Main

Central Asia suffered from an extreme agricultural drought in its early growing season in 2021. Crops withered due to insufficient water supplies, and thousands of sheep, cows and horses had died because of the lack of water and forage. The harsh drought of 2021 is not an independent event but a part of a dryer trend over the past half century1,2,3. Soil moisture has been used as an indicator of agricultural and ecological drought4. The loss of soil moisture has accelerated in recent decades, which has intensified desertification in this arid region and led to a more fragile ecosystem5,6,7. For water resource-limited environments such as Central Asia, the reduction in soil moisture is closely related to the changes in regional precipitation and temperature-based evapotranspiration8,9,10. While the record-breaking heat and severe lack of precipitation are blamed for the parched land in 2021, the ongoing worsening droughts are accompanied by increased evapotranspiration due to high temperature and decreased precipitation5,11,12. Attribution studies have found that the rapid warming trend in the past half century over Central Asia is mainly forced by anthropogenic activities13. The observed changes in Central Asian precipitation contain both forced responses and internal variations14,15,16. However, soil moisture, as the crucial indicator of agricultural and ecological droughts, has received less attention in understanding the cause of its observed changes in Central Asia.

There is a growing concern regarding the detection and attribution of long-term changes in soil moisture at regional or global scales. While soil moisture deficits are attributed to anthropogenic warming in America17,18, Europe9,19,20, the Northern Hemisphere21 and global land22, natural climate variability is also suggested to drive severe agricultural drought at regional scales8,18,23. Understanding the influence of external forcing and internal variability on soil moisture change over Central Asia needs to separate the forced and unforced responses. The US Climate Variability and Predictability Program (CLIVAR) Working Group on Large Ensembles has set up a data archive containing large ensembles of historical and scenario simulations for several models24. Unlike Coupled Model Intercomparison Project (CMIP)-like multi-model simulations with each model having few members, which are mainly used to reveal externally forced signals, large ensemble simulations from a single climate model are powerful tools to understand and quantify the relative contribution of internal variability and external forcing24,25,26,27,28. In this study, based on three sets of large ensemble simulations, we provide evidence that the combined impact of external forcing and the phase change of the Interdecadal Pacific Oscillation (IPO), a long-term oscillation of sea surface temperatures (SST) in the Pacific Ocean that can last from 20 to 30 years (refs. 29,30,31), has exacerbated agricultural droughts over water-scarce southern Central Asia in its early growing season since 1992. The IPO evolution could also modulate the externally forced drying trend in the near-term projection.

Observed drying trend

We focus on the surface soil moisture content in April–May–June (AMJ), which is the early time of the growing season in Central Asia32. Plants in this region mainly depend on shallow soil water and shallow groundwater to survive5. Since the 1950s, a decline in surface soil moisture has been seen over southern Central Asia (Extended Data Fig. 1). The drying tendency has accelerated since the 1990s with decreasing rates at −4.53%, −1.29% and −2.12% per decade during 1992–2020 for ERA5-Land, GLEAM and ESACCI, respectively (Fig. 1 and Extended Data Table 1). The soil moisture deficit in 2021 was one of the most severe droughts in the past 70 years, which was −11.79% lower than the climatological mean (1995–2014) based on ERA5-Land (Fig. 1a).

Fig. 1: Drying trend over southern Central Asia.
figure 1

a, Surface soil moisture content anomaly (%) in AMJ 2021 relative to the climatological mean (1995–2014) derived from ERA5-Land. b, The trends of AMJ surface soil moisture content anomaly (% per decade) during 1992–2020 derived from ERA5-Land. Stippling indicates that the linear trends are significant at the 10% level according to the Mann–Kendall (MK) test. c, The trends in AMJ surface soil moisture content (% per decade) during 1950–1992 (grey) and 1992–2020 (blue) over southern Central Asia (red box in b) obtained from ERA5-Land, GLDAS (ended 2014), GLEAM and ESACCI, and the ensemble mean of CESM1-CAM5, CSIRO-Mk3-6-0 and MPI-ESM-LR. The ensemble mean is calculated based on 40, 30 and 100 members for the three models, respectively. The error bars denote the range of the trends derived from all members for each model. As GLEAM and ESACCI are started in 1980 and 1978, respectively, the trends in P1 are not calculated.

