 and European MetOp platforms, and the Special Sensor Microwave Imager/Sounder (SSMIS) instruments on U.S. Defense Meteorological Satellite Program (DMSP) satellites. These PMW observations are assembled and converted to estimates of precipitation intensity using GPROF. These high-quality, half-hourly PMW estimates are then used to adjust the GEO-IR-based PERSIANN-CCS. Finally, a semi-Lagrangian time-interpolation scheme, adapted from the CMORPH–Kalman filter (Joyce and Xie 2011), is used to approximate the temporal evolution of the PMW-observed precipitation, based on motion vectors estimated from the GEO-IR. This process is repeated to produce the IMERGearly and IMERGlate products, with the latter being of higher quality due to it using more data and a time interpolation of data, rather than the forward-only propagation calculated in IMERGearly. After a two month latency, an interpolated monthly station data analysis provided by the Global Precipitation Climatology Centre (GPCC) is combined with IMERGlate to create an IMERGfinal product. This blending procedure is similar to that used in the TRMM Multisatellite Precipitation Analysis (Huffman et al. 2007) and was originally developed for the Global Precipitation Climatology Project (Huffman et al. 1997). The satellite-gauge blending process is achieved through 1) large-scale bias correction of the satellite estimates, followed by 2) blending with interpolated gauge fields. The blending process is guided by weights proportional to the inverse of the associated error variances—this process gives greater weight to the satellite estimates in data-sparse regions. This blending produces the IMERGfinal.
Here, we describe and evaluate a bias-adjusted version of IMERGlate [Climate Hazards Center IMERG (CHIME)] and a station-enhanced version of CHIME [CHIME with Stations (CHIMES)]. CHIMES enhances the IMERGlate product using a high-resolution climatology and low-latency rain gauge observations, resulting in a product that is similar to IMERGfinal, but with a lower latency. The value of this bias-correction procedure arises from the strong performance of the CHC’s approach to developing high-resolution climatologies. The core insight guiding this work is that long-term averages of satellite QPE provide excellent inputs into the production of high-resolution climatologies (Funk et al. 2015a). The spatial correlation between in situ precipitation normals and mean QPE fields is large and consistent (∼>0.6), while the relationship with elevation is local, complicated, and typically weaker (Fig. 1 in Funk et al. 2015a). Furthermore, many gauge-based climatologies use complex mathematical fits, such as thin-plate splines, that can perform poorly in data-sparse areas, producing complicated orographic features that are not well supported by the available data. Other climatologies, like those underlying the Global Precipitation Climatology Centre’s interpolated gauge product, are based on interpolated climate normals—which may miss important details because of station limitations.
(a) The CHIMES development process. (b) CHIME(S) products. (c) Map of the wettest three months used in the validation study; numerals correspond to the first month. (d) REGEN (Contractor et al. 2020) validation locations and regions. High-quality interpolated gauge data in these regions are compared with satellite-only estimates.
Citation: Bulletin of the American Meteorological Society 103, 2; 10.1175/BAMS-D-20-0245.1
To support early warning efforts like the Famine Early Warning Systems Network (FEWS NET) (Funk et al. 2019) and hydrologic modeling efforts like the FEWS NET Land Data Assimilation System (FLDAS), we have developed an update to the Climate Hazards Center Precipitation Climatology. As shown in (Funk et al. 2015a), this approach performs substantially better than the GPCC climatology and other competing products. Focusing on Ethiopia (Figs. 10–12 in Funk et al. 2015a), we see that the GPCC captures the general shape of mean Ethiopian rainfall, but still exhibits substantial biases. A sample west-to-east transect at 7°N shows that the GPCC captures the overall gradient reasonably well, but misses local maxima associated with the Rift Valley escarpments at 37° and 40°E. Other global assessments have confirmed the CHC climatology’s relatively strong performance (Beck et al. 2017). Just as Funk et al. (2015a) suggested improvements over the GPCC mean fields, the results presented here suggest that the IMERGlate’s mean performance can be improved by use of an updated CHC climatology.
We conclude our introduction with a few concrete examples of how these data might be used in the context of food insecurity. Since 1999 (Verdin et al. 1999), FEWS NET and the CHC have focused on developing “climate science for early warning” (Verdin et al. 2005). Today, blended gauge-satellite QPE can be combined with forecasts to provide “staged” (Funk and Shukla 2020; Shukla et al. 2021) early warning systems that combine climate, weather, land surface models, and satellite observations to provide multiple opportunities to anticipate and monitor hydroclimatic extremes. These efforts contribute to FEWS NET’s sophisticated Food Security Outlook process (Magadzire et al. 2017), reports and alerts produced by the Group on Earth Observations Global Agricultural Monitoring Initiative (GEOGLAM) Crop Monitor project (http://www.cropmonitor.org/), the capacity-building and decision-support activities of NASA SERVIR (https://www.nasa.gov/mission_pages/servir/index.html), and NASA hydrologic modeling and forecasting activities, based on the FLDAS (McNally et al. 2017) and NASA Hydrological Forecast and Analysis System (NHyFAS; Arsenault et al. 2020).
While these efforts are already leveraging satellite observations to help guide billions of dollars of humanitarian assistance that help provide aid to millions of people every month, more needs to be done to address rapid increases in hydroclimatic risks. The human and economic costs of hydroclimatic shocks are massive and increasing. Between 2015 and 2019, the impacts and losses associated with droughts, floods, and hurricanes surged through the interaction of increasing exposure and more extreme weather and climate (Funk 2021). In 2020, the Aon-Benfield Reinsurance Company (2021) reported that cyclones, floods, and droughts resulted in $165 billion in losses. And between 2015 and 2021, the number of extremely food-insecure people has tripled; in 2021, more than 147 million people, one out of every 55 people, requires emergency food assistance, as the world faces a severe food security crisis (UN Food Security Information Network 2021). While conflict, food price inflation, and COVID-19 largely fuel this crisis, both droughts and extreme precipitation/flooding have played a substantial role. Improved QPE products, like CHIMES, will help us monitor, mitigate, and manage these increased hazards.
An overview of the CHIMES data products and process
While no dataset is perfect, authoritative data are trustworthy and capable of supporting important decisions that can save lives and livelihoods and trigger positive financial and societal responses to hazards. Authoritative data are actionable. Surveys of the user community (Kirschbaum 2020) indicate that continuity and gauge enhancement for IMERGearly and IMERGlate are key interests for the IMERG user community. The incorporation of timely gauge data can increase the utility and accuracy of gridded datasets. Instead of two often-quite different sources of information (gauge and satellite), users are provided a single harmonized data product. CHIMES builds on the techniques and resources used to produce the widely used CHC Infrared Precipitation with Stations (CHIRPS2) product (Funk et al. 2015b). The satellite-only component of CHIRPS2 (CHIRP2) is based solely on GEO-IR observations, while IMERGlate also draws on the PMW and GEO-CO constellation. As we will demonstrate, IMERGlate can be substantially improved by the bias adjustment and gauge enhancement provided by CHIMES. As our validation results indicate, this will provide a product that performs better than CHIRP2, but with a shorter period of record, beginning in 2001 rather than 1981. While IMERG begins in mid-2000, CHIMES starts in 2001 to provide complete years. Performance will be evaluated via comparison with a completely independent set of high-gridded quality gauge observations (Contractor et al. 2020), and quantified via correlation, mean bias error, ratios of standard deviations, mean absolute errors, and dry spell hit ratios.
