The US crop data used in this study includes harvest area, production, yield time series, and crop calendars, covering a 42 year period from 1978 to 2019. State-level soybean harvest data was obtained from the USDA's NASS Quick Stats database (https://quickstats.nass.usda.gov/, last access: 1 May 2024), while planting and harvest dates at a 0.5 min grid resolution were sourced from Sacks et al. (Sacks et al., 2010). Across the US, the primary soybean growing season typically spans May to November.

The annual US soybean yield (Yt) was calculated by aggregating yields across US states, weighted by harvested area. The formula is as follows: 1Yt=∑Yt,s×Hat,s∑Hat,s Where t and s denote the year and political unit, respectively, Yt,s and Ha_t,s represent the yield (tha-1) and harvested area (ha).

To eliminate non-climatic influences on yield (Anderson et al., 2017a; Iizumi et al., 2014), we calculated expected yields (YEXt,s) by applying a 5 year running mean for each state. The percent yield anomaly (ΔYt,s) relative to expected yield was then computed as follows: 2ΔYt,s=Yt,s-YEXt,sYEXt,s×100 Where t and s again represent the year and political unit, respectively. Yt,s represents the yield value (tha-1). Due to the 5 year running mean (averaging from t-2 to t+2), anomalies could not be calculated for 1978, 1979, 2018, and 2019.

To capture the variability of the tropical Indian Ocean SST, we used monthly mean SST grid data at a 1.0°×1.0° resolution from the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST, Rayner et al., 2003) spanning 1979–2017. The IOB index was defined as the average SST anomaly (SSTA) in the region 20° S–20° N, 40–110° E, while the Niño 3.4 index was calculated as the SSTA averaged over 5° S–5° N, 170–120° W. Before calculating these indices, SST fields were standardized by multiplying each value by the square root of the cosine of latitude.

To assess the impact of meteorological factors on soybean yields in the United States, we selected ten key variables from the Climatic Research Unit (CRU) TS v4.07 dataset and the ERA5 reanalysis (Harris et al., 2020; Hersbach et al., 2020). These include temperature [maximum (Tmx, °C), mean (Tmp, °C), minimum (Tmn, °C), diurnal temperature range (DTR, °C)], precipitation (Precip, mmd-1), wet day frequency (Wet, days), cloud cover (Cld, %), downward shortwave radiation flux (DSRF, Wm-2), root-zone soil moisture (SMroot, m3m-3; Layer 2, 7–28 cm depth), and vapor pressure deficit (VPD, hPa). Eight of these variables were obtained from CRU, which provides monthly mean gridded data at 0.25°×0.25° resolution, while SMroot was obtained from ERA5 as a proxy for soybean root water uptake. The choice of variables is consistent with previous studies highlighting the role of temperature, precipitation, radiation, soil moisture, and humidity in soybean yields (Gaupp et al., 2020; Gobin and Van de Vyver, 2021; Hamed et al., 2021; Joshi et al., 2021; Leng and Hall, 2019; Ray et al., 2015; Schauberger et al., 2017), with VPD included as an additional dryness indicator (Ergo et al., 2018). VPD was calculated using the following formulas: 3e0=6.108exp⁡17.27×TmpTmp+237.34VPD=e0-ea where Tmp is the monthly average temperature (°C), and ea is the average actual vapor pressure (hPa), both from the CRU dataset. e0 represents the monthly mean saturated vapor pressure (hPa). A summary of all variables and references, including units, is provided in Table S1 in the Supplement.

To investigate how the IOB mode influences key meteorological factors, we selected monthly values for mean sea level pressure (SLP, Pa), geopotential height at 200 hPa (GPH200, m), and wind components at 925 hPa (meridional, v925 in ms-1, and zonal, u925 in ms-1). All variables were obtained from the ERA5 reanalysis dataset (Hersbach et al., 2020).