Full size image

Agricultural drought is related to meteorological factor changes, including temperature and precipitation33. The surface soil moisture over southern Central Asia was negatively correlated with contemporaneous surface air temperature (r = −0.62; p < 0.01) (Extended Data Fig. 1a,b). The rising temperature increases the rate of evapotranspiration and depletes surface soil moisture34,35. Central Asia has witnessed a significant warming trend in the past half century (Extended Data Fig. 2e, f). The observed warming trend was stronger after 1992, as indicated by the warming rate of 0.36 K per decade and 0.49 K per decade in 1950–1992 and 1992–2020, respectively. The enhanced warming rate after 1992 partly explained the amplified drying trend in soil moisture. Agricultural drought is also related to deficient rainfall. The surface soil moisture content was highly correlated with early precipitation in March–April–May (r = 0.73; p < 0.01) (Extended Data Fig. 1c,d). The deficiency of precipitation over an extended period of time can result in a water shortage for soil and environment, and a lag usually occurs between the meteorological drought and agricultural drought36,37,38. The decreased precipitation in spring and increased temperature-based evapotranspiration in AMJ result in imbalanced water supplies over southern Central Asia after 1992, which further leads to the significant drying trend (Fig. 1b and Extended Data Fig. 2).

Impact of external forcing

Is the accelerated drying trend since 1992 related to anthropogenic external forcing? The trends in surface soil moisture derived from the three sets of large ensemble simulations were examined (Fig. 1c). Under the same external forcing, 40 ensemble members of CESM1-CAM5 exhibited diverse soil moisture trends ranging from −1.36% to 1.40% per decade in 1950–1992 and −2.44% to 2.30% per decade in 1992–2020. There was a weak but significant drying trend of −0.07% per decade (p < 0.05) before 1992 and a larger drying trend of −0.08% per decade after 1992 for the ensemble mean (Fig. 1c and Extended Data Table 1). The significant weak decreasing trends can also be found for CSIRO-Mk3-6-0 and MPI-ESM-LR in both periods, which were −0.23% and −0.60% per decade (p < 0.01) for CSIRO-Mk3-6-0 and −0.40% and −0.65% per decade (p < 0.01) for MPI-ESM-LR in 1950–1992 and 1992–2020, respectively (Fig. 1c). Hence the ensemble means of the large ensembles indicate that the external forcing has a weak but statistically significant contribution to the drying surface soil moisture over southern Central Asia since 1950, in particular after 1992.

Anthropogenic activities, especially the emission of greenhouse gases, has resulted in the significant warming trend in the past half century and the accelerated warming rate after 1990s13. The human-induced warming further leads to the increased evapotranspiration and decreased soil moisture. In addition, the spatial pattern of the long-term warming trend due to external forcing has suppressed spring precipitation over southern Central Asia15,39, which was also favourable for the reduction in soil moisture content in AMJ.

Impact of the internal mode of IPO

While the ensemble mean simulation, as a measure of the response to external forcing, can reproduce the accelerated drying trend over southern Central Asia after 1992, the simulated drying trend was far weaker than the observation (Fig. 1c and Extended Data Table 1). Among the large ensemble members, there were some realizations that can reproduce the observed magnitude change, indicating the contribution of internal variability (Fig. 1c). The SST trend difference between the Min10 (the mean of 10 members with the largest negative trends of soil moisture in 1992–2020) and the ensemble mean of all members were analysed and a negative phase of an IPO-like pattern was found, which featured a cooling trend over the tropical central eastern Pacific and a warming trend over the subtropical western Pacific (Fig. 2a–c). The IPO index decreased during 1992–2020 for the Min10 of all models (Fig. 2f). Following the accelerated drying trend over southern Central Asia since 1992, the positive-to-negative phase transition of the IPO was indeed seen in the observations (Fig. 2d,e).