The basic objective of CHIMES is to enhance IMERGlate using in situ station observations (Fig. 1a). This enhancement results in three products: the climatologically adjusted daily satellite-only adjusted IMERGlate (CHIME), the moderate-quality station-adjusted submonthly (pentadal) CHIMESprelim, and the high-quality station-adjusted monthly CHIMESfinal. We use “S” to denote station enhancement.
Supporting CHIMES is an extensive station archive developed and curated by the CHC. Table 1 lists the data sources used in the CHC Station Climatology Database (CSCD), and appendix A describes the CHC’s extensive quality control and curation processes. Appendix B describes the CHIMES procedures mathematically. CHIMES benefits from three categories of station data. Used judiciously, information from these sources can “stack” in an additive manner (Fig. 1b). Numerous climate normals are used to generate high-resolution mean fields, which are used to produce unbiased daily CHIME fields. Then, every “pentad,”3 CHIMESprelim estimates blend CHIME pentads with relatively sparse, but low-latency, submonthly gauge data. The pentadal station observations are quality controlled using automated procedures. Because of the difficulty in identifying problematic daily station data, interpolation is not performed at the daily time scale, but daily CHIMESprelim fields are produced via disaggregation based on the satellite-only daily CHIME. In the last phase, after the careful manual curation (appendix A), the sum of the previous months’ CHIMESprelim pentads are blended with in situ observations to produce CHIMESfinal. Proportional disaggregation is then used to produce final pentad and daily CHIMES values.
CHC rain gauge archive contributions by source, along with approximate numbers of stations.
While involved, this type of staged process can support effective early warning and early action, as rapid initial estimates (daily CHIME) are refined using available submonthly station observations (CHIMEprelim pentads). Then, carefully screened monthly gauge totals are used to produce an authoritative but timely “final” product. Analysts and modelers have access to interoperable information sources at a range of latencies that extends from 2 to 18 days, with performance at the “final” stage similar to other high-quality monthly gauge-enhanced products (Becker et al. 2013).
The updated CHC Precipitation Climatology (CHPclim)
As periods of accumulation and latencies decrease, going from long-term means, to months, to pentads, the number of available stations declines. Hence, there is more information at the long-term mean time scale, and complex geospatial modeling techniques (Funk et al. 2015a) can be used to build high-resolution (0.05°) long-term mean fields (information source 1 in Fig. 1a). We refer to these climatologies as the “Climate Hazards Precipitation Climatology,” or CHPclim. As described in appendix B, monthly background mean fields are created using local regressions trained to the previous CHC climatology (Funk et al. 2015a). IMERGfinal and elevation data are used as predictors. If b0, b1, and b2 represent the local regression coefficients, the estimate at a grid cell will be Yest = b0 + b1IMERGfinal + b2Elevation. These Yest fields are then adjusted using monthly gauge-undercatch-corrected precipitation gauge averages. In the next step, 1981–2010 CSCD averages are calculated and combined with GPCC observations (Schneider et al. 2013), with preference given to the CSCD values. As described in appendix B, a modified inverse distance interpolator is used to blend the station observations and Yest fields. The 12 monthly climatologies are disaggregated to daily and pentadal means by 1) creating a padded 14-month 427 array of days, 2) assigning each day the associated monthly value, and finally 3) running a smoothing filter many times to produce a smoothly varying annual cycle.
Ironically, the most rapidly produced product, the satellite-only daily CHIME, is adjusted using the slowest source of station information. Long-term climate normals are incorporated into CHPclim, which is then used to bias adjust the IMERGlate (information source 2, Fig. 1a). As we shall see, this substantially improves performance, as measured by mean bias errors, mean absolute errors, ratios of standard deviations, and hit ratios. This type of bias adjustment also improves the alignment with station observations, especially in data-sparse areas like East Africa (Dinku et al. 2018). Reducing systematic differences between gauge observations and satellite estimates can reduce inhomogeneities as station networks shift over time.
The CHC station archive and blending process
During the “preliminary” and “final” modeling steps, pentadal, and monthly station observations are blended with either the CHIME or CHIMESprelim (information source 3, Fig. 1a, appendix B). Numerous international and national station data sources (Table 1, described in more detail in appendix A) are combined and quality controlled to produce sets of in situ pentad and monthly totals.
Central to both the CHC quality control and pentadal/monthly interpolation procedures is the development of a set of well-organized “anchor stations”—a superset of all the possible gauge observation locations. Multiple data sources (Table 1) are then used to produce as-complete-as-possible time series, with higher-quality data sources given preference. These time series support the identification of improbable outliers. The pentadal quality control (QC) process is automatic. The final monthly station archive is also carefully checked, visually and analytically, by a team at the CHC. The pentadal station observations are obtained from rapidly updated automated station archives, with a 2-day latency. At the monthly time step, many additional gauge observations are contributed by numerous partner agencies, including national meteorological agencies. These contributions often dramatically enhance national coverage.
Pentads, as opposed to days, were chosen as a preferred accumulation period for several reasons. One pragmatic constraint is uncertainty about the actual reporting time associated with many data sources. Another issue involves quality control. Statistical error detection typically becomes easier over longer time periods. The spatial coherence of precipitation fields increases with accumulation period, increasing the accuracy of spatial interpolation. And finally, many early warning practitioners, especially those interested in precipitation deficits (i.e., not floods) find that a 5–10-day submonthly accumulation period is an appropriate time scale (Ross et al. 2009).
This process uses IMERGfinal to generate spatial distance decay functions for every pixel and time period. The spatial coherence of IMERGfinal informs the spatial influence of station observations. For each pixel, decay functions, neighbors, and neighbor distances are precomputed, supporting rapid global interpolation speeds.
The CHIMES process (Fig. 1a) draws from four information sources: 1) dense sets of climate normals, 2) the GPM constellation/IMERGlate, 3) in situ station archives, and 4) IMERGfinal’s spatial coherence. Appendix B provides a mathematical description of the CHIMES algorithms.
Evaluating the performance of IMERGlate and CHIME
We next present the first of two brief evaluation studies. The first study focuses on the performance of the satellite-only IMERGlate and CHIME products. The second study uses cross validation to assess the contribution provided by station enhancement. While a detailed multiproduct assessment is beyond the scope of this brief paper, we also include three other representative satellite-only products: CHIRP2, PERSIANN-CCS (Hong et al. 2004), and the GOES precipitation index (GPI) (Arkin and Meisner 1987). The GPI and CHIRP2 are both based on cold cloud duration (CCD) values calculated from the Climate Prediction Center GEO-IR archive (Janowiak et al. 2001) using a fixed 235-K temperature threshold. The 235-K threshold represents the expected temperature of cloud-top temperatures, so CCD is related to deep convection, especially in tropical and subtropical monsoon regions. The GPI estimates are calculated by multiplying the CCD totals by 3 mm h−1.
Validations are carried out using the high-quality daily 1° “Rainfall Estimates on a Gridded Network” (REGEN; or rain in German) (Contractor et al. 2020), which was obtained from the Frequent Rainfall Observations on Grids (FROGS) database (Roca et al. 2019). To facilitate global analyses, we define the wettest three months for each location (Fig. 1c), and focus our evaluations on the wettest three months and the 18 pentads within those months. Figures 2–5 show results based on seasonal accumulations. For each pixel, a fixed wettest 3-month period was identified (Fig. 1c), and then 16 years of 3-month total precipitation was calculated for the interpolated station (REGEN) and five satellite-only datasets. The five satellite-only datasets do not incorporate time-varying gauge observations. Table 2 shows results based on the 16 years × 18 pentadal accumulations for each pixel’s wet season. REGEN ends in 2016, hence this study uses a 2001–16 evaluation period, which is the longest whole-year period containing complete data for all sources.
