The tropical Indian Ocean matters for U. S. winter precipitation variability and predictability

The El Niño-Southern Oscillation (ENSO) is the key predictor for operational seasonal climate prediction in the United States (U. S.). Compared with the impact of the tropical Pacific associated with ENSO, the role of the Indian Ocean on U. S. climate variability and predictability is less documented. In this work, we noted that the impact of the tropical Indian Ocean is stronger than the tropical Pacific on winter precipitation variability in a part of the southeastern contiguous U. S. (CONUS), mainly including Kentucky, Tennessee, Mississippi, and Alabama. Different from the north-south contrastive impact of ENSO, the influence of the Indian Ocean is confined to the southeastern CONUS. Basin-wide warming (cooling) in the tropical Indian Ocean is tied to above (below) normal winter precipitation in the southeastern CONUS. The observed relationship is reproduced in model forecasts and simulations. Physically, Indian Ocean heating anomaly communicates its influence by inducing a teleconnection from the Indian Ocean to the North Atlantic Ocean via the North Pacific. The connection provides an additional source of predictability of the winter precipitation in CONUS, and monitoring the heat condition in the Indian Ocean may benefit winter precipitation prediction in the southeastern CONUS.


Introduction
As the strongest interannual variability in the tropical oceans, El Niño-Southern Oscillation (ENSO) is the most important source of global climate predictability on seasonal-to-interannual time scales (National Research Council 2010. The United States (U. S.) is one of the regions significantly affected by ENSO (e.g. Ropelewski and Halpert 1986, Leathers et al 1991, Zhu et al 2013, L'Heureux et al 2015, Peng et al 2018. The impact of ENSO on the U. S. climate has been well documented which is through a Pacific-North American-like teleconnection pattern (Wallace and Gutzler 1981, Straus and Shukla 2002. ENSO has been the key predictor for the operational seasonal climate prediction over the U. S. (O'Lenic et al 2008).
Despite being significantly influenced by ENSO, the overall skill of operational seasonal predictions in the U. S. is low (O'Lenic et al 2008, Peng et al 2012. For example, the U. S. Climate Prediction Center official seasonal precipitation forecast skill (Heidke score) is 0.025 for the average in 1995-2009 (Peng et al 2013). Although it has been documented that the influence of the atmospheric internal variability may limit seasonal predictability due to ENSO (Kumar et al 2007, Jha et al 2019, influences from boundary forcing from the other oceans, such as the Indian Ocean (Sun et al 2019), may play a role in the seasonal climate predictability in the contiguous United States (CONUS). Such a hypothesis is supported by previous works. For example, Bader and Latif (2005) pointed out that anomalous warming/cooling in the Indian Ocean affects atmospheric circulation in the North American and North Atlantic sectors with the South Asian jet acting as the waveguide. These results imply a potential that U. S. climate variability may be affected by the anomalous heating in the Indian Ocean. That is the focus of this work.
In this work, we first examine the statistical connection of winter precipitation in CONUS with the Indian Ocean sea surface temperature (SST) anomaly (SSTA) and compare it with the corresponding connection with ENSO in observations. Results based on observations are verified with forecasts from a state-of-the-art climate model and Atmospheric Model Intercomparison Project (AMIP)-like simulations forced by observed SST in the tropical Indian Ocean. Establishing such a connection implies a predictability source that may be useful in improving seasonal forecasts and in model validation.

Data and methods
Monthly mean SSTs on a 1 • × 1 • grid are from Optimum Interpolation v2.1 (OIv2.1) SST since September 1981 (Huang et al 2021). The Niño3.4 index is defined as the averaged SSTA in the region (5 • S-5 • N, 170 • W-120 • W) to represent ENSO. SSTA averaged in (20 • S-20 • N, 40 • -110 • E) is used to represent the Indian Ocean basin mode (IOBM), while the difference of averaged SSTAs between the western (10 • S-10 • N, 50 • -70 • E) and eastern (10 • S-0 • , 90 • -110 • E) Indian Ocean is used to represent the Indian Ocean Dipole (IOD; Saji et al 1999). IOBM and IOD are the first and second leading modes of monthly SSTAs in the tropical Indian Ocean, respectively (e.g. Prajeesh et al 2022), thus, they represent a large fraction of SST variability in the tropical Indian Ocean.