Fig. 2: Impact of IPO on the AMJ soil moisture over southern Central Asia.
figure 2

a–d, Spatial distribution of trends in SST (°C per decade) for the difference between Min10 and ensemble mean of CESM1-CAM5 (a), CSIRO-Mk3-6-0 (b), MPI-ESM-LR (c) and the trends in SST derived from Kaplan (d) during 1992–2020. Stippling indicates that the linear trends are significant at the 10% level according to the MK test. e, Time series of the 9-year running mean IPO index (°C) derived from Kaplan SST (black) and ERSST (grey). The vertical line denotes the year of 1992. f, The trends of IPO indices (°C per decade) derived from the ensemble mean of Max10 (Min10) during the period of 1950–1992 (1992–2020) derived from CESM1-CAM5 (black), CSIRO-Mk3-6-0 (blue) and MPI-ESM-LR (red).

Full size image

Without the impact of external forcing, the positive-to-negative phase transition of IPO alone can lead to a decrease in soil moisture over southern Central Asia during 1992–2020. Following the transition of the IPO to a negative phase, anomalous SST trends will strengthen the Walker circulation and further enhance (suppress) convective activities over the Indo-western Pacific warm pool (central eastern Pacific)40. The increased latent heating associated with the enhanced precipitation in the Indo-western Pacific warm pool would stimulate westward-propagating baroclinic Rossby wave trains emanating from the western Pacific, which can lead to reduced rainfall over southern Central Asia15,39,41,42. In addition, the suppressed convective heating over the central eastern Pacific would stimulate poleward-propagating equivalent barotropic Rossby wave trains, which could extend eastward over North America and the North Atlantic into Eurasia, modulating the precipitation over southern Central Asia14,42,43,44,45. The precipitation changes further decreased the surface soil moisture over Central Asia. In addition, the IPO shifted from a negative to a positive phase during 1950–1992 in observations (Fig. 2e), which was favourable for more precipitation and wetter soil moisture over southern Central Asia before 1992 (Fig. 2f and Extended Data Fig. 3).

Attribution of observed drying trends to climate drivers

To quantify the relative contribution of external forcing and IPO, the IPO evolution was adjusted in each member based on the observed IPO index during the whole period (Methods). The original IPO-related soil moisture changes for each member were replaced by those induced by the observed IPO evolution. Therefore, the ensemble mean of the adjusted members represents the combination of the responses to external forcing and the observed IPO phase transition.

While the externally forced soil moisture trends during 1950–1992 ranged from −0.40% to −0.07% per decade for the three models, the negative-to-positive phase transition of IPO was favourable for wetter conditions over southern Central Asia at a rate of 0.22% to 0.55% per decade (Fig. 3a,c, Extended Data Fig. 4 and Extended Data Table 1). After adjustment, the IPO-induced increase in soil moisture offset the externally forced response, showing a weak trend in this period (−0.05% to 0.48% per decade). During 1992–2020, the adjustment with the observed positive-to-negative IPO transition decreased the soil moisture trend from −0.65% to −0.08% per decade to −1.34% to −1.08% per decade, which was closer to the observed trend (Fig. 3b,c, Extended Data Fig. 4 and Extended Data Table 1). The simulated IPO-induced trend was −1.00% to −0.61% per decade during 1992–2020, which was comparable to the externally forced drying trend. All the models converged towards a high probability of an accelerated drying trend over southern Central Asia after adjustment (Fig. 3b,c).

Fig. 3: Adjustments of AMJ surface soil moisture content over southern Central Asia according to the observational IPO phase transition.
figure 3

a,b, Histograms (bars) and fitted distribution (line) of the frequency of occurrence of the soil moisture content trends relative to 1995–2014 (% per decade) for the 170 members of the three models during 1950–1992 (a) and 1992–2020 (b) derived from the original results (orange) and those adjusted accounting for the influence of the observational IPO phase transition (blue). The black and red dots denote the trends for the ensemble mean without and with adjustments. The numbers denote the percentages for members with negative trends. c, The original (blue) and adjusted trends (orange) in soil moisture content during 1992–2020 were obtained from CESM1-CAM5, CSIRO-Mk3-6-0 and MPI-ESM-LR, respectively. ‘Pac-pacemaker’ denotes the trends of the ensemble mean for the historical simulations (blue) and IPO pacemaker simulations (orange) of CESM1.1 during 1992–2013. The numbers after each of the climate models are ensembles analysed. The error bars denote the 25–75% range of all members. The grey bar denotes the average of ERA5-Land, GLEAM and ESACCI, and the error bar denotes the range of the three soil moisture products.

