2021 Volume 99 Issue 2 Pages 423-436
The theory of extreme precipitation has matured over the last decade and stipulates that the intensity of the extreme precipitation is balanced with the surface humidity. The changes in surface humidity can further be approximated by the changes in surface temperature. The analytically derived scaling coefficient based on the Clausius–Clapeyron derivative is ∼ 6 % K−1 in the tropics. While frequently confronted with observations over land, the theory has so far only been marginally evaluated against precipitation data over the ocean. Using an ensemble of satellite-based precipitation products and a suite of satellite-based sea-surface temperature (SST) analyses at 1°-1day resolution, extreme scaling is investigated for the tropical ocean (30°S–30°N). The focus is on the robust features common to all precipitation and SST products. It is shown in this study that microwave constellation-based precipitation products are characterized by a very robust positive scaling over the 300–302.5-K range of 2-day-lagged SST. This SST range corresponds to roughly 60 % of the amount of tropical precipitation. The ensemble mean scaling varies between 5.67 ± 0.89 % K−1 and 6.33 ± 0.81 % K−1 depending on the considered period and is found to be very close to the theoretical expectation. The robustness of the results confirms the suitability of the current generation of constellation-based precipitation products for extreme precipitation analysis. Our result further confirms the extreme theory for the entire tropical ocean. Yet, the significant differences in the magnitude of the extreme intensity of the products require dedicated validation efforts.
The exchanges between energy and water within the Earth system are related to the major feedback processes responsible for the fate of the Earth's climate under increasing greenhouse gas concentrations (Stephens et al. 2020). The second principle of thermodynamic and the Clausius–Clapeyron relation indicate that water vapor in the atmosphere increases at a rate of 6–7 % with each increasing degree warming, which is at the heart of the strongly positive water vapor feedback. The increased loading of water vapor in the atmosphere has a strong impact on both the mean precipitation (Stephens and Ellis 2008) and distribution of its extreme (Trenberth 1999).
The theory of extreme precipitation dependence on surface temperature has indeed matured significantly over the last decade, and currently, a physical framework is well established (O'Gorman and Schneider 2009; O'Gorman 2015; Fischer and Knutti 2016). However, cloud-resolving model-idealized simulations over the tropical oceans exhibit diverging sensitivities (Muller and Takayabu 2020). Land-based studies using conventional precipitation observations over the tropical land also obtained contrasting results (Westra et al. 2014). Nevertheless, recent investigations using satellite observations have identified robust extreme precipitation regimes, which is in agreement with the theoretical expectations from thermodynamics (Roca 2019). However, over the ocean, the scarcity of conventional in situ precipitation observations (Serra 2018) precluded investigations on the scaling of extreme precipitation with surface temperature with notable exceptions.
Using early instantaneous satellite observations from special sensor microwave/imager (SSM/I) and monthly sea surface temperature (SST) from Hadley Centre Sea Ice and Sea Surface Temperature dataset (Rayner et al. 2003), Allan and Soden (2008) calculated the rate of change of 2.5° × 2.5° daily precipitation to tropical mean SST anomalies. Extreme precipitation indicates a scaling ranging from slightly above to much larger than their value derived from the Clausius–Clapeyron response, depending on the baseline for the anomaly computations. The analysis further pointed out systematic underestimation of the response of the climate models, thus prompting a more in-depth analysis of the behavior of the extremes over oceans. Nevertheless, the use of tropical mean SST anomaly prevents the processes at play from being further addressed.
More recently, using the OceanRAIN dataset of ship-based disdrometer precipitation measurements (Klepp et al. 2018), the scaling of instantaneous (at a 1-min scale) extreme precipitation to the local simultaneous SST has been revisited for the global ocean (Burdanowitz et al. 2019). The disdrometer data exhibits a single increasing regime of the 99th percentile with local SST. The scaling is computed by pooling available data over a large range of SST (0–30°C) and is shown to vary between 6 % K−1 and ∼9 % K−1, depending on the regression technique used. Yet, the small sample of the ship-based measurements prevents a definitive conclusion, and the relative role of the convective dynamics and thermodynamics remains to be elucidated at these scales.
In summary, numerous observation-based assessments of the scaling theory have been conducted under continental conditions, whereas model-based assessments are conducted under oceanic conditions. In comparison, the tropical ocean so far benefited from very few investigations based on observations. In this paper, we aimed to improve this situation by pooling recent satellite observations at the 1° × 1° daily scale of both precipitation and surface temperature all over the tropical oceans.