Monthly mean geopotential heights at 200 hPa (H200) and the meridional component of wind at 200 hPa (v200) are derived from the fifth-generation atmospheric reanalysis of the global climate of the European Centre for Medium-Range Weather Forecasts (ERA5) at a 2.5 • × 2.5 • resolution for the period January 1959-December 2022 (Hersbach et al 2020). Monthly mean precipitations are from a global satellite-rain gauge merged product at a 2.5 • × 2.5 • spatial resolution since January 1979 (Janowiak and Xie 1999), which are the blended analysis with monthly rain gauge totals from the Climate Anomaly Monitoring System (CAMS) and satellitebased estimates from outgoing longwave radiation (OLR) anomalies that are generated by the OLR precipitation index. OLR data on a 2.5 • × 2.5 • grid since January 1974 are from Liebmann and Smith (1996).
As an additional validation of the connection between the Indian Ocean and winter precipitation in the U. S., we examine the retrospective seasonal predictions (or hindcasts) for January 1982- To further verify the impact of Indian Ocean SST on the predictability and variability of U. S. winter precipitation, AMIP-like experiments are analyzed. In the AMIP experiments, the atmospheric general circulation model is the atmospheric component (Global Forecast System) of CFSv2 at a T126 resolution (Saha et al 2014). The experiments are forced by observed SST in the tropical Indian Ocean (25 • S-5 • N, 38 • -100 • E) and climatological SST elsewhere, and it is referred to as IOGA. Sponge-like smoothing is applied near the boundary of the Indian Ocean to avoid extreme gradient (see the red contours in figure 5). The SST data from 1981 to 2008 are from HadISST and afterward from OIv2 SST which is available in real-time for climate attribution operation purpose. The experiments are from January 1981 to December 2022 with 18 ensemble members, each starting from different atmospheric ICs (Hu et al 2020a). From supplementary figure S1(a), we can see that the domain used in IOGA (the green dashed line rectangle) and the domain used to define the IOBM index (the blue line rectangle) largely overlap. The correlation between the IOBM index and SSTA averaged in the IOGA-specified observed SST region is 0.96 (figure S1(b)). Thus, with the domain used in IOGA, the variability of the IOBM index is captured.
To isolate the connection of precipitation (SST, OLR, v200) variation with the tropical Indian Ocean without the influence of ENSO, or the connection of a precipitation index with ENSO without the influence of IOBM, and the connection of the precipitation index with IOBM without the influence of ENSO, partial correlations are analyzed (Pedhazur 1997). The partial correlation between A and B adjusted for C (to eliminate C's influence) is: where r AB , r AC , and r BC are the correlations between A and B, A and C, and B and C, respectively.
All the anomalies are relative to the climatologies of 1991-2020. All data in DJF 1981/82-2021/22 are detrended.