Correlations for the wettest three months between REGEN and satellite-based estimates. Cells with dense station support are noted with small black dots. These 1° grid cells have at least, on average, 7 or more observations per day. Cells with limited station support are noted with small gray dots. These 1° grid cells have at least, on average, less than 1 observation per day.
Citation: Bulletin of the American Meteorological Society 103, 2; 10.1175/BAMS-D-20-0245.1
Ratios of standard deviations (σsatellite/σREGEN) for the wettest three months. Cells with dense or limited station support are noted with small black or gray dots.
Citation: Bulletin of the American Meteorological Society 103, 2; 10.1175/BAMS-D-20-0245.1
Ratios of means (μsatellite/μREGEN) for the wettest three months. Cells with dense or limited station support are noted with small black or gray dots.
Citation: Bulletin of the American Meteorological Society 103, 2; 10.1175/BAMS-D-20-0245.1
Maps of mean absolute errors (MAE) for the wettest three months. Units are millimeters over the 3-month accumulation period. Cells with dense or limited station support are noted with small black or gray dots. MAE expressed as ratio of long term REGEN average precipitation.
Citation: Bulletin of the American Meteorological Society 103, 2; 10.1175/BAMS-D-20-0245.1
Validation results based on comparison of pentadal satellite-based precipitation estimates with gridded REGEN rain gauge observations. Results averaged over well-gauged 1° grid cells during the climatologically wettest three months. Validation seasons and regions are shown in Fig. 1d. CenAM is Central America.
We use the number of average daily observation stations in each 1° grid cell as a metric of data certainty. Cells with averages of at least seven daily observations are identified with blue circles in the bottom panel of Fig. 1c. The summary statistics presented in Table 2 are based on the average of results for these well-instrumented locations. To facilitate visual interpretations, these cells are also noted in Figs. 2–5. Poorly monitored locations, with less than an average of one station observation per day, are identified with gray circles.
Correlation results.
Figure 2 shows 2001–16 wet season correlation maps focused on the four best-instrumented regions. The performance of the CHIME/IMERGlate estimates is compelling. Over large portions of North America, Europe, Iran, and Australia, seasonal correlations exceed 0.8, and most areas have correlations of at least 0.6. Information from GPM-CO and PMW instruments is clearly enhancing performance, since correlations for GEO-IR-based CHIRP2/GPI and the PERSIANN-CCS are substantially lower and less consistent. The PERSIANN-CCS correlations are the lowest of the five products, with typical seasonal correlations of less than 0.6. The pentadal correlation results, shown in Table 2, present a similar tableau. Overall, the CHIME/IMERG pentadal correlations were high (0.78 on average) and consistently above the CHIRP/GPI and PERSIAN-CCS. The ability of IMERGlate to track wet season precipitation from space with such a high degree of fidelity bodes well for applications utilizing IMERGlate and CHIME.
Ratio of standard deviations results.
Figure 3 shows maps displaying 2001–16 ratios of interseasonal standard deviations (satellite/REGEN). In North America, Europe, and Iran, the performance of CHIME is close to parity. In these regions, the IMERG-to-CHIME bias correction brings the variance structure into closer alignment with REGEN. In central Australia, CHIME may underestimate the variance, but this extremely dry region is also poorly instrumented.
In terms of variance, CHIRP2 performs poorly, substantially underestimating variability. The CHC team has tracked this problem back to the intercept terms used to calculate CHIRP2 pentads. PERSIANN-CCS and GPI perform better than CHIRP2, but both have areas where the variance of the satellite precipitation is much greater than the observations, particularly in the western United States and central and northern Iran.
Ratio of means results.
Figure 4 shows maps displaying 2001–16 ratios of seasonal wet season means. The CHIRP2 and CHIME means are very close to REGEN averages. This fidelity enhances their accuracy and utility, especially for applications sensitive to the absolute quantity of precipitation. For example, groups like FEWS NET rely on satellite rainfall grids to estimate the start of crop growing seasons, and estimate “crop water satisfaction” (Senay and Verdin 2003; Verdin and Klaver 2002). In semiarid areas, a small bias (say ±5 mm per pentad), can add up to a large bias when accumulated over a 3- or 4-month crop-growing season. These biases can also lead to inaccurate growing season start estimates. Hydrologic modeling applications like FLDAS (McNally et al. 2017), face similar challenges. Relatively small precipitation biases can accumulate within these models’ soil moisture reservoirs. IMERGlate exhibits a modest tendency to overestimate, except for the northwestern United States and portions of Iran and western Turkey. The GPI and PERSIANN-CCS overestimate rainfall in parts of the western United States and Iran. The PERSIANN-CCS and GPI bias patterns are quite similar.
Ratio of mean absolute error to observed mean results.
Figure 5 shows maps displaying 2001–16 seasonal mean absolute errors (MAE) expressed as ratios of the average REGEN wet season precipitation. Pentadal results are listed in Table 2. MAE will be influenced by both systematic and random errors (Willmott et al. 1985). These results, therefore, reflect both the strength of seasonal correlations (Fig. 2) and systematic distribution biases (Figs. 3 and 4). The CHIME performance is consistently good across all four regions, with typical percent MAE values of less than 30%. The IMERG and CHIRP2 appear to be the next-best performers, though probably for different reasons, with the CHIRP2 benefitting from strong mean performance and moderate correlation performance, and the IMERG benefitting from excellent correlation performance and moderate mean performance. The GPI and PERSIANN-CCS’s large mean biases translate into very large MAE values (Fig. 5). Capturing the mean precipitation structure is an important aspect of having overall low MAE values.
The pentadal MAE statistics (Table 2) indicate global percent errors ranging from 48% (CHIME) to 75% (PERSIANN-CCS, GPI) of the long-term mean, with IMERGlate and CHIRP2 falling in the middle of this range (60% and 64%). Unbiasing reduces the overall IMERG errors by ∼20%. Regionally, the IMERGlate to CHIME adjustment seems most beneficial in mid-latitude regions. The U.S. and European validation sites saw the largest improvements, as MAE values were reduced by 20%–30%.
One striking feature of the CHIME is its robust performance across all the regions. All products tended to have the highest MAE values in the two western U.S. regions, southern Europe, and Iran. CHIME performance was substantially better in these regions, due to the bias-correction procedure. The CHIME percent MAE errors only exceeded 50% in one region (Iran). The other products exceeded this error level in 8–12 regions.
Dry spell analysis.
While a detailed analysis is beyond the scope of this article, to briefly explore CHIME performance vis-à-vis agricultural applications, we evaluate hit ratios based on dry events (Table 3), with dry events defined as wet-season pentads receiving less than 10 mm of precipitation. All products performed quite well, with hit rates over 75%. Satellites’ ability to observe dry conditions from space is tremendously valuable in the context of agricultural drought monitoring. IMERGlate, however, clearly performed better than CHIRP2, and CHIME performed better than IMERG. CHIRP2’s variance underestimation leads to poor dry-spell detection, especially in Brazil and Mexico. CHIME performance, on the other hand, is strong in every region, reinforcing the consistently high levels of correlation, low bias, and variance fidelity shown in Figs. 2–5 and Table 2. This analysis underscores the strong performance of IMERGlate and CHIME across a very diverse collection of geographies and climate. Please also note that the IMERGlate/CHIME data being evaluated are completely independent of the station-based REGEN archive. So this high level of agreement indicates that both of these independent data sources are accurate, which leads to a high level of agreement.