Over ocean, these satellite-based products is generally evaluated using the few available data obtained from buoy networks (Wu and Wang 2019), atolls rain gauges observations (Greene et al. 2008), or island-based radar measurements (Henderson et al. 2017). More recently, the release of a new ship-based disdrometer measurement database (Klepp et al. 2018) provides a complementary reference dataset over some commercial ship routes and research vessel campaigns. For large space and time scales, consistency analysis can be conducted through water and energy budget conservation analysis (L'Ecuyer et al. 2015). As a consequence of these limited verification references, compared with land, the capability of these satellite-based products to describe the precipitation field is generally not very well documented (Sun et al. 2018), not mentioning under the extreme rainfall conditions. Instead of elaborating on the difficulty to quantify the accuracy of the products, the rationale of this study is to focus on the robustness of the analysis across numerous satellite products to investigate the thermodynamic scaling of extreme precipitation with SST.
Section 2 introduces a number of satellite-based products of both precipitation and SST and details the methodology followed in this study. Section 3 presents the results and a sensitivity of the results to various assumptions. Finally, a summary and discussion section conclude the paper.
The list of the various products under consideration is presented in Table 1. All the datasets are used at the same 1° × 1° daily resolution over the 30°S–30°N region and originate mostly from the Frequent Rainfall Observations on GridS (FROGS) database (Roca et al. 2019a, b). While the products share some of the raw satellite observations, they differ in numerous aspects, which can influence their capability of describing precipitation. They are produced from different instantaneous rain rate algorithms. The daily accumulation from the various products further benefits from different sampling from single or multiple platforms, from infrared and/or microwave (MW) imagers and/or sounders. Finally, some products also incorporate in situ corrections.
The Global Precipitation Climatology Project (GPCP) is a pioneer effort to provide satellite-based precipitation estimates globally (Huffman et al. 1997). Here, we use the 1° daily climate data record (CDR) v1.3 detailed by Huffman et al. (2001). The GPCP 1° daily (1DD) product relies on a single MW platform, infrared measurements, and the GPCC analysis over land. The daily estimates are adjusted to mimic the GPCP monthly product when aggregated over a month. Note that the GPCP monthly product is used in various other products as an adjustment reference. The Precipitation Estimation from Remotely Sensed Information using Artificial Neural Network (PERSIANN) family of products has various incarnations. Thus, here, we use the climate-oriented product, namely, the CDR version, also known as PERSIANN CDR v1 (Ashouri et al. 2015). PERSIANN is based on infrared imagery and is adjusted to the GPCP monthly mean. It can be considered as an alternative downscaling of the GPCP monthly data to that of GPCP 1DD. The National Oceanic and Atmospheric Administration (NOAA) climate prediction center morphing technique (CMORPH) satellite precipitation estimates (Xie et al. 2017) rely on the constellation of both MW imagers and sounders and infrared-derived cloud motion winds. The blending of the various data is performed, thanks to Kalman filtering (Joyce and Xie 2011). Over ocean, the product is adjusted to the GPCP accumulations. Compared with buoys, the CMORPH under(over) estimates rainfall over the Atlantic (Pacific) ocean (Wu and Wang 2019). The Tropical Rainfall Measuring Mission Multi-satellite Precipitation Analysis (TMPA) product is a widely used dataset that combines radar, infrared, MW imagers, and sounder satellite observations (Huffman et al. 2007). Here, we use v7. Over tropical Atlantic (Pacific), the TMPA product shows under(over)estimation (Wu and Wang 2019). Over the Northern Indian Ocean, the TMPA product generally overestimates precipitation compared with the buoys but underestimates heavy precipitation events over 100 mm d−1 (Prakash and Gairola 2014). Note that earlier investigation noted that TMPA monthly