Full size image

We further examined the contributions of IPO and anthropogenic external forcing by analysing the 20-member IPO pacemaker simulations of CESM1.1 and its corresponding historical simulations. In IPO pacemaker simulations, while the atmosphere and ocean are freely coupled in the historical simulations, the SST anomalies over the tropical Pacific vary synchronously with the observations (Methods). Whereas the historical simulations represent the response to external forcing, the pacemaker simulations represent the combined impact of observed IPO evolution and external forcing. The soil moisture decreased at rates of −0.79% per decade and −2.44% per decade for the ensemble mean of historical and pacemaker simulations, respectively. By considering both anthropogenic external forcing and IPO-related internal SST changes, the pacemaker simulations reasonably reproduced the observed drying trend after 1992 (Fig. 3c), further demonstrating the crucial contribution of internal variability of IPO.

Future changes

Under the representative concentration pathway version 8.5 (RCP8.5) emission scenario, the externally forced surface soil moisture content over southern Central Asia features a continued decreasing trend at rates of −0.87%, −0.09% and −1.24% per decade to the end of the century for the ensemble mean of CESM1-CAM5, CSIRO-Mk3-6-0 and MPI-ESM-LR, respectively (Fig. 4a). Multi-model projections of CMIP6 models also show a reduction in surface soil moisture over southern Central Asia under external forcing in both low- and high-emissions scenarios10. While the aggravated agricultural drought in AMJ will continue, the spring precipitation is projected to increase under the external forcing46. To reveal the discrepancy among the future changes in precipitation and soil moisture content, we binned spring precipitation and AMJ surface air temperature over southern Central Asia into 5 × 5 or 10 × 10 bins and assessed the anomalies of surface soil moisture in each bin in observation and large ensemble simulations (Extended Data Fig. 5). For precipitation with similar magnitudes, surface soil moisture decreases as surface air temperature rises with promoted evapotranspiration. Among springs with positive precipitation anomalies, only 6.33% to 19.39% and 11.35% to 21.89% springs experienced negative soil moisture anomalies during 1950–1992 and 1992–2020, and the ratio would come to 16.78% to 30.32% and 29.24% to 56.16% for the 2 °C and 4 °C warming levels (Fig. 4c,d). Hence, although the precipitation will increase during the twenty-first century associated with the moistening atmosphere under external forcing, agricultural drought still will become more severe due to human-induced continuous warming and related increase of evapotranspiration.

Fig. 4: Future changes in AMJ surface soil moisture over southern Central Asia.
figure 4

a, Time series of the 9-year running mean soil moisture content anomalies (%) derived from CESM1-CAM5 (black), CSIRO-Mk3-6-0 (blue) and MPI-ESM-LR (red). The lines denote the results of the ensemble mean and the number denote the corresponding linear trends (% per decade) in 2021–2099; the light shades are for one standard deviation about the ensemble mean. b–e, The relationship between the area-averaged AMJ surface soil moisture content anomaly (%) and spring (MAM) precipitation anomaly (%) during 1950–1992 (b) and 1992–2020 (c) and under 2 °C (d) and 4 °C (e) levels of global warming derived from CESM1-CAM5 (black), CSIRO-Mk3-6-0 (blue) and MPI-ESM-LR (red). The numbers denote the percentage of springs with negative soil moisture anomaly among springs with positive precipitation anomaly.

Full size image

The near-term externally forced projection can be amplified or reduced by internal variability40,47. Given the evidence that the recent externally forced drying trend in southern Central Asia is accelerated due to the IPO evolution, will the IPO modulate the externally forced drying trend in the near term? We excluded all of the IPO’s influence by removing the soil moisture variations that are linearly related to the IPO index in each member (Methods). After the removal, the standard deviations (SDs) of the internal component of the linear trends during 2021–2040 among all members are reduced by 15% to 32% (Fig. 5a), indicating that the spread among different members can be partly explained by the IPO phase transition.