Observational diagnoses
It is well known that SST variability in the Indian Ocean is influenced by ENSO (e.g. Wang et al 2019). Statistically, the correlation between the IOD and Niño3.4 indices is largest when the IOD index leads the Niño3.4 index by 1-3 months, while the correlation between the IOBM and Niño3.4 indices is largest when the IOBM index lags the Niño3.4 index by 3-5 months (figure S2). To examine the impact of SSTAs in the Indian Ocean without the influence of ENSO on winter precipitation in CONUS, the partial correlation approach is applied. Figure 1  shows the simultaneous correlations of winter (DJF) precipitation anomalies with (a) the Niño3.4 index, and the simultaneous partial correlations of winter precipitation anomalies with the (b) IOBM and (c) IOD indices adjusted for the Niño3.4 index. Overall, in CONUS, the area with significant correlations is the largest for the correlations with the Niño3.4 index, the smallest for the partial correlations with the IOD index, and in between for the partial correlations with the IOBM index. In addition, the spatial patterns of the correlations are also different. For the correlations with the Niño3.4 index, it is a north-south pattern with strong and broad-area of significant positive correlations in the southern and central CONUS, and some areas with weaker negative correlations in the northern Great Plains ( figure 1(a)). Similar patterns have been documented in previous works (e.g. Ropelewski and Halpert 1986, Hu and Huang 2009. For the partial correlations with the IOBM index, it is largely a west-east dipole pattern with significant positive correlations in a part of the southeastern CONUS, mainly including Kentucky, Tennessee, Mississippi, and Alabama, and some minor negative correlations in the western CONUS ( figure 1(b)). The partial correlations with the IOD index (figure 1(c)) are mostly not significant. However, Saji and Yamagta (2003) noted that the significant influence of IOD in U. S. precipitation was confined to the western portion and in summer only. Thus, the impact of IOD on U. S. precipitation may vary with location and season. These results suggest that winter precipitation of CONUS may be affected by both ENSO and Indian Ocean basin-wide warming or cooling. Since the impact of ENSO has been well documented and the connection of the winter precipitation in CONUS with IOD is mostly not significant, we will focus on the impact of IOBM on the winter precipitation in CONUS in the following discussion.
The correlations of the southeastern CONUS precipitation with the Niño3.4 and IOBM indices in DJF 1981/82-2021/22 are compared next (figure 2). Here, the precipitation variation in the southeastern CONUS is represented by an index which is the precipitation anomaly averaged in (32 • -42 • N, 80 • -95 • W; the green rectangle in figures 1(b) and 3(a)). The correlation between the southeastern CONUS precipitation and IOBM indices ( figure 2(b)) is 0.54 and is significant at the level of 99.9%, while the correlation between the southeastern CONUS precipitation and Niño3.4 indices (figure 2(a)) is 0.37 and is significant at the level of 98%. Interestingly, the partial correlation between the IOBM and precipitation indices adjusted for the Niño3.4 index is 0.43 and the partial correlation between the Niño3.4 and precipitation indices adjusted for the IOBM index is −0.06. Thus, for the winter precipitation in the southeastern CONUS, the correlation is larger and more significant with the IOBM index than with the Niño3.4 index. Furthermore, in addition to the same sign in the extreme IOBM years in the 1997/98 and 2015/16 winters, there are 26 winters having the same sign of the IOBM and precipitation indices with 13 winters having the opposite signs, implying that the connection between the southeastern CONUS precipitation and IOBM indices is present in both the extreme and no-extreme IOBM winters.
To further analyze this relationship, the correlations between the southeastern CONUS precipitation index and SST and H200 anomalies are examined ( figure 3(a)). Significant positive correlations with SST are present in both the tropical Indian Ocean and the central and eastern tropical Pacific Ocean (shading in figure 3(a)). Interestingly, the positive correlations are larger in the tropical Indian Ocean than in the central and eastern tropical Pacific Ocean, implying a closer connection between the winter precipitation variation in the southeastern CONUS and tropical Indian Ocean basin-wide SSTAs, consistent with the results in figure 2. The influence of the tropical Indian Ocean on winter precipitation in the southeastern CONUS is supported by the similarity of the correlations of H200 anomalies with the southeastern CONUS precipitation index (contours in figure 3(a)) with the partial correlations of H200 anomalies with the IOBM index adjusted for the Niño3.4 index (contours in figure 3(b)). For example, over the CONUS, the correlations show that both above-normal winter precipitation in the southeastern CONUS and abovenormal SST in the tropical Indian Ocean are associated with positive H200 anomalies in the southeastern CONUS and negative H200 anomalies in the western CONUS. Meanwhile, the correlation pattern also shows some similarities over the Indian Ocean, East Asia, and the North Pacific Ocean. With warming (cooling) in the Indian Ocean, the positive (negative) H200 anomalies are generated in the southeastern CONUS and the negative (positive) ones in the western CONUS, leading to the convergence (divergence) and above (below) -normal precipitation in the southeastern CONUS (figures 3(a) and (b)).