means are similar to the GPCP product over the tropical ocean (Huffman et al. 2007). The Hamburg Ocean-Atmosphere Parameters and Fluxes from Satellite Data (HOAPS) product provides a satellite-based suite of freshwater budget parameters, including precipitation over sea-ice-free ocean surface (Andersson et al. 2010). This MW-only product relies on multiplatforms inter-calibrated measurements from SSM/I and special sensor microwave imager sounder (SSMIS) (Fennig et al. 2020). Here, v3.2 of the precipitation product is used (Andersson et al. 2017). When compared with the OceanRAIN in situ data, the HOAPS instantaneous rain rate underestimates the intensity in the intertropical convergence zone, especially for high rain rate, but at the same time, the HOAPS retrieval overestimates the occurrence of precipitating cases (Bumke et al. 2019). The Tropical Amount of Rainfall with Estimation of ERors (TAPEER) algorithm uses geostationary infrared imagery, together with MW imager instantaneous rain rate estimates plus the SAPHIR sounder to estimate the accumulation of daily precipitation (Roca et al. 2020). The addition of the Megha-Tropiques platform to the constellation is shown to improve the product compared with the implementation of imagers alone (Roca et al. 2018).The recently released TAPEER product has been extensively evaluated over West Africa (Gosset et al. 2018) but is not yet well characterized over the tropical ocean. The Global Satellite Mapping of Precipitation (GSMaP) product provides high-resolution precipitation estimations using satellite observations from multiple platforms (Kubota et al. 2020). This product is mainly based on the MW estimation of rainfall from a suite of MW imagers and sounders. The MW instantaneous rain rate estimates (Aonashi et al. 2009; Shige et al. 2009) are propagated based on cloud motion wind vectors originally derived from IR geostationary imagery (Ushio et al. 2009). Here, the near real time v6 product is used. The Integrated Multi-satellite Retrievals for Global Precipitation Measurement (GPM) Mission (IMERG) is developed at NASA based on infrared observations and both MW imager and sounder data (Huffman et al. 2020). It takes advantage of the TMPA product, the CMORPH-Kalman filter approach (Joyce and Xie 2011), and the PERSIANN-Cloud Classification System algorithm (Hong et al. 2004). The most recent v6 release of the final run product is used. Previous evaluation efforts demonstrated that the older version (v4) overestimates rain rates compared with buoy observations at an hourly 0.1° × 0.1° scale over the North Indian Ocean (Prakash et al. 2017). Version 5 daily 1° × 1° estimates are shown to underestimate the OceanRAIN measurements on average, which is likely due to the inclusion of very light rain rates in the statistics (Khan and Maggioni 2019). While IMERGv5 has similar biases to CMORPH and TMPA with respect to in situ data, it performs better in the mean value estimation with the exception of the Atlantic Ocean regime over 4 mm d−1 where the product is shown to significantly underestimate precipitation (Wu and Wang 2019). Version 6 of the product benefits from a refined intercalibration procedure and better interpolation between the platforms that demonstrate improvements on some metrics (Tan et al. 2019) and should also be reflected in these prior evaluations. The multi-source weighted-ensemble precipitation (MWSEP) is a product corresponding to a pragmatic approach that average existing products to provide the best estimate. Over land, an optimization method based on hydrological modeling and observed stream gauge data is used (Beck et al. 2017). Over the ocean, a merging procedure is also followed. The various sources of precipitation used encompass satellite-based products, rain gauge measurements, and reanalysis results. Here, we use v2.2 (Beck et al. 2019). From 2000 onward, the MSWEP estimates over the tropical ocean are weighted much more toward satellite products than the reanalysis. CMORPH, GSMaP-MKV v5, and TMPA v7 RT are blended in the MSWEP. Note that over the ocean, TMPA v7 RT and TMPA v7 are similar, whereas over land, they are different. The geographical distribution of the 99th percentile of 3-hourly precipitation of the MSWEP v2.2 is very close to that of the CMORPH over the tropical ocean (Beck et al. 2019).