Fig. 5: The changes in AMJ surface soil moisture content over southern Central Asia during 2021–2040 with and without IPO’s influence.
figure 5

a, The linear trends of the internal component of the soil moisture content (% per decade) relative to 1995–2014 derived from 40, 30 and 100 members of CESM1-CAM5, CSIRO-Mk3-6-0 and MPI-ESM-LR, respectively. The grey thin lines denote the original results, and the red thin lines denote the trends with the IPO’s influence being removed. The dots (thick lines) denote the corresponding ensemble mean (standard deviation, ±SD). b, The original and adjusted trends in soil moisture content were obtained from CESM1-CAM5, CSIRO-Mk3-6-0 and MPI-ESM-LR, respectively. The bars denote the ensemble mean of the three models, and the error bars denote the 25–75% range of all members. The grey bars denote the original results, and the blue and yellow bars denote the trends with the same amplitude of a positive (+2 SD per 2 decades) or a negative IPO (−2 SD per 2 decades) phase transition, respectively.

Full size image

To reveal the role of the IPO on the near-term projection, we calculated the SD (0.23 °C) of the observational IPO index during 1950–2020 and assumed that the IPO will shift from +1SD to −1SD or from −1SD to +1SD in 2021–2040. The IPO phase change of 2SDs over a 20-year period (± 0.23 °C per decade) is comparable to the observational trends in 1950–1992 (0.15 °C per decade) and 1992–2020 (−0.28 °C per decade). We then replaced the original IPO-related soil moisture changes for each member by those induced by the assumed fixed IPO evolution (Methods). If an amplitude of −2SD of the IPO trend was predicted to be superimposed onto the external forcing from 2021 to 2040, the multi-model average would show a significant decrease in soil moisture by −1.52 (−2.19 to −0.46 for the model spread) % per decade (Fig. 5b), which is 0.65 (0.42, 0.83) % per decade drier than the externally forced drying trend (−0.87 (−1.36, −0.04) % per decade), indicating nearly 75% enhancement of the drying trend. In contrast, if the IPO shifts from −1SD to +1SD, the negative-to-positive phase transition will weaken the externally forced drying trend by 0.52 (0.12, 0.86) % per decade, indicating nearly 60% weakening of the drying trend. The phase transition of the IPO could modulate the externally forced drying trend in the following decades and has a notable impact on the soil moisture trends over southern Central Asia in the near term.

Implications of a continued drying trend

The southern part of Central Asia has been facing an increase in aridity in its early growing season since the 1990s. The drying trend is not limited to surface soil but can also be found in deeper soil layers (Extended Data Fig. 6). Dominated by anthropogenic warming, agricultural droughts over southern Central Asia were expected to be exacerbated by the end of the century. The region with the highest population density at the foothill of the mountain was more likely to face a further enhanced agricultural drought (Extended Data Fig. 7). Superimposed onto the external forcing, the IPO evolution not only has intensified the decreasing tendency of soil moisture during 1992–2020 but would also modulate the externally forced drying trend in the near-term projection. Thus, improved IPO predictions are necessary for developing effective adaptation and mitigation strategies for decisionmakers of Central Asian countries. If the IPO turns into its warm phase in the next few decades, as some previous studies supposed40, an amplitude of +2SD of the IPO trend will weaken the externally forced decreasing rate of surface soil moisture content by nearly 60% during 2021–2040.

In Central Asia, crop and livestock production is the dominant contributor to GDP and employment share33,48. The increasingly serious agricultural droughts not only directly influence crops and livestock but are also a challenge for production, living, ecological environments, social stability and even regional security49,50. A catastrophic drought during 2000–2001 revealed that due to structural factors, including rural poverty and unsustainable management of natural resources, agricultural drought in Central Asia was extremely likely to further lead to socioeconomic drought33. Our results show that the decreasing trend in surface soil moisture since the 1990s will continue in the twenty-first century under the RCP8.5 scenario regardless of how the IPO changes unless we take ambitious climate action and reach global peaking of greenhouse gas emissions as soon as possible to achieve a climate-neutral world. In addition to agricultural drought, there are increasing risks of meteorological drought51 and hydrological drought52,53 in southern Central Asia. Overall, our findings have highlighted an urgent need of developing a thorough risk-management plan at sectoral, local, country and even multi-country levels to adapt and mitigate the enhanced agricultural droughts for food security and livestock production.