Dry pentad detection results based on comparison of pentadal satellite-based precipitation estimates with gridded REGEN rain gauge observations. Dry events were defined as pentads with less than 10 mm of precipitation. Validation seasons and regions are shown in Fig. 1d. Hit ratios are calculated as 100 × #hits/(#hits + #misses). Guatemala and Ecuador were excluded because they very rarely exhibited REGEN pentads with less than 10 mm of precipitation.
Evaluating the relative benefits of rain gauge enhancement
This section uses monthly in situ rain gauge observations and a take-one-away cross-validation procedure to assess the relative benefits of blending station data with CHIME. It is extremely important for readers to know that the density of precipitation gauge data varies dramatically from region to region. While space limitation precludes a detailed description here, resources at the CHC website support detailed visualization of time series1, national maps of stations locations,4 and global geotiffs5 that quantify station density.
For each year and month between 2001 and 2019, each monthly station observation is withheld, and at-station gauge/satellite adjustment factors are calculated at surrounding stations. In the previous section, the 2016 end point of the REGEN dataset limited us to 2016; here we can use a longer period because we use the CHC’s routinely updated station archive. A modified inverse distance weighting (IDW) interpolation is then used to estimate the weighted adjustment value at the station location [as in Eqs. (B12)–(B14) in appendix B].
This procedure allows us to evaluate the decreases in MAE and the increases in variance associated with the gauge-blending procedure. Some fraction of the total variability of the global station dataset will be explained by the original IMERGfinal archive. Unbiasing with CHPclim will then explain some additional variance. Finally, gauge-based station enhancement can explain another additional amount of variance. The remaining variance is the mean-squared error (MSE).
Figure 6 presents our cross-validation results in terms of variance explained for the entire domain (60°S–60°N, 180°–180°) and by continent.6 Three sources of information (variance explained) stack: IMERGlate, the incremental improvement due to unbiasing, and the information added when stations are blended with CHIME. The remaining variance is the MSE. Please note that no attempt was made to spatially weight these results, so densely instrumented regions will tend to dominate the global values. Overall, IMERGlate performs well, explaining 50% of the station variance. IMERGlate performance in Europe, North Asia, and Australia, however, is lower than 27%. Unbiasing in Africa, North America, Europe, and North Asia increases the variance explained, respectively, by 19%, 24%, 30%, and 33%—large improvements in performance. In Australia, South Asia, and South America, gains due to bias adjustment were minor or modest, 1%–7%. The original IMERGlate, exhibited little bias in these areas, so unbiasing adds little benefit.
Pie charts showing the relative fraction of monthly precipitation variance for each region. These results are based on at-station cross validation. The variance explained by the original IMERGlate data are shown in dark blue. The additional variance explained by the unbiased satellite-only CHIME is shown in cyan. The additional variance explained by the stations is shown in light blue. These values “stack,” such that globally, these results indicate that the CHIMES product explains 50% + 13% + 12% = 75% of the monthly in situ precipitation variability. The unexplained variance, or mean-squared error, is shown in red.
Citation: Bulletin of the American Meteorological Society 103, 2; 10.1175/BAMS-D-20-0245.1
Station enhancement adds additional information. The largest benefits occurred in Australia, North America, and Europe, with increases of 47%, 20%, and 17% variance explained. South America and North Asia saw modest 13% enhancements. Africa and South Asia exhibited limited enhancements; the results in these regions underrepresent the potential merits of enhancement, because the interstation distances are too large to well support cross-validation-based analyses. Note that spatial gauge distributions will also influence these results. For example, gauges in Australia are much more frequent near the populated wetter coastal regions, and the cross-validation procedure identifies large increases in variance, explained here (Fig. 6). The number of available monthly stations varies dramatically by country.7 In 2021, the United States had around 6,000 monthly observations, while Tanzania had 10. This heterogeneity underscores why the consistency of CHIME is important; our results suggest that CHIME is likely to perform reasonably well, even in areas without gauges. These results can also be expressed as changes in MAE (Table 4). Globally, the unbiasing and gauge blending reduces MAE values from 46 to 39 to 27 mm month−1, implying that CHIMES’s MAE values are almost half of IMERGlate MAE values, with the caveat that our “global” results tend to favor data-rich regions like Europe, Australia, and North America. While the contribution from unbiasing and station enhancement varies from region to region, CHIMES variance explained is substantially better in every case.
Cross-validated monthly MAE estimates (mm month−1), based on the CHC station data archive.
Finally, it should be noted that this evaluation approach will underestimate the value of stations in data-sparse areas. The actual value of station observations is probably greatest where they are least dense, as in Africa, but because the actual interstation distances are so great, cross-validated assessments of these benefits are less conclusive.
Discussion and conclusions
There are two major findings in the results presented here. The first finding has been that the IMERGlate product does a very good job of representing pentadal and wet season precipitation variability, both in terms of correlation and variance. The second finding, which comes as no surprise, is that high-resolution climatologies and in situ gauge observations can be used to make substantial enhancements to the satellite-only IMERGlate product.
CHIMES effectively marries two approaches to precipitation estimation—a complex, physically based estimation process (Huffman et al. 2020) driven by GPM-CO, PMW, and GEO-IR observations (Hou et al. 2014; Skofronick-Jackson et al. 2017), as interpreted by GPROF (Kummerow et al. 2015), PERSIANN-CCS (Hong et al. 2004), and the IMERG algorithm (Huffman et al. 2020)—and the CHC’s geostatistical modeling approach (Funk et al. 2015a,b), which can be seen as extension of presatellite approaches to “smart interpolation” (Willmott and Matsuura 1995; Willmott and Robeson 1995) and current efforts led by the Global Precipitation Climatology Centre (GPCC) (Becker et al. 2013; Schneider et al. 2017) and the Climatic Research Unit (CRU) of East Anglia (Harris et al. 2020). The satellite-only CHIME combines the strong temporal fidelity of IMERGlate, which ultimately originates in the ability of space-borne microwave and radar sensors to actually sense hydrometeors, with the high-spatial accuracy of CHPclim and CHIRP2. While sparse in their coverage, microwave and radar observations can be used to tune thermal infrared (TIR) estimates. These estimates are the only source of information when the surface is icy or snowy. To IMERGlate’s strong temporal fidelity, CHIME can add very low systematic bias and consistently low MAEs (Fig. 5, Table 1).
One striking attribute of the version 6 IMERGlate algorithm is very consistent correlation performance across most of the REGEN validation regions (Fig. 2). The median pentadal correlation for the 12 study regions was 0.75 (Table 2), a value substantially higher than the 0.58 median correlation for the GEO-IR-based CHIRP2 and PERSIANN-CCS. While the difference between 0.58 and 0.75 may seem small, this can represent a substantial change in variance explained. Another positive aspect of the IMERGlate performance was quite consistent estimates of seasonal (Fig. 3) and pentadal precipitation standard deviations (Table 2). While IMERGlate overestimated the variance slightly, the estimated standard deviations were within ±25% of the observed REGEN standard deviations in all of the study regions. The PERSIANN-CCS performance was also quite good, with 10 out of 12 regions within ±25%. CHIRP2 performance, by this metric, was very poor. The CHC has tracked this issue back to the use of intercept values in the pentadal CHIRP2 cold cloud duration (CCD) estimates (CHIRP2 = b0 + b1CCD). The constant b0 terms incorrectly reduce the year-to-year variance.