The resulting variety of precipitation distribution from these nine products is presented in Fig. 1. The two products that do not rely on the MW constellation (GPCP and PERSIANN) exhibit a strongly decreasing occurrence with the amount of daily precipitation. GPCP does not exhibit values above 170 mm d−1. The TAPEER product exhibits a systematic significant underestimation of the probability of the largest rain accumulation compared with other MW-based products. This is likely due to the relative lack of light rain situations (Roca and Fiolleau, 2020). In addition, note that the TAPEER product is only available for a limited time period. Due to their outlier distribution and documented biases, these three products are not considered in the following. The figure further demonstrates two clusters of products showing both the occurrence smoothly declining exponentially toward the very high accumulation of daily precipitation. The first cluster is composed of the CMORPH, TMPA, and MSWEP. The members of this cluster are very close to the MSWEP being built from the CMORPH and TMPA. Both the CMORPH and TMPA further share the same adjustment to the common reference of GPCP pentads over the ocean. The second cluster includes GSMaP, HOAPS, and IMERG that do not share much in terms of algorithms, methods, and input data. While the two clusters demonstrate a similar distribution of up to 100 mm d−1, above that threshold, the first cluster systematically underestimates the occurrence compared with the second cluster. This is consistent with previous analysis indicating that the climatological mean behavior of the products is a poor predictor of the products' extreme precipitation (Masunaga et al. 2019). The products of the first cluster have further been shown to underestimate the intense rain rates compared with buoys (Prakash and Gairola 2014; Wu and Wang 2019), which could be related to their difference with the second cluster.
Probability of exceedance of daily 1° × 1° accumulated precipitation over the tropical ocean (30°S–30°N) for the period 2007–2017 except for the TAPEER product where it is restricted to the period 2012–2016. The probability of exceedance is computed with respect to wet-days with precipitation above 1 mm d−1.
Satellite-based SST products readily provide estimations of the foundation SST, which is defined as the ocean temperature at a depth of ∼ 1 m, below the diurnal layer, at a high space and time resolution over a long temporal record. Most of the products rely on optimal interpolation techniques to merge various satellite observations with in situ measurements. While such products have a well characterized accuracy under clear sky conditions of ∼ 0.5 K (Donlon et al. 2012), their representativity under (extreme) rainfall conditions is not well documented. Indeed, the cloudiness associated with the rainfall prevents the use of infrared measurements, and the MW signal, while being helpful in cloudy but non-precipitating cases, is altered by rain drop emission during the heavy precipitation situations, also preventing SST estimation in this case (Wentz et al. 2000). As a consequence, the SST estimates under precipitating conditions mainly rely on the result of the optimal interpolation method and are smoothed over several days at a large distance. In the following, we interpret these fields as “synoptic” SSTs and evaluate whether such synoptic SSTs are a good and robust proxy for the thermodynamics of the extreme precipitation events over the ocean. A suite of three products is used to evaluate the sensitivity of the scaling to the input data and to the optimal technique implementation and are summarized in Table 2. All SST products have been regridded at a daily 1° × 1° resolution over the 30°S–30°N region to match the precipitation data using a simple conservative averaging procedure.
The Operational Sea Surface Temperature and Ice Analysis (OSTIA) (Donlon et al. 2012) product provides a global foundation SST field derived from satellite observations in the MW (two platforms) and the infrared (six platforms) and in situ measurement at 1/20° and daily resolution since 2007. The analysis is a multi-scale optimal interpolation using a 3-day assimilation window centered on the day of the analysis, and the background field is the analysis of the previous day. The error correlation length scale used in the OSTIA in the tropics is around 100 km. The product optimum interpolation sea surface temperature (OISST) (Banzon et al. 2016) uses IR measurements from the Advanced Very High-Resolution Radiometer (AVHRR) instruments onboard the NOAA satellites since 1981. The interpolation method is similar to that of the OSTIA. The error correlation length scales in the tropics are about 150–200 km and 3 days. The nominal resolution is daily at 0.25°. OISST v2 is used here. Finally, the MW optimally interpolated (OI) SST product from Remote Sensing Systems (Gentemann et al. 2010) is used. As some rain-contaminated SST estimate may persist in the MW-derived dataset, a stringent quality control is performed to prevent potentially biased SST retrievals. The error correlation scales used are 100 km and 3 days, the background field is the analysis of the previous day, and the SST represents a foundation SST remapped at a daily 1° × 1° resolution. This product is referred to as the OIRSS in the following.