Physically, the heating anomalies over the tropical oceans associated with ENSO or IOBM can generate anomalies in the extratropics via Rossby wave propagation (e.g. Wallace and Gutzler 1981, Jin and Hoskins 1995, Hoerling et al 1997. Here, we follow Jin and Hoskins (1995) using zonal departure v200 to track the extratropical response to heating in the tropical Indian Ocean. The partial correlations with the IOBM index adjusted for the Niño3.4 index suggest that basin-wide warming (cooling) leads to enhanced (suppressed) convection in the western and central tropical Indian Ocean (represented by OLR, shading in figure 3(c)) which produces a teleconnection from the tropical Indian Ocean to CONUS via East Asia and the North Pacific (represented by zonal departure v200, contours in figure 3(c)). Such a teleconnection is similar to Bader and Latif (2005). They argued that basin-wide warming or cooling in the Indian Ocean can excite a teleconnection pattern from the Indian Ocean to the North Atlantic Ocean via the North Pacific and North America, and modulates North Atlantic oscillation (NAO).
The observational analysis suggests that for the winter precipitation anomalies in the southeastern CONUS, Indian Ocean basin-wide warming or cooling plays a more important role than ENSO. Indian Ocean heating anomaly influences the winter precipitation anomaly in the southeastern CONUS by exciting a teleconnection from the Indian Ocean to the North Atlantic Ocean via the North Pacific and North America.

Evidence from model forecasts and simulations
Due to the short length of the observational record, the robustness of statistical relation needs to be verified with additional independent data, such as from model forecasts and simulations. Here, we use CFSv2 forecasts and AMIP simulations to validate the observed connections between SSTAs and the winter precipitation in CONUS.
Shown in figure 4 is the simultaneous correlation of the southeastern CONUS precipitation index with global SSTAs in different lead-month forecasts from CFSv2. The correlations are computed based on the 20-member ensemble mean, and therefore, highlight the variability in rainfall response to SST. The observed statistical connection between the precipitation anomalies in the southeastern CONUS and the tropical Pacific SSTAs is reproduced in CFSv2. In the connection with tropical Indian Ocean SSTA, although positive correlations are present but they are less significant than that in the observations ( figure 3(b)), except for the 6 month lead forecasts. Interestingly, the significance of the correlations in the tropical Indian Ocean increases with lead-time increase. We speculate that may be associated with the relatively small sample size as well as the model's IC shock.
To further identify the roles of SSTA in the Indian Ocean on the winter precipitation in CONUS, figure 5 shows simultaneous correlations of IOGA simulated land precipitation (shading) and H200 (contour) anomalies with observed IOBM index in DJF 1981/82-2021/22. Similar to the observations (figures 3(a) and (b)), the positive (negative) phase of IOBM in the Indian Ocean is associated with a teleconnection from the Indian Ocean to the North Atlantic Ocean via the North Pacific, favoring above (below) normal precipitations in southeastern CONUS. That is consistent with the signal-to-noise ratios (SNRs) of the winter precipitation in CONUS in the IOGA experiment ( figure S3). Here, SNR is defined as the ratio of the standard deviations of the ensemble mean anomaly to the standard deviation of the departure of each ensemble member from its ensemble mean. Larger SNR corresponds to higher potential predictability (Kumar and Hoerling 1995, Scaife and Smith 2018, Jha et al 2019. It is noted that SNRs with values larger than 0.1 are present in the southeastern CONUS, generally collocated with significant positive partial correlations between the southeastern CONUS precipitation and the IOBM index adjusted for the Niño3.4 index in the observation (figures 1(b) and 2(b)) and positive correlations between the southeastern CONUS precipitation and the IOBM index in the IOGA simulations (shading in figure 5). According to Kumar and Hoerling (2000;   Thus, both the CFSv2 forecasts and AMIP-like simulations confirmed the observed connection of tropical Indian basin-wide warming or cooling with winter precipitation in the southeastern CONUS. Here, the overall small SNR values are consistent with low predictability of U. S. precipitation (Peng et al 2012(Peng et al , 2013. Consistently, Hu et al 2020a argued that the pattern correlations of monthly mean land precipitation anomalies between observations and ensemble mean of AMIP simulations forced by observed global SST are around 0.1, implying that only about 1% of monthly land precipitation variance seems forced by SST (see their figures 6 and 7).