Despite its systematic underrepresentation of precipitation variance, CHIRP2’s performance was fairly comparable to IMERGlate (Fig. 5), largely due to CHIRP2’s strong compensating performance in terms of low bias (Fig. 4). In the United States, Europe, and Central America, IMERGlate means were 14%–34% too large, while CHIRP2 means were within a range from −2% to 3% (Table 2). Recent evaluations of IMERG (Nguyen et al. 2020) indicate that this may relate to the overestimation of the wettest precipitation events.
Another rather striking result was the relatively strong performance of CHIRP2, relative to PERSIANN-CCS, in the United States, Iran, and Europe. Given that the products have similar inputs—GEO-IR data—it is interesting to note how CHIRP2’s more consistent seasonal correlations (Fig. 2) and lower bias (Fig. 4) translate into lower MAE values (Fig. 5). Given that the CHIRP2 algorithm is much simpler than the PERSIANN-CCS, which uses a neural net (Hong et al. 2004) and cloud classifications (Hong et al. 2004) to estimate precipitation intensities, these results provide support for spatially explicit algorithms, i.e., relatively simple algorithms which are tuned to specific locations and times.
While the advantages of reducing systematic errors in satellite-only QPE are obvious, an equally important aspect of bias reduction involves the spatial interpolation of intermittent station data. Station networks are not static, and gauge observations appear and disappear, especially in many data-sparse regions. This gauge intermittency predated the satellite era, and led to the widespread use of combinations of static climatologies and time-varying station anomalies, beginning in the late 1990s (Willmott and Matsuura 1995; Willmott and Robeson 1995). Consider, for example, interpolated precipitation estimates halfway between a wet mountain and a dry coastal region (Funk and Shukla 2020).8 If the mountain gauge stops reporting, one might anticipate a spurious “drought” emerging at the point between the mountain and the coast. Modern interpolation schemes, therefore, interpolate station anomalies, and then combine these anomalies with static climatologies to get rainfall in millimeters. Now, allow that static climatology to shift based on information from IMERG, and we have CHIME, a very good basis for “SMART” interpolation (Willmott and Matsuura 1995) of precipitation gauge observations, which results in CHIMES.
Station quality control is also critical. The CHC science team’s extensive experience with station data, and sophisticated data management and quality control procedures (appendix A), help ensure that gauge enhancement avoids most serious quality issues associated with poor-quality station observations. The CHC process and inputs are quite similar to the GPCC’s, but they are more timely, and comparisons indicate high levels of correlation between the GPCC and CHIRPS (Funk et al. 2015b). The IMERGfinal has a latency of about two months, the CHIMESprelim and CHIMESfinal will have latencies of 2 and 18 days.
The cross-validation results, presented in the “Evaluating the relative benefits of rain gauge enhancement” section and Fig. 6, highlight the value of station blending. Using station data, as opposed to gridded station data, as a basis for evaluation is demanding. So the fact that CHIMES can explain 75% of monthly at-station variance is very promising (Fig. 6). In the United States and Africa, IMERGlate performance is fairly good, 38% and 48%, but CHIMES is substantially better (82% and 63%). Another aspect of gauge enhancement is the benefit that stations provide in data-sparse areas, especially when those regions are highly vulnerable. For example, the CHIRPS product only had four or five stations in northern Ethiopia in 2015, but those stations helped capture the worst drought in 50 years, and in conjunction with the Famine Early Warning System Land Data Assimilation System (FLDAS) (McNally et al. 2017), captured the extreme and persistent soil moisture deficits that contributed to creating extreme food insecurity conditions for more than 11 million people. Supported by, and supporting, the United Nations Framework for Climate Services, the CHC engages directly with many national agencies, leading to timely and very valuable data contributions (Table 1). Where station densities are low or declining, the CHIME results presented here suggest the satellite-only CHIME represents a substantial improvement over CHIRP2, PERSIANN-CCS, and GPI.
Thus, in summary, the very promising performance of CHIMES should be understood as arising from dozens of contributions: the GPM satellites, approximately homogenized using the GPM-CO via the numerous algorithms that have been painstakingly assembled by the GPM science team, the CHC efforts focused on building world-class high-resolution climatologies, station archives, and interpolation strategies, which, in turn, rely on the exceptional global, regional, and national data collection and coordinating systems for satellite and in situ observations.
Time series available at http://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/stations-perMonth-byRegion/pngs/.
Pentads break months into six accumulation periods. The first five pentads always have five days. The sixth pentad will contain the remaining days of every month.
North America: 15°–60°N, 165°–45°W; South America: 60°S–15°N, 105°–30°W; Europe: 38°–60°N, 15°W–30°E; Africa: 40°S–38°N, 20°W–45°E; North Asia: 45°–60°N, 30°E–180°; South Asia: 10°–45°N, 50°–150°E; Australia: 50°–13°S, 105°E–180°.
Chapter 8 in the copyright free chapters at https://data.chc.ucsb.edu/people/chris/DroughtEarlyWarningBook/.
Acknowledgments.
Primary support for this work came from the National Aeronautics and Space Administration (NASA) GPM mission Grant 80NSSC19K0686 and the U.S. Agency for International Development (USAID) Cooperative Agreement 72DFFP19CA00001. Additional support for CF was provided by the U.S. Geological Survey Drivers of Drought and the Famine Early Warning Systems Network programs, U.S. Geological Survey (USGS) Cooperative Agreement G14AC00042. Support for GJH and CP-L was provided by NASA under ISFM WBSs 573945.04.80.01.01 and 573945.04.80.01.03.
Data availability statement.
The CHPclim climatology and CHIMES data will be freely available at data.chc.ucsb.edu, and routinely updated on pentadal and monthly basis. The updated schedule will be identical to the CHIRPS processing stream. Once every pentad, with a delay of approximately two days, a “preliminary” product will be created using submonthly stations. Once a month, following the CHC curation process, a “final” monthly CHIMES field will be produced, and “final” pentadal CHIMES be produced. CHIMES will be licensed under a Creative Commons license.
Appendix A: The CHC station curation process
The CHC Station Climatology Database (CSCD) contains over 200,000 (not necessarily unique) stations with over a billion daily records from 65 different data sources. This appendix summarizes CSCD curation and quality-control procedures. In general, each data source has two or more linked relational databases: a station database containing station metadata, and daily and/or monthly tables containing time-varying observations. The metadata contains an important “use rank” identifier. Some stations are marked as low quality and are not used. Other stations, however, are marked as being of intermediate quality. These are typically from rapidly updated automatic reporting systems like the Global Telecommunication System (GTS) and the Global Summary of the Day archive (GSOD). Finally, stations from carefully curated sources, such as the Global Historical Climatology Network (GHCN) and national meteorology agencies, are given high-use rank values. Such observations are given priority, but augmented as necessary by intermediate-quality gauge totals.
GTS and GSOD screening procedures.
While the GTS and GSOD archives are valuable sources of daily precipitation, careful analyses of these sources reveal the occurrence of “false zeros.” False zeros arise when missing data are coded as zero. False zeros can create very misleading and inconsistent results, such as large, erroneous rainfall deficits. To increase the reliability of the GTS and GSOD archives, a comparison between the daily station data and local daily CHIRP values is used to screen and exclude likely false-zero candidates. CHIRP is used so that the station archive can extend back to 1981. This screening dramatically improves performance in some data-sparse regions.
Initial station screening procedures.
The following steps are used to identify and isolate questionable rain gauge stations.