Methods

Soil moisture

In this study, we focused on agricultural drought by analysing the surface soil moisture content. Multi-source soil moisture products were used, including: (1) ERA5-Land reanalysis soil moisture data derived from the Land Surface Scheme of the operational Integrated Forecasting System used at European Centre for Medium-Range Weather Forecasts with a spatial resolution of 0.1 × 0.1° during 1950–2021, soil moisture contents over 0–7 cm, 7–28 cm, 28–100 cm and 100–289 cm are available54; (2) the soil moisture product derived from the NASA’s Global Land Data Assimilation System (GLDAS Noah) with a spatial resolution of 1 × 1° during 1948–2014, soil moisture contents over 0–10 cm, 10–40 cm, 40–100 cm and 100–200 cm are available55; (3) the soil moisture product derived from the Global Land Evaporation Amsterdam Model (GLEAM v3.3a) with a spatial resolution of 0.25 × 0.25° during 1980–2021, soil moisture contents at surface layer (0–10 cm) and root zone are available56; (4) the multi-sensor soil moisture dataset obtained from European Space Agency (ESA) in the implementation of the Global Basic Climate Change Initiative (CCI) with a spatial resolution of 0.25 × 0.25° during 1978–2021, which merged soil moisture observations derived from multiple active and passive satellite remote sensing instruments in the microwave domain and was adjusted based on break correction; only soil moisture content at the surface layer is available57.

Climate data

The monthly precipitation provided by the Met Office Hadley Center gridded land surface extremes indices (HadEX3)58 was used to examine the relationship between surface soil moisture and regional precipitation. Monthly gridded observational temperature from the Berkeley Earth Surface Temperature dataset (BEST)59 was used to reveal the relationship between surface soil moisture and regional temperature. Details of the datasets are provided in Supplementary Table 1.

Large simulations

To investigate the impact of internal variability and external forcing, three sets of large ensemble simulations from the US CLIVAR Working Group on Large Ensembles were used, including CESM1-CAM5, CSIRO-Mk3-6-0 and MPI-ESM-LR. These simulations are selected because at least 30 members are available and monthly surface soil moisture content is provided. A detailed introduction of the models is provided in Supplementary Table 2. The historical simulations covering the period of 1850–2005 were driven by observed historical changes in radiative forcing following phase 5 of the Coupled Model Intercomparison Project (CMIP5)60. To satisfy the analysis of long-term trends during 1950–2020, the simulations were extended to 2020 following the representative concentration pathway version 8.5 (RCP8.5) emissions scenario. The three models can reasonably reproduce the climatological mean of surface soil moisture (Supplementary Fig. 1), the long-term trends in surface temperature (Supplementary Fig. 2) and the relationship between precipitation over southern Central Asia and moisture transport from the Indian Ocean (Supplementary Fig. 3).

Pacemaker simulations

To confirm the impact of IPO on Central Asian soil moisture, the results of 20-member coupled model ensembles of tropical Pacific ‘pacemaker’ simulations performed with the National Center for Atmospheric Research Community Earth System Model version 1 (NCAR CEMS1) were compared with the corresponding historical simulations. The Pacific pacemaker simulations were forced by historical radiative forcing through 2005 and the RCP8.5 radiative forcing thereafter like the historical simulations, but the SST anomaly was nudged to be observations in the eastern tropical Pacific. Further details on the model and simulations can be found in Deser et al.61,62.