Before conducting the specific analyses, we employed Gram-Schmidt orthogonalization to remove the linear influence of ENSO (represented by Niño3.4) from the IOB index, other climate indices, meteorological factors, and large-scale circulation fields. This method transforms correlated variables into orthogonal sets by sequentially projecting each target variable onto the space orthogonal to ENSO. The ENSO-independent component of a variable X was calculated as: 5X⊥E=X-〈X,E〉〈E,E〉E here X is the original variable, E the ENSO signal, and 〈⋅,⋅〉 denotes the inner product. Through this procedure, only the variability linearly independent of ENSO is retained, enabling a clearer attribution of Indian Ocean-related effects. We note that while this approach ensures zero-lag statistical independence from ENSO, lead-lag influences cannot be fully eliminated, as ENSO and Indian Ocean warming often co-evolve and interact across seasons. Similar approaches have been applied in recent climate studies (Hou et al., 2024).

We began by conducting a correlation analysis between multiple climate index time series and US soybean yield data. In this step, we systematically adjusted the time range of each climate index to examine its lagged impact on soybean yield, which allowed us to identify the climate modes with the most significant influence. Next, we applied ridge regression to analyze the relationship between ten meteorological factors and soybean yield, isolating the three factors with the greatest impact on yield outcomes. For clarity, in this study we define meteorological factors as local surface climate variables that directly affect crop growth (e.g., Tmx, Precip, SMroot, DSRF, and VPD). In contrast, we define atmospheric circulation patterns as large-scale circulation fields that characterize regional and hemispheric variability, including SLP, GPH200, and 925 hPa winds. We then divided the regression analysis into two components: (1) assessing the relationship between climate modes, key meteorological factors, and atmospheric circulation patterns, and (2) examining the link between key meteorological factors and soybean yield. This framework enabled us to construct an impact chain that connects climate modes, large-scale circulation features, local weather conditions, and ultimately soybean yield outcomes.

Many studies have focused on the climate variability of the Pacific and Atlantic Oceans, including phenomena such as ENSO and Tropical Atlantic variability (TAV), which will affect the weather conditions in the United States, resulting in fluctuations in soybean production (Anderson et al., 2019; Qian et al., 2020). However, in addition to these well-studied oceanic drivers, other climatic modes may also exert significant impacts on soybean yields. To explore this further, we conducted a comprehensive analysis of the correlation between soybean yields in key production regions and a broad range of climate indices. These indices include Niño3, Niño4, Niño3.4, Pacific North American Index (PNA), Pacific Decadal Oscillation (PDO), Southern Oscillation Index (SOI), AMO, Atlantic Meridional Mode (AMM), NAO, Tropical South Atlantic (TSA), Tropical North Atlantic (TNA), Indian Ocean Dipole (IOD), and IOB, among others. See Table S2 in the Supplement for sources of these indices.

As shown in Fig. 1, during the soybean growth period (MAM-JAS), Niño3.4 and Niño4, which are associated with the ENSO phenomenon, exhibit a statistically significant correlation with US soybean yields at the 95 % confidence level based on a two-tailed t test. This finding aligns with previous studies, which have well-documented the influence of ENSO on soybean production (Anderson et al., 2017b; Hamed et al., 2023).

However, our analysis extends beyond the growing season to consider climate variability during the winter months prior to soybean planting. We find that the IOB mode, which captures large-scale warming or cooling in the Indian Ocean, plays a substantial role in determining soybean yields before sowing. The year-to-year anomalies in soybean yield exhibit the strongest Pearson correlation (-0.41) with the IOB index during ND(-1)J (November and December of the year preceding harvest and January of the harvest year), which was identified as the optimal 3 month window following an exhaustive correlation screening across all possible periods; this relationship is statistically significant at the 99 % confidence level (two-tailed t test; Fig. 1). In a simple linear regression framework, this corresponds to a coefficient of determination of R2=0.1681(≈16.8%), indicating that approximately 16.8 % of the interannual variability in soybean yield is explained by the IOB mode during this pre-growing season window.

Figure 2a illustrates the spatial distribution of soybean production in the United States. It is evident that soybean production is concentrated in the central United States, with Illinois, Iowa, Minnesota, Indiana, and Nebraska having the highest yields, accounting for more than 50 % of the total soybean production in the country.