- 1)Station locations are sanity checked to ensure they are in reasonable locations.
- 2)Stations are compared visually and statistically with neighbors. These evaluations are based on precipitation totals for the wettest three months. For each station, the following process is carried out.
- (i)Four high-quality neighboring stations are identified.
- (ii)A time series of distance weighted mean (DWM) of these neighbors is calculated, and the correlation (R) and differences between the individual station and DWM is calculated. The mean of the absolute differences is divided by the mean of the DWM time series to create a fraction of average error (FAE) value. The R and FAE are then combined into a single quality index:
- 3)The sum of the monthly QI values for the three wettest months is used to calculate a “total quality index” (TQI) for each station. Locations with TQI values above 100 are examined visually. Time series plots for each station and its neighbors are evaluated. Maps of elevation and climatological average precipitation provide context. CHC scientists assign each station a “high-,” “intermediate-,” or “low-”quality metric. High-quality stations are used to create the anchor station list (described below). Intermediate stations are considered valid “fill sources.” Low-quality stations are not used in any way.
Creating the anchor station list.
The next step in the CHC processing creates a unique set of “anchor stations” for each month. All stations that pass the screening, as described above, are used to define, for each month, a unique set of “anchor station” locations. These anchor stations are then used in two ways: z
- 1)to build as complete-as-possible time series for each given location, which supports the identification of suspicious values, and
- 2)to support the precomputation of neighbors and neighbor distances, which supports rapid interpolation.
In the first step, all available data sources are used to produce a continuous series of values. A search radius of 0.05° is used to identify potential fill values. In the infilling procedure, higher-quality curated station sources (like the GHCN or national-level meteorological agency data) are preferred over lower-quality sources like the GTS or GSOD. In the second step, for every 0.05° land pixel, a set of nearest neighboring stations is identified, and the pixel-to-station distances are calculated.
Automatic statistical quality control procedures.
Gamma-distribution-based standardized precipitation index (SPI) and absolute magnitude criteria are used to identify, and mark as problematic, suspicious observations. For monthly data, stations failing the following checks are flagged and excluded:
- 1)CHIRP > 75 AND [(CHIRP − station)/CHIRP] > 1
- 2)CHIRP > 75 AND climatology > CHIRP AND 4 × CHIRP < station
- 3)(CHIRP − station)/CHIRP < −1 AND CHIRP > 75
- 4)[Station > (5 × CHIRP) AND CHIRP < 20] OR station > 2000
- 5)(CHIRP > 20 AND station > 0 AND source = (GTS OR GSOD)
- 6)The time series of station at anchor location has at least 100 observation and SPI > 4.
Pentad stations are excluded if any of the following criteria are met:
- 1)[Station > (5 × CHIRP) AND CHIRP > 20] OR station > 300
- 2)(CHIRP > 7 AND station = 0 AND source = (GTS OR GSOD)
- 3)Time series of station at anchor location has at least 100 observation AND SPI > 4
Suspicious observations are not used in further calculations. These automatic procedures are applied every pentad and every month.
Visual quality control via Reality Checks.
In the middle of every month, a set of monthly station totals from the previous month is rigorously evaluated by the “Reality Checks” team at the CHC. In addition to statistical quality parameters, special “Reality Check” images are generated and examined using an interactive web mapping tool, the Early Warning Explorer (EWX). In these monthly Reality Checks images, draft CHIRPS and CHIMES values are overlain with station values, with station values represented as boxes. Individual station values can be rapidly compared to neighboring station observations, as well as the satellite precipitation fields. The comparisons examine absolute values (monthly totals), relative differences (arithmetic anomalies), and statistical likelihood (SPI). The EWX also displays topography and a wide variety of remotely sensed datasets, which facilitates cross-checking extreme precipitation outcomes and identifying stations with false zeros. Analysts also take into account weather model output, placing potential extremes in climatic context. The team discusses suspicious station observations and removes the most likely problematic stations. Additional ancillary data sources, such as NOAA and WMO resources, news reports, and government meteorological reports, are frequently used in the Reality Checks process. These discussions and outcomes are documented on a CHC wiki: https://wiki.chc.ucsb.edu/CHIRPS_Reality_Checks.
Resources for viewing the locations and counts of CHC station archives.
The CHIMES product will build on the same CSCD stations as those used in the CHIRPS product. Readers interested in learning more about the spatial and temporal distributions can use the resources located at http://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/, as described here—https://chc.ucsb.edu/data/chirps/diagnostics.
Global station density.
Every month, we provide GeoTiff files of the number of stations within each pixel for 0.05° resolution and 0.25° resolution. This product is useful for the identification of station-rich and station-poor regions of the globe. The GeoTiff files are available for download https://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/global_monthly_station_density/tifs/, where you can select the resolution of interest (0.05° or 0.25°).
In addition to the spatial representation of the global data, we also make available the locations of all the stations that make it into the CHIRPS final product. These files are available at https://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/list_of_stations_used/monthly/. In this directory you can find files named “global.stationsUsed.YYYY.MM.csv,” which cover the full time series of CHIRPS, and only stations which were available in the historical record at the time of the release of CHIRPS in 2015. Additional stations made available after the initial release are also available in a similar format, but named “extra.stationsUsed.YYYY.MM.csv.” For periods after 2015, you can combine the two files to find the complete list of station locations making their way into CHIRPS.
Stations by country.
There are two diagnostic products to evaluate stations going into CHIRPS for an individual country. The first is a map of the stations’ locations falling within a country’s boundaries, as well as neighboring areas, for every month of the historical record. You can find a listing of the mapped countries at http://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/chirps-n-stations_byCountry/ and then click on the country of interest to find the map for each historical month. At the top of each graphic there is also a count of the number of stations within the country, and in the mapped area (including the country’s surrounding regions) for reference.
In addition to the spatial map, there is also a graphic showing the time series of station input to CHIRPS. This can show the trend in how many stations are reporting over time and reveal where station support may be increasing or decreasing. These plots for each country can be found at http://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/stations-perMonth-byCountry/pngs/. Additional versions of this same plot are made over a geographic region and can be found at https://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/stations-perMonth-byRegion. The process described here generates two sets of station values: preliminary and final. The preliminary station values are produced every pentad, with a latency of about two days. Automatic quality control procedures are applied to these values. The final station values are produced monthly, with a latency of two to three weeks. Automatic statistical analyses and visual curation results in reliable, timely, high-quality gauge archives.
Appendix B: The CHIMES process in equations
Creating the CHPclim climatology.
This regression equation is fit to a sequence of 9 × 9 moving windows, centered on every pixel from 60°S to 60°N. The predictand is the CHPclim climatology. The predictors are
Thus, CHPclim incorporates mean IMERGfinal and elevation fields [Eq. (B1)] and station normals [Eqs. (B2)–(B6)] to produce a continuous high-resolution climatology. These are created for each month, and then the 12 monthly values are disaggregated to daily and pentadal means using a nonparametric smoothing procedure. Visual quality control was also performed on all the monthly fields by the CHC team, in a manner similar to the Reality Checks process described above. Questionable stations were removed, and the gauge enhancement rerun. This process was repeated twice.
Temporal versus spatial homogeneity.