2.3 Method a. Background on the theoryThe background on the scaling of extreme precipitation with surface temperature has been detailed and reviewed in various publications (O'Gorman 2015; Allan and Liu 2018; Roca 2019; Muller and Takayabu 2020) and is only briefly summarized below. Based on the dry static energy budget of the atmosphere (Muller et al. 2011), the rate of extreme precipitation in the tropics can be expressed as follows:
 |
with warming highlights three contributions, that is, the change in microphysics related to the efficiency term, the change in dynamics through ρω, and the change in thermodynamics through 
. At the daily 1° × 1° scale, with the assumption that the related change of the dynamic and efficiency contributions remains low with warming and by neglecting the vertical variations in ω, the scaling of extreme precipitation becomes (Muller et al. 2011).
 |
with temperature, that is, the Clausius–Clapeyron rate. Under typical tropical sea surface conditions, the expected value ranges between 5.5 % K−1 and 6.5 % K−1.
b. Methodological aspects
Extreme daily precipitation is characterized using high percentiles of the wet-day (P > 1 mm d−1) distribution. This index is chosen over others (Zhang et al. 2011) due to its relevance for scaling investigations (Schär et al. 2016). The scaling of the precipitation extremes with the SST is calculated using the binning method, which is well suited to our investigation given the large number of available observations (Roca 2019). Indeed, for each 0.5 K degree SST interval, the precipitation data from the whole tropical ocean are pooled together, and the 99.9th percentile of the precipitation distribution is estimated. Then, after identifying a relevant regime, linear regression (in the logarithm space) is employed to compute the scaling factor defined as the rate of change of the 99.9th percentile with SST.
c. Sensitivity studies for robustness determinationTo identify the robust aspects of the thermodynamic scaling estimation using satellite observations, we perform a sensitivity study in complement to the use of various precipitation and SST products. We also explore how the timing between the SST and precipitation influences the scaling and how the overall statistical analysis is sensitive to the selected period.
As already well explored for continental cases (e.g., Bao et al. 2017), intense precipitation events can strongly alter the surface heat budget and the surface temperature. The strong gust and downdrafts associated with deep convection result in a lower surface temperature compared with the case of no rain (Lafore et al. 2016). As a result, the surface conditions may not represent the large-scale environment of the event but are impacted by the event itself. As discussed above, the synoptic SST product assimilation scheme prevents these impacted surface conditions from being employed in the estimation; however, the risk of misattributing a rain event to a given temperature range due to this coupling is unclear. As a mean to evaluate the impact on our findings, the timing of the association between the SST and precipitation event is varied from simultaneous and lagged analysis up to 2 days before.
The constellation-based precipitation products are characterized by a changing configuration of the constellation over the last two decades, depending on the availability of MW imagers and sounders that could influence their capability of steadily monitoring the precipitation in the tropics (Roca et al. 2020).The SST products are also sensitive to the availability of the MW imagers and the infrared radiometers. The availability of these platforms since 2001 is presented in Fig. 2. The baseline period for our investigation spans from 2007 to 2017, which is the longest period shared by all SST products. It corresponds to the homogeneous cycle of OSTIA production (Donlon et al. 2012) and to the start of the systematic use of two AVHRR instruments for the OISST product (Banzon et al. 2016). Two other different periods are considered. The first one is longer, which extends back in time until 2001 for which the OIRSS and OISST are available and correspond to the availability of the GSMaP precipitation product. During the period of 2001–2007, the MW imagers fluctuated from four to six platforms and the sounders from two to four, whereas the IR radiometers increased from one to three. These constellation configurations are all less populated than the forthcoming period post 2007, and the products are associated with less sampling than during the period of 2007–2017 that could impact our analysis. The second period is a shorter one and corresponds to a precipitation data-rich period that includes the Megha-Tropiques operations (Roca et al. 2018) and partly includes the TRMM and GPM platforms along with the SSM/I and SSMIS platforms. This period is characterized by up to 14 MW sensors operating simultaneously. The use of three periods, namely, 18, 11, and 5 years, also allows for the assessment of the sensitivity of the estimation of the extreme percentile of the precipitation distribution.
Time series of the availability of microwave imagers and sounders used in the precipitation products of the study. Vertical dashed lines indicate the different time periods explored in the study.