Summary and discussion
The impacts of the tropical Pacific SSTA associated with ENSO on the interannual variability of U. S. climate have been well documented and are the major source of the predictability of climate variability in the region. Compared with the impact of the tropical Pacific SSTA, other ocean basins' impact on U. S. climate variability and predictability is less known. In this work, through observational diagnoses and model forecasts and simulations, we proposed the role of the tropical Indian Ocean in the winter precipitation variation in CONUS. The connection may provide an additional source of predictability for the winter precipitation variation in CONUS.
The impact of the tropical Pacific SSTA associated with ENSO mainly leads to opposite winter precipitation anomalies between the southern and northern tiers of CONUS, while the influence of Indian Ocean basin-wide warming or cooling is mainly in a part of the southeastern CONUS. Basinwide warming (cooling) in the tropical Indian Ocean is tied to above-(below-) normal winter precipitation in the southeastern CONUS, mainly including Kentucky, Tennessee, Mississippi, and Alabama. Statistically, compared with the impact of the tropical Pacific Ocean, Indian Ocean basin-wide warming or cooling plays a more important role in the winter precipitation variation in the southeastern CONUS. However, the impact of the dipole-like SSTA in the tropical Indian Ocean on the winter precipitation variation in the southeastern CONUS is not significant. The observed connection between winter precipitation in the southeastern CONUS and the tropical Indian basin-wide warming or cooling is reproduced in model forecasts and AMIP-like simulations. Physically, Indian Ocean heat anomaly carries on its influence on the winter precipitation anomaly in the southeastern CONUS through exciting a teleconnection from the Indian Ocean to the North Atlantic Ocean via the North Pacific and North America. Such a connection means another source of predictability of the winter precipitation in CONUS in addition to that of the tropical Pacific SSTA associated with ENSO. Interestingly, we note that the connection between the IOBM index and southeastern CONUS precipitation is only present in winter, and there are no significant (partial) correlations in the other seasons ( figure S4).
The IOGA simulations were forced by spatially varied SST observations in the domain. However, it is unclear for the key region of the Indian Ocean SSTAs in affecting the U. S. winter precipitation variations. For instance, what are the possible impacts of the Arabian Sea and the Bay of Bengal? These two key regions of the Indian Ocean with extremely strong deep convections were not included in the IOGA simulations (green box in figure S1). Also, this work focuses on the connection between interannual variations of IOBM and southeastern CONUS winter precipitation. However, the pronounced warming tendency and interdecadal variations in the Indian Ocean (e.g. Huang et al 2022) and interdecadal shift of ENSO (e.g. Hu et al 2020b) as well as the Pacific decadal oscillation (e.g. Hu and Huang 2009) and the Atlantic multi-decadal oscillation (e.g. Kerr 2000) may modulate the interannual connection. Moreover, in addition to ENSO and the tropical Indian Ocean, the interannual variations of winter precipitation in the middle and high latitudes of the northern hemisphere (including CONUS) may be affected by other factors, such as NAO, Arctic Oscillation, and Arctic stratospheric polar vortex (Hardiman et al 2020, Juzbašić et al 2020. Lastly, since partial correlation only accounts for linear relationships and represents the first-order approximation of individual ocean basins' impacts on U. S. precipitation variability. The tropical Pacific and Indian Ocean might interact in a more complex nonlinear manner, which is not considered here.