It should be noted that there is a mismatch between the 1981 and 2010 time period used to produce the CSCD climate normals and the 2001–present time period of the IMERGlate. This mismatch arises from the declining number of stations.B1 We selected a 1981–2010 baseline to capture the much greater station density over that time period. We selected this time period to maximize the number of stations used in our climatology. In general, precipitation means change much more rapidly from place to place than from decade to decade. CHIMES has been designed to work well in countries like Ethiopia, in which mean precipitation values can change by 200% or 300% over very short distances. See, for example, (Funk et al. 2015a) which describes our climatology-building process. The boreal spring season in eastern Ethiopia has been declining, but that decline is on the order of 15%–20%, this is also discussed in our CHIRPS paper (Funk et al. 2015b). As a general principle, spatial variability in mean precipitation rates is much stronger than decadal variability, so we have tried to be as contemporaneous as feasible while including a large number of stations. Another factor relates to the shrinking number of available station observations, which can be visualized for all stations (https://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/stations-perMonth-byRegion/pngs/all.station.count.CHIRPS-v2.0.png) and for stations in Africa (https://data.chc.ucsb.edu/products/CHIRPS-2.0/diagnostics/stations-perMonth-byRegion/pngs/Africa.station.count.CHIRPS-v2.0.png). Unfortunately, there are many fewer stations available over the 2001–present time period, and this is why we selected a 1981–2010 mean period. Note also that if there are station data that have shifted between 1981–2010 and 2001–20, then this information would show up in the blending procedure and be incorporated in the CHIMES dataset.
Reducing bias in IMERGlate.
Calculating decorrelation slopes.
The station blending procedure.
References
Aon-Benfield Reinsurance Company, 2021: Weather, climate and catastrophe insight: 2020 annual report. Aon-Benfield Reinsurance Company Rep., 81 pp., https://www.aon.com/global-weather-catastrophe-natural-disasters-costs-climate-change-2020-annual-report/index.html?utm_source=email&utm_medium=clients&utm_campaign=natcat21.
Arkin, P. A. , and B. N. Meisner , 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–84. Mon. Wea. Rev., 115, 51– 74, https://doi.org/10.1175/1520-0493(1987)115<0051:TRBLSC>2.0.CO;2.
Arsenault, K. R. , and Coauthors, 2020: The NASA hydrological forecast system for food and water security applications. Bull. Amer. Meteor. Soc., 101, E1007– E1025, https://doi.org/10.1175/BAMS-D-18-0264.1.
Beck, H. E. , and Coauthors, 2017: Global-scale evaluation of 22 precipitation datasets using gauge observations and hydrological modeling. Hydrol. Earth Syst. Sci., 21, 6201– 6217, https://doi.org/10.5194/hess-21-6201-2017.
Becker, A. , P. Finger, A. Meyer-Christoffer, B. Rudolf, K. Schamm, U. Schneider, and M. Ziese , 2013: A description of the global land-surface precipitation data products of the Global Precipitation Climatology Centre with sample applications including centennial (trend) analysis from 1901–present. Earth Syst. Sci. Data, 5, 71– 99, https://doi.org/10.5194/essd-5-71-2013.
Contractor, S. , and Coauthors, 2020: Rainfall Estimates on a Gridded Network (REGEN)—A global land-based gridded dataset of daily precipitation from 1950 to 2016. Hydrol. Earth Syst. Sci., 24, 919– 943, https://doi.org/10.5194/hess-24-919-2020.
Dinku, T. , C. Funk, P. Peterson, R. Maidment, T. Tadesse, H. Gadain, and P. Ceccato , 2018: Validation of the CHIRPS satellite rainfall estimates over eastern of Africa. Quart. J. Roy. Meteor. Soc., 144, 292– 312, https://doi.org/10.1002/qj.3244.
Funk, C. , 2021: Drought, Flood, Fire: How Climate Change Contributes to Recent Catastrophes. Cambridge University Press, 332 pp.
Funk, C. , and S. Shukla , 2020: Drought Early Warning and Forecasting: Theory and Practice. Elsevier, 238 pp.
Funk, C. , A. Verdin, J. Michaelsen, P. Peterson, D. Pedreros, and G. Husak , 2015a: A global satellite assisted precipitation climatology. Earth Syst. Sci. Data, 7, 275– 287, https://doi.org/10.5194/essd-7-275-2015.
Funk, C. , and Coauthors, 2015b: The Climate Hazards Infrared Precipitation with Stations—A new environmental record for monitoring extremes. Sci. Data, 2, 150066, https://doi.org/10.1038/sdata.2015.66.
Funk, C. , and Coauthors, 2019: Recognizing the Famine Early Warning Systems Network: Over 30 years of drought early warning science advances and partnerships promoting global food security. Bull. Amer. Meteor. Soc., 100, 1011– 1027, https://doi.org/10.1175/BAMS-D-17-0233.1.
Harris, I. , T. J. Osborn, P. Jones, and D. Lister , 2020: Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci. Data, 7, 109, https://doi.org/10.1038/s41597-020-0453-3.
Hong, Y. , K.-L. Hsu, S. Sorooshian, and X. Gao , 2004: Precipitation estimation from remotely sensed imagery using an artificial neural network cloud classification system. J. Appl. Meteor., 43, 1834– 1853, https://doi.org/10.1175/JAM2173.1.
Hou, A. Y. , and Coauthors, 2014: The Global Precipitation Measurement mission. Bull. Amer. Meteor. Soc., 95, 701– 722, https://doi.org/10.1175/BAMS-D-13-00164.1.
Hsu, K. , X. Gao, S. Sorooshian, and H. V. Gupta , 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36, 1176– 1190, https://doi.org/10.1175/1520-0450(1997)036<1176:PEFRSI>2.0.CO;2.
Huffman, G. J. , and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5– 20, https://doi.org/10.1175/1520-0477(1997)078<0005:TGPCPG>2.0.CO;2.
Huffman, G. J. , and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38– 55, https://doi.org/10.1175/JHM560.1.
Huffman, G. J. , and Coauthors, 2018: NASA Global Precipitation Measurement (GPM) Integrated Multi-Satellite Retrievals for GPM (IMERG), version 5.2. National Aeronautics and Space Administration Algorithm Theoretical Basis Doc., 35 pp.
Huffman, G. J. , and Coauthors, 2020: Integrated Multi-Satellite Retrievals for the Global Precipitation Measurement (GPM) mission (IMERG). Satellite Precipitation Measurement, V. Levizzani et al., Eds., Advances in Global Change Research. Springer Nature., 67, 343– 354, https://doi.org/10.1007/978-3-030-24568-9_19
Isaaks, E. H. , and R. M. Srivastava , 1989: An Introduction to Applied Geostatistics. Oxford University Press, 561 pp.
Janowiak, J. E. , R. J. Joyce, and Y. Yarosh , 2001: A real-time global half-hourly pixel-resolution infrared dataset and its applications. Bull. Amer. Meteor. Soc., 82, 205– 217, https://doi.org/10.1175/1520-0477(2001)082<0205:ARTGHH>2.3.CO;2.
Journel, A. G. , and C. Huijbregts , 1978: Mining Geostatistics. Academic Press, 600 pp.
Joyce, R. J. , and P. Xie , 2011: Kalman filter–based CMORPH. J. Hydrometeor., 12, 1547– 1563, https://doi.org/10.1175/JHM-D-11-022.1.
Kidd, C. , A. Becker, G. J. Huffman, C. L. Muller, P. Joe, G. Skofronick-Jackson, and D. B. Kirschbaum , 2017: So, how much of the Earth’s surface is covered by rain gauges? Bull. Amer. Meteor. Soc., 98, 69– 78, https://doi.org/10.1175/BAMS-D-14-00283.1.
Kirschbaum, D. B. , 2020: GPM applications overview. NASA GPM Annual Meeting, Virtual, NASA.