Figure 3 presents the scaling of the 99.9th percentile as a function of the simultaneous SST from the OSTIA product for the six constellation-based products. The two clusters of precipitation products previously identified are also presented. The low-cluster 99.9th percentile ranges from 120 mm d−1 to 150 mm d−1, whereas the high cluster 99.9th percentile ranges from 170 mm d−1 to 240 mm d−1. There is roughly a factor of two among the least value of CMORPH at 297 K and the highest one of IMERG at 302.5 K. The dependence of the high percentile on the SST is characterized by three regimes. Up to 300.25 K, the products do not exhibit any robust behavior. The GSMaP is mainly increasing over this SST range, whereas the HOAPS extremes decrease; the other products rather exhibit no sensitivity to the SST. This regime accounts for only 19 % of the total tropical precipitation accumulation. The second regime spans SST from 300 up to 302.5 K and is characterized by an increase of all the products. Almost 56 % of the total rainfall belongs to this regime. For the last regime, above 302.25 K and corresponding to 25 % of the total rainfall amount, the precipitation products exhibit a robust decrease in the value of the high percentile. For the second regime, the ensemble mean of the scaling is ∼ 5 % K−1, with a small coefficient of variation of around 10 % and a range of 4.21–5.75 % K−1 (Table 3a). A robust scaling, which is close to the Clausius–Clapeyron, is found in this case.
The value of the 99.9th percentile of the 1° × 1° daily accumulated precipitation as a function of the contemporaneous SST from the OSTIA product. Each color corresponds to a precipitation product. For the period 2007–2017. Regimes are separated by vertical dashed lines. The grey shaded area indicates the non-robust cold regime between precipitation products. Black dash-dotted lines correspond to the Clausius-Clapeyron 6 % K−1 rate. See text for details.
The replacement of the OSTIA by the OISST does not change much the regime decomposition (not shown), but the ensemble scaling over regime 2 is much smaller (3.4 % K−1) and spreads almost twice much more (Table 3b). However, the use of the OIRSS data drastically alters the picture with no well-defined regimes at all. Over the previous SST range of regime 2, the ensemble scaling is even slightly negative (−0.45 % K−1). Hence, the scaling obtained from the simultaneous measurements of precipitation and surface temperature are not robust to the selection of the SST products.
3.2 Lagged analysisThe lagged analysis using the SST of the day before the precipitation does not change much the overall picture (not shown). Quantitatively, the ensemble mean of the scaling slightly increases to 5–6 % K−1 with a coefficient of variation less than 11 % for all the OSTIA and OISST products (Tables 3a, b). The OIRSS product now exhibits a scaling of 4.4 ± 0.6 % K−1. However, the lagged analysis at 2 days before confirms the previously identified low- and high-precipitation product clusters and reveals a very robust pattern across the SST products. Figure 4 confirms the relevance of the range of the scaling regime identified for the OSTIA for all the SST products. The sensitivity of the selection of the upper-bound 303 K instead of 302.5 K (not shown) does not significantly modify the values. With a 2-day lag, the ensemble mean scaling is very similar for each SST product: 5.9, 5.9, and 6.0 % K−1 for the OSTIA, OISST, and OIRSS, respectively. The ensemble variance is also similar among the products at around 10 % (or 0.65 % K−1). The multi-precipitation and multi-SST product ensemble of 18 combination statistics reads 5.93 ± 0.60 % K−1. The “cold” regime with SST < 300 K remains non-robust at a 2-day lag, with the precipitation products not agreeing on the sensitivity of the 99.9th percentile to the SST. The warm regime with SST ≥ 302.5 K is characterized by a robust behavior across the precipitation products that appear not to be robust across the SST selection. The marked decrease in the extreme in the OSTIA after 302.5 K is rather associated with a smooth, steady evolution for the OISST and OIRSS. Note that the GSMaP product, unlike any other product, indicates a CC-like positive scaling over most of the cold regime. This unique feature deserves further attention and will be investigated in the future.
The value of the 99.9th percentile of the 1° × 1° daily accumulated precipitation as a function of the SST lagged by 2 days. Each color corresponds to a precipitation product. Solid line for OSTIA, dashed line for OISST and dash-dotted lines for OIRSS. For the period 2007–2017. Regimes are separated by vertical dashed lines. The grey shaded areas indicate the non-robust cold regime between precipitation products (left) and the non-robust warm regime between SST products (right). Black dash-dotted lines correspond to the Clausius-Clapeyron 6 % K−1 rate. See text for details.