Separating externally forced and internally unforced signals

Take CESM1-CAM5 as an example. A certain variable V(i) of member i in CESM1-CAM5 is influenced by both external forcing and internal variability. As all members are driven by the same external forcing, the ensemble means of V(i) for all members can be regarded as the response to external forcing Vforced and are represented as:

$$V_\mathrm{forced} = \left( {V\left( i \right)} \right)_\mathrm{EM},i = 1,2, \ldots ,n - 1,n$$
(1)

Thus, V(i) can be separated as:

$$V\left( i \right) = V_\mathrm{forced} + V_\mathrm{internal}(i),i = 1,2, \ldots ,n - 1,n,$$
(2)

where Vinternal(i) is the internal component estimated as the residual of the original V(i) minus the model forced response Vforced.

IPO definition

The IPO index is defined as the difference between the SST anomalies averaged over the central equatorial Pacific (10° S−10° N, 170° E−90° W) and the average of the SST anomalies in the Northwest (25° N−45° N, 140° E−145° W) and Southwest Pacific (50° S−15° S, 150° E−160° W) following Henley et al.31. The definition of the IPO index varies among different studies with similar phase transitions30,31,63. We calculated the IPO index based on area-averaged SST anomalies rather than the leading mode of SST in the Pacific Ocean for easier calculation and direct comparison between observations and model simulations. There is a high consistency between the IPO indices derived from the Kaplan SST dataset64 and Extended Reconstructed Sea Surface Temperature (ERSST)65 with a correlation coefficient of 0.99 (p < 0.01). In particular, the IPO index for each ensemble member of the simulations was obtained by the internal part of SST, namely SSTinternal(i).

Adjustment of soil moisture based on observational IPO

An ‘adjustment’ method following Salzmann and Cherian66 was used in this study. The core idea of this method is replacing the original simulated IPO-related soil moisture trend with that induced by observed IPO evolution, which is expressed as:

$$\partial_{t}V_{\mathrm{adj}}\left( i \right) = \partial_{t}V_{\mathrm{forced}} + \partial_{t}V_{{\mathrm{internal}}\_{\rm{adj}}}\left( i \right),i = 1,2, \ldots ,n - 1,n,$$
(3)

where:

$$\partial_{t}V_{{\mathrm{internal}}\_{\rm{adj}}}\left( i \right) = \partial_{t}V_{\mathrm{internal}}\left( i \right) + A_{\mathrm{internal}}\left( i \right).$$
(4)

Ainternal (i) is the adjusted term and can be expressed as:

$$A_{\mathrm{internal}}\left( i \right) = - r_{V,\mathrm{IPO}}(i)\left[ {\partial_{t}{\mathrm{IPO}}\left( i \right) - \partial_{t}{\mathrm{IPO}}_{\rm{OBS}}} \right],$$
(5)

where \(r_{V,\mathrm{IPO}}(i)\) is the regression coefficient of the soil moisture regressed onto the IPO index during the study period, reflecting the response rate of soil moisture to IPO evolution for member i. \(r_{V,\mathrm{IPO}}(i){\partial_{t}}\mathrm{IPO}\left( i \right)\) denotes the original IPO-related soil moisture trend in member i, and \(r_{V,\mathrm{IPO}}(i)\partial_{t}{{\mathrm{IPO}}_{\rm{OBS}}}\) denotes the soil moisture trend related to the observed IPO evolution with the same amplitude. While \(\partial _{t}V_{\mathrm{forced}}\) represents the response of soil moisture to external forcing, the ensemble mean of \(\partial_{t}V_{{\mathrm{internal}}\_{\rm{adj}}}\left( i \right)\) can represent the soil moisture trend contributed by the observed IPO evolution.

Removing IPO’s influence

To remove the IPO’s influence, we calculated the IPO index for each member and removed the soil moisture variations that are linearly related to the IPO index through a linear regression, which is expressed as:

$$V_{{\mathrm{non}} - {\rm{IPO}}}\left( {i,t} \right) = V\left( {i,t} \right) - r_{V,\mathrm{IPO}}\left( i \right)\times\mathrm{IPO}\left( {i,t} \right),i = 1,2, \ldots ,n - 1,n.$$
(6)

Statistical analysis

The area-weighted method was used to calculate the regional average. A 9-year running mean was applied to observation and model outputs to isolate the interdecadal signal. The Theil–Sen trend estimation method was used in this study to estimate the linear trend, which is not sensitive to outliers67. The non-parametric MK test was used to detect the presence of the trends68,69.