Kirschbaum, D. B. , and Coauthors, 2017: NASA’s remotely sensed precipitation: A reservoir for applications users. Bull. Amer. Meteor. Soc., 98, 1169– 1184, https://doi.org/10.1175/BAMS-D-15-00296.1.
Knapp, K. R. , and Coauthors, 2011: Globally gridded satellite observations for climate studies. Bull. Amer. Meteor. Soc., 92, 893– 907, https://doi.org/10.1175/2011BAMS3039.1.
Kozu, T. , 2001: Development of Precipitation Radar onboard the Tropical Rainfall Measuring Mission satellite. IEEE Geosci. Remote Sens. Lett., 39, 102– 116, https://doi.org/10.1109/36.898669.
Kummerow, C. , W. S. Olson, and L. Giglio , 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 1213– 1232, https://doi.org/10.1109/36.536538.
Kummerow, C. , W. Barnes, T. Kozu, J. Shiue, and J. Simpson , 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 809– 817, https://doi.org/10.1175/1520-0426(1998)015<0809:TTRMMT>2.0.CO;2.
Kummerow, C. , and Coauthors, 2000: The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. J. Appl. Meteor., 39, 1965– 1982, https://doi.org/10.1175/1520-0450(2001)040<1965:TSOTTR>2.0.CO;2.
Kummerow, C. , and Coauthors, 2001: The evolution of the Goddard profiling algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801– 1820, https://doi.org/10.1175/1520-0450(2001)040<1801:TEOTGP>2.0.CO;2.
Kummerow, C. , D. L. Randel, M. Kulie, N.-Y. Wang, R. Ferraro, S. Joseph Munchak, and V. Petkovic , 2015: The evolution of the Goddard profiling algorithm to a fully parametric scheme. J. Atmos. Oceanic Technol., 32, 2265– 2280, https://doi.org/10.1175/JTECH-D-15-0039.1.
Magadzire, T. , G. Galu, and J. Verdin , 2017: How climate forecasts strengthen food security. WMO Bull., 66 ( 2), https://public.wmo.int/en/resources/bulletin/how-climate-forecasts-strengthen-food-security.
McNally, A. , and Coauthors, 2017: A land data assimilation system for sub-Saharan Africa food and water security applications. Sci. Data, 4, 170012, https://doi.org/10.1038/sdata.2017.12.
Nguyen, P. , M. Bador, L. Alexander, T. Lane, and C. Funk , 2020: On the robustness of annual daily precipitation maxima estimates over monsoon Asia. Front. Climate, 2, 578785, https://doi.org/10.3389/fclim.2020.578785.
Roca, R. , and Coauthors, 2019: FROGS: A daily 1° × 1° gridded precipitation database of rain gauge, satellite and reanalysis products. Earth Syst. Sci. Data, 11, 1017– 1035, https://doi.org/10.5194/essd-11-1017-2019.
Ross, K. , M. E. Brown, J. P. Verdin, and L. Underwood , 2009: Review of FEWS NET biophysical monitoring requirements. Environ. Res. Lett., 4, 024009, https://doi.org/10.1088/1748-9326/4/2/024009.
Ruane, A. C. , and Coauthors, 2021: Strong regional influence of climatic forcing datasets on global crop model ensembles. Agric. For. Meteor., 300, 108313, https://doi.org/10.1016/j.agrformet.2020.108313.
Schneider, U. , A. Becker, P. Finger, A. Meyer-Christoffer, M. Ziese, and B. Rudolf , 2013: GPCC’s new land surface precipitation climatology based on quality-controlled in situ data and its role in quantifying the global water cycle. Theor. Appl. Climatol., 115, 15– 40, https://doi.org/10.1007/s00704-013-0860-x.
Schneider, U. , P. Finger, A. Meyer-Christoffer, E. Rustemeier, M. Ziese, and A. Becker , 2017: Evaluating the hydrological cycle over land using the newly-corrected precipitation climatology from the Global Precipitation Climatology Centre (GPCC). Atmosphere, 8, 52, https://doi.org/10.3390/atmos8030052.
Senay, G. B. , and J. Verdin , 2003: Characterization of yield reduction in Ethiopia using a GIS-based crop water balance model. Can. J. Remote Sens., 29, 687– 692, https://doi.org/10.5589/m03-039.
Shukla, S. S. , and Coauthors, 2021: Multistage and scale drought prediction to support food insecurity early warning (invited presentation). 101st Annual Meeting, Virtual, Amer. Meteor. Soc., 1B.5, https://ams.confex.com/ams/101ANNUAL/prelim.cgi/Paper/379647.
Skofronick-Jackson, G. , and Coauthors, 2017: The Global Precipitation Measurement (GPM) mission for science and society. Bull. Amer. Meteor. Soc., 98, 1679– 1695, https://doi.org/10.1175/BAMS-D-15-00306.1.
Turk, F. J. , E. E. Ebert, H.-J. Oh, B.-J. Sohn, V. Levizzani, E. A. Smith, and R. Ferraro , 2003: Validation of an operational global precipitation analysis at short time scales. 12th Conf. on Satellite Meteorology and Oceanography, Long Beach, CA, Amer. Meteor. Soc., JP1.2, https://ams.confex.com/ams/annual2003/techprogram/paper_56865.htm.
Turk, F. J. , S. Miller, and C. Castello , 2010: A dynamic global cloud layer for virtual globes. Int. J. Remote Sens., 31, 1897– 1914, https://doi.org/10.1080/01431160902926657.
UN Food Security Information Network, 2021: Global report on food crises 2021. UN FSIN Rep., 307 pp., https://www.fsinplatform.org/sites/default/files/resources/files/GRFC%202021%20050521%20med.pdf.
Verdin, J. , and R. Klaver , 2002: Grid-cell-based crop water accounting for the Famine Early Warning System. Hydrol. Processes, 16, 1617– 1630, https://doi.org/10.1002/hyp.1025.
Verdin, J. , C. Funk, R. Klaver, and D. Roberts , 1999: Exploring the correlation between southern Africa NDVI and Pacific sea surface temperatures: Results for the 1998 maize growing season. Int. J. Remote Sens., 20, 2117– 2124, https://doi.org/10.1080/014311699212380.
Verdin, J. , C. Funk, G. Senay, and R. Choularton , 2005: Climate science and famine early warning. Philos. Trans. Roy. Soc. London, 360B, 2155– 2168, https://doi.org/10.1098/rstb.2005.1754.
Willmott, C. J. , and K. Matsuura , 1995: Smart interpolation of annually averaged air temperature in the United States. J. Appl. Meteor., 34, 2577– 2586, https://doi.org/10.1175/1520-0450(1995)034<2577:SIOAAA>2.0.CO;2.
Willmott, C. J. , and S. M. Robeson , 1995: Climatologically aided interpolation (CAI) of terrestrial air temperature. Int. J. Climatol., 15, 221– 229, https://doi.org/10.1002/joc.3370150207.
Willmott, C. J. , S. G. Ackleson, R. E. Davis, J. J. Feddema, K. M. Klink, D. R. Legates, J. O’Donnell, and C. M. Rowe , 1985: Statistics for the evaluation and comparison of models. J. Geophys. Res., 90, 8995– 9005, https://doi.org/10.1029/JC090IC05P08995.
Xie, P. , and P. A. Arkin , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539– 2558, https://doi.org/10.1175/1520-0477(1997)078<2539:GPAYMA>2.0.CO;2.