The analysis for the longer period is presented in Fig. 5. The “cold” regime results still hold in this case, although its upper limit can be revised to 299.5 K. Conversely, the “warm” regime is now more robust among the SST products, but the 302.5 K limit is less clearly marked than for the 2007–2017 period. The values of the 99.9th percentile are slightly higher for all the precipitation products compared with the previous period. The scaling over the 300–302.5 K regime demonstrates a smoother sensitivity than before with an ensemble mean value of 4.3 and 5.3 % K−1 for the two SST products (Tables 3b, c). When the range is slightly adjusted to 299.5–302.0 K, the regime corresponds to 45 % of the total precipitation, and the scaling now reads 5.65 % K−1 and 5.69 % K−1 for the OISST and OIRSS, respectively. In this case, the spread of the ensemble scaling for the OISST is diminished to ∼ 17 % instead of 24 % (Table 3b) and remains the same at ∼ 15 % for the OIRSS. In this case, the ensemble of 12 combinations for all the SST and precipitation products is 5.67 ± 0.89 % K−1.
The value of the 99.9th percentile of the 1° × 1° daily accumulated precipitation as a function of the SST lagged by 2 days. Each color corresponds to a precipitation product. Dashed line for OISST and dash-dotted lines for OIRSS. For the period 2001–2017. Regimes are separated by vertical dashed lines. The grey shaded areas indicate the non-robust cold regime between precipitation products (left) and the non-robust warm regime between SST products (right). Black dashdotted lines correspond to the Clausius-Clapeyron 6 % K−1 rate. See text for details.
Figure 6 confirms the delineation of the three regimes for the shorter period. The magnitude of the extreme precipitation is close to that of the 2001–2017 period. The ensemble mean scaling is slightly larger and now ranges between 6.0 % K−1 and 6.8 % K−1, whereas the 18 combinations for all the SST and precipitation products is 6.33 ± 0.81 % K−1, which is also in close agreement with the theoretical expectation. Owing to its availability over that period, the TAPEER product is also shown for the sake of completeness, but it is not included in the ensemble statistics. It is in line with the other products, although with a much lower magnitude of the 99.9th percentile marker as presented in Fig. 1. However, the scaling mean value is ∼ 5.1 % K−1, which is slightly less than that for the other products.
The value of the 99.9th percentile of the 1° × 1° daily accumulated precipitation as a function of the SST lagged by 2 days. Each color corresponds to a precipitation product. Solid line for OSTIA, dashed line for OISST and dash-dotted lines for OIRSS. For the period 2012–2016. Regimes are separated by vertical dashed lines. The grey shaded areas indicate the non-robust cold regime between precipitation products (left) and the non-robust warm regime between SST products (right). Black dash-dotted lines correspond to the Clausius-Clapeyron 6 % K−1 rate. See text for details.
This study aimed to explore the scaling of extreme precipitation with surface temperature over the tropical oceans. The pooling of data originating from an ensemble of six constellation-based precipitation products reveals two clusters of products that differ in terms of the magnitude of the extreme daily accumulation. The difference between the low-cluster (MSWE, TMPA, and CMORPH) and high cluster (HOAPS, GSMaP, and IMERG) spans roughly a factor of two at the most. No such clustering is found over land, likely due to the inclusion of rain gauges in most of the products (Roca 2019; Masunaga et al. 2019). While buoy-based comparison suggests that the low-cluster products indeed underestimate intense precipitation intensity, no definitive argument yet permits to prefer one cluster from the other. Despite this difference in the magnitude, the present results highlight a very robust behavior of the satellite product when depicting extreme precipitation sensitivity with SST. The explanation for the similarities and differences of the individual product would require a dedicated study and is out of the scope of the present work, which focuses on the robust features of precipitation product ensemble.
The timing of the SST–precipitation relationship has also been explored. Moreover, it has been demonstrated that lagged analysis at 2 days exhibits robust regimes across the SST products. This is likely due to the employment of the SST analysis that blends data over 3 days and at ∼ 100 km to provide an SST estimate under precipitating conditions. Our study confirms the suitability of these analyzed SST for precipitation-related investigations. Specifically, the lagged analysis at 2 days identifies three distinct regimes:
A “cold” regime with SST < 300 K corresponds to ∼ 19 % of the total tropical precipitation amount. In this case, while the results are not sensitive to the SST product selection, or to the timing of the precipitation–temperature association or the length of the record, the various precipitation products exhibit inconsistent behaviors. The lack of robustness of the results might be caused by some structural errors in the precipitation retrievals and/or by the weak data sampling that prevents a robust estimation of the high percentile of the precipitation distribution. The 99th percentile (not shown) that ranges between 60 mm d−1 and 100 mm d−1 over the cold regime is less sensitive to the data sampling than the 99.9th percentile and demonstrates a lack of sensitivity that is more reproducible among the products. Only the HOAPS product differs from the other five products.