To explore how soybean yields respond to variations in the IOB mode across different US regions, we conducted a regression analysis. The findings reveal that the IOB mode during ND(-1)J exerts a significant negative influence on soybean yields, with an inverse relationship between the IOB index and yields observed in most states (Fig. 2b). This negative correlation was especially pronounced in several key agricultural states. Notably, South Dakota, Kansas, Missouri, Illinois, Michigan, Indiana, Ohio, Kentucky, Tennessee, and New York displayed significant sensitivities to fluctuations in the IOB mode. The relationship between the IOB index and soybean yields in these regions passed the 90 % significance threshold, indicating that the impact of the IOB mode on soybean production in these states is both statistically robust and consistent.

To quantify the overall impact of IOB fluctuations on US soybean production, we calculated a weighted sum of the regression coefficients, with the weights based on the average harvested area and expected yields across individual regions. This approach provides a more comprehensive assessment of the IOB mode's effect on national production. The findings indicate that an increase in the IOB index is associated with a decrease in soybean production in nearly all regions of the United States (Fig. 2c). Specifically, a one standard deviation increase in the IOB index during ND(-1)J corresponds to a yield reduction of approximately 4.7 million tonnes in regions experiencing significant declines, while regions with notable yield increases were virtually nonexistent. Among these regions, Illinois saw the largest decline, with an estimated loss of around 800 000 t. In aggregate, an increase of one standard deviation in the IOB index during ND(-1)J corresponds to a reduction of approximately 4.0 % in total soybean production across the United States.

To identify the key factors influencing soybean yields in the United States, we considered ten climate indicators (Tmp, Tmx, Tmn, DTR, DSRF, Precip, VPD, Cld, wet, and SMroot). Given the interdependence among these predictors, ridge regression was employed to stabilize coefficient estimates by penalizing multicollinearity. The regression results indicate that diurnal temperature range (DTR) during the reproductive stage (JAS) exerts the largest influence on yield anomalies, followed by root-zone soil moisture (SMroot) and maximum temperature (Tmx) (Table S3 in the Supplement).

To more precisely assess the influence of these meteorological factors on soybean yields in the United States, we conducted a regression analysis examining their relationships with soybean yield across various states. Figure 3a demonstrates that an increase in the DTR is consistently associated with a decrease in soybean yields across nearly all regions of the United States. Statistically significant reductions in yields are observed in approximately 58 % of the regions (P<0.1), indicating that large temperature swings between day and night negatively affect soybean productivity. Figure 3b indicates a positive correlation between SMroot and soybean yields in most areas, particularly in the Midwest and Southeast, where significant positive correlations (P<0.1) are evident. We also find that soybean yields are significantly impacted by DTR in areas similar to those affected by SMroot, suggesting that the effects of DTR are particularly pronounced in regions with limited water supply. Figure 3c illustrates that higher maximum temperatures (Tmx) are linked to decreased soybean yields in the southern and central United States, with significant negative correlations observed in these areas (P<0.1). Notably, the spatial patterns of DTR and Tmx closely resemble each other, both showing an inverse relationship with SMroot.

During ND(-1)J, an increase in the IOB index results in a widespread reduction in SMroot across the United States. Significant soil moisture anomalies (P<0.1) are particularly evident in North Dakota, South Dakota, Minnesota, Illinois, Georgia, South Carolina, North Carolina, and parts of Nebraska, Kansas, Iowa, Missouri, and Indiana (Fig. 4e). This decrease in SMroot appears to be associated with anomalies in Precip and VPD, as regions experiencing reduced SMroot generally align with areas of increased VPD and decreased Precip.