A “warm” regime with SST > 302.5 K corresponds to ∼ 25 % of the total precipitation amount and is characterized by a systematic decrease in the values of the 99.9th percentile from 302.5 K to the warmest SST under considerations. This warm regime is also observed over land (Roca 2019) and is usually attributed to relative humidity-limited conditions at a warm surface temperature that decreases the intensity of extreme precipitation over these semi-arid areas. The decrease in the wet-day duration for this regime has also been identified as a key mechanism under mid-latitude land conditions (Utsumi et al. 2011). Further analysis is required to determine whether these very high SSTs are associated with large-scale subsidence and simultaneous relatively dry boundary layer that could impact the intensity and/or the duration of the precipitation events as well as explain the decreasing trend. Burdanowitz et al. (2019) do not report any such decreasing regime when analyzing instantaneous disdrometer-derived precipitation rate from ship data. They attribute this departure from the continental extreme behavior to the absence of decreasing event duration in this regime. However, their analysis is restricted to 1 K SST bin up to 303 K and hence would miss the warm regime identified here. Furthermore, our study demonstrates that the use of simultaneous SST is detrimental to the scaling computation. It is likely that using a time-lagged analysis of the instantaneous precipitation scaling over an extended SST range would allow a better delineation of the various regimes relevant for that instantaneous scale.
The third regime, which we call the “Clausius–Clapeyron” regime, ranges from 300 K to 302.5 K and includes almost 56 % of the total precipitation amount. It is characterized by a steady increase in the extremes with surface temperature. This regime is robust to the precipitation product, the SST product, and the length of the record. When simultaneous SST data are used, the scaling is diminished, but for the OIRSS where the scaling simply does not exist. The results of the 1-day lagged scaling are robust to the SST product selection and lead to a scaling of around 5.17 ± 0.85 % K−1. The 2-day lagged scaling value ranges from 5.67 ± 0.89 % K−1 for the 2001–2017 period to 6.33 ± 0.81 % K−1 for the 2012–2016 period for all the SST and precipitation products considered here. While the actual direct validation or evaluation of the representation of intense rain accumulation in these products remains challenging, their common and physically sound behavior indicates that the products are suitable for the exploration of the extreme precipitation over the ocean. The robustness of the scaling analysis across the satellite precipitation products is further very close to the theoretical expectation for the thermodynamic scaling of ∼ 6 % K−1. This gives us even more confidence in this generation of satellite precipitation products at the 1° 1-day resolution over the ocean. The occurrence frequency of the precipitation greater than the 99.9th percentile and the surface conditions corresponding to the Clausius–Clapeyron regime have been mapped in Fig. 7. The figure demonstrates that this regime is not associated with the climatological ITCZ location as previously noted (Masunaga et al. 2019). However, it is associated with the known climatological rainfall maxima, which is off the Colombian Coast and in the northern Bay of Bengal and also seems to align with the climatological distribution of cyclone occurrence in the East Pacific, West Atlantic, South Indian Ocean, and China Sea.
Map of the ensemble mean frequency of occurrence (%) of precipitation greater than the percentile 99.9th for the CC SST regime [300 K, 302.5 K]. SST from OSTIA over the period 2007–2017, lagged by 2 days, are used to delineate the regime.
An avenue for further research lies in the identification of the contribution of the organized convection to the scaling physics (Pendergrass 2020; Muller and Takayabu 2020; Roca and Fiolleau 2020). Our results further prompt for a dedicated investigation of the contribution of the cyclone precipitation to the scaling physics that is so far not promoted much. Our results also suggest that community efforts, possibly under the umbrella of the International Precipitation Working Group (Levizzani et al. 2018) and the GEWEX/WCRP core project, are needed to further characterize the absolute accuracy of precipitation products over ocean and elucidate which clusters of products are to be understood as a reference, if any.
This study benefited from the IPSL mesocenter ESPRI facility, which is supported by CNRS, UPMC, Labex L-IPSL, CNES, and Ecole Polytechnique. The authors acknowledge the CNES and CNRS support under the Megha-Tropiques program. They also thank Dr T. Fiolleau for helpful discussions and Dr. M. Schröder for his help with the HOAPS dataset.