As illustrated in Fig. 4a, the warming signals from the Indian Ocean can affect the United States across the Pacific. The warming of the Indian Ocean enhances convective activities and upward motion, leading to the development of negative outgoing longwave radiation (OLR) anomalies. These anomalies generate Rossby waves that propagate from the tropical Indian Ocean to the United States via East Asia and the North Pacific (Hu et al., 2023; Ratnam et al., 2012). The response pattern of 200 hPa geopotential heights consists of a low-high system over the United States (Fig. 4a). This leads to southeasterly wind anomalies, which bring warmer and wetter equatorial air to the United States, causing widespread temperature increases in the United States and more precipitation in the southeastern United States (Fig. 4b and c). However, the southeasterly winds that bring warm and moist air shift to southwesterly winds in the northern United States (Fig. 4a), and the northeasterly wind anomaly transfers drier air from the polar regions to the Northern US (Wang et al., 2013). Concurrently, the north-central United States becomes a water vapor divergence area (Fig. S3 in the Supplement), reducing Precip and contributing to decreased SMroot in that area (Fig. 4b and e). In the southeastern United States, although Precip increases, the concurrent rise in Tmp intensifies VPD, contributing to the observed decrease in SMroot (Fig. 4c–e). Elevated Tmp could amplify VPD by accelerating soil and plant moisture loss, as shown in previous studies (Yin et al., 2014; Zhao et al., 2023).

For the root zone, soil moisture memory ranges from several days to up to a year and shows a long memory for dry soil moisture regimes (Stacke and Hagemann, 2016). In this study, we found that soil moisture anomalies during ND(-1)J are significantly positively correlated with soil moisture anomalies in JAS (Fig. S3). Specifically, the strongest positive correlations are observed over the US Midwest and Northern Plains, which represent the core soybean production regions, indicating that early-season soil moisture conditions in these areas persist into the summer. This “memory effect” suggests that reductions in ND(-1)J soil moisture can influence root-zone moisture during the reproductive stage, thereby affecting soybean yields and highlighting the role of early-season hydrological conditions in modulating summer drought risk.

The warming of the Indian Ocean during the ND(-1)J could indirectly influence JAS meteorological conditions in the United States by affecting Pacific SSTs. It is associated with a Matsuno-Gill-type response, which excites atmospheric Kelvin waves propagating eastward (Gill, 1980; Matsuno, 1966). This results in easterly wind anomalies over the western-central equatorial Pacific. Strengthened easterly winds lead to a shallower thermocline, which brings colder waters closer to the surface, contributing to the cooling of SSTs in the eastern Pacific. This cooling effect can persist throughout the summer months, as illustrated in Fig. 5a, and is a key component of the development of La Niña events (Cai et al., 2019; Wang, 2019; Wu and Kirtman, 2004; Xie et al., 2016). During La Niña events, there are significant atmospheric circulation anomalies that further impact the United States. For example, Fig. 5b shows that during La Niña years, 200 hPa geopotential height anomalies are predominantly positive in the mid-latitudes, while negative anomalies occur in the tropics and certain high-latitude regions (Luo and Lau, 2020). These geopotential height anomalies reflect the establishment of an anomalous high-pressure system over much of the United States.

Such conditions lead to changes in meteorological patterns across the United States, including warmer temperature anomalies and decreased precipitation (Jia et al., 2016; Jong et al., 2020, 2021; Mo et al., 2009; Wang et al., 2007). Consequently, substantial reductions in Precip and increases in Tmx are shown in Fig. 6a and b. The elevated Tmx exacerbates evapotranspiration, driving higher moisture loss from both the soil and the plants. This warming effect intensifies the drought-like conditions across several regions of the United States. Additionally, a noticeable rise in the VPD (Fig. 6c) further compounds the stress on crops by increasing the demand for moisture in the atmosphere. This heightened atmospheric moisture deficit, combined with reduced precipitation, directly leads to declining SMroot levels (Fig. 6d). These patterns indicate a clear decline in soil water availability, heightening the risk of drought conditions across affected regions. Moreover, Fig. 6f reveals a significant increase in the DTR across soybean-growing regions. This increase is likely associated with a reduction in Cld (Fig. 6e), which allows for more extreme temperature variations between day and night. The pattern observed in this study is similar to those examined by Doan et al. (2022).

These results indicate that the variability of the IOB can affect the meteorological elements in the United States by influencing the development of ENSO. The hot and dry conditions resulting from these meteorological changes are particularly detrimental to soybean yields, as highlighted by Anderson et al. (2017a) and Hamed et al. (2023). The combined impacts of increased Tmx and DTR, decreased Precip, elevated VPD, and reduced SMroot create an environment highly unfavorable for soybean production. This highlights the critical role of these climate drivers in influencing agricultural outcomes under positive IOB conditions (Fig. S5 in the Supplement).