1 Introduction

Water is an essential resource that has a broad impact on socioeconomic systems and natural ecosystems (Schaldach et al. 2012). Water usage by cultivated crops is a primary issue facing modern agriculture and changing patterns of precipitation make regional and local water scarcity events more probable around the world (Gosling and Arnell 2016; Mancosu et al. 2016; Wang et al. 2017). Driven by climate change, there are serious risks of water shortage in China, even in the places with relatively abundant water resources including precipitation, such as the southern China mainland (Wang et al. 2014; Ye et al. 2015; Yu et al. 2016; Gao et al. 2018). In general, water is far from sufficient for crop growth and irrigation is thus necessary to maintain production. Currently, agricultural irrigation water accounts for > 70% of total domestic water consumption in China (Han et al. 2019). In particular, rice is one of the most water-consuming crops and China produces 32% of the total worldwide production during 1981–2018 (FAO 2020). Therefore, careful regulation of irrigation water resources is essential to ensure sustainable rice production.

There are two factors to the efficient use of irrigation water to optimize water resources. One is to optimize irrigation systems and in practice, a number of approaches have been used including traditional, shallow wet, intermittent, furrow, drip, and controlled irrigation. The optimal procedure for rice water-use efficiency depends on local and regional soil composition and precipitation patterns (He et al. 2013; Zhang et al. 2013; Wang et al. 2016). The second factor is determination on the actual amount of irrigation water needed for rice cultivation, the irrigation water requirement (IWR). IWR is a tool that in general encompasses the water requirements for rice and has been increasingly used as a predictive tool (Wang et al. 2014, 2017; Ding et al. 2017; Han et al. 2019). Deterministic tools and methods have been developed to assist in irrigation planning and water management. Crop growth simulation models, merging dynamic and mechanistic properties of crop growth, are widely used proposed to calculate the IWR over a growing period. For example, ORYZA2000 and ORYZA v3 are two common simulation models for IWR calculations across rice cultivation regions in China (Wang et al. 2017; Ding et al. 2017, 2020). The soil water balance method is another traditional method for calculation IWR and has also been applied to rice cultivation (Shahid 2011; Mancosu et al. 2016).

The IWR is highly dependent on associated climatic conditions and this becomes more important as global climatic patterns affect regional and local water availability (Wang et al. 2014; Nie et al. 2019). This will be inevitably ongoing in the future. Global climate model (GCM) and regional climate model (RCM) are the current most useful tools to estimate future dynamic IWR predictions for rice (Gosling and Arnell 2016; Ding et al. 2017). In the outputs derived from these models, temperature and precipitation have been closely linked to the actual observational data (Donat et al. 2017; Sangelanton et al. 2019). There are still small biases in these outputs but they are superior to other simulations, e.g., net radiation and wind speed (Liu and Sun 2016). These differences in the integrity and accuracy of the data limit the use of one particular method such as a crop growth simulation model that requires multivariable inputs and may generate greater uncertainty than the method using reliable inputs. Overall, the input of fewer variables is most appropriate for quantifying IWR.

A simpler soil water balance method that couples crop evapotranspiration and effective precipitation has been proposed to quantify IWR (Ye et al. 2015). Crop evapotranspiration has been evaluated by a variety of methods, such as Penman-Monteith, Thornthwaite, Holdridge, Hargreaves, Priestley, and Taylor (Bormann 2011; Feng et al. 2017; Zhang et al. 2020b). The Penman-Monteith model requires the most detailed data inputs and this has precluded in satisfactory estimation of future evapotranspiration. It is thus reasonable and necessary to choose proper models before they are applied to a specific region.

Until now, efforts to include the impacts of climate change on IWR calculations have been used across the southern rice-plantation areas of China but not for Northeast China (NEC). NEC encompasses the largest plain areas and is a significant base of rice production for China. The planted areas and yields of single-season rice in NEC have increased over the past decades, and on average account for 21.4 and 20.9% in total single-season rice planted areas and yields in China during 2001–2017, in accordance with the statistics from the National Bureau of Statistics of China. The inclusion of IWR dynamics into irrigation scheduling is important to maintain current production yields. This can be accomplished under two representative concentration pathway scenarios, moderate greenhouse gas emissions (RCP4.5) and highest greenhouse gas emissions (RCP8.5), along with utilization of 20 fine-resolution downscaled GCMs as the primary tools to predict potential IWR changes. The current study describes a method to calculate IWR dynamic processes into future models of rice production and is especially important considering the current state of climatic alterations now being experienced across the globe.

2 Material and methods

2.1 Study region and datasets

Our study area encompassed the northeastern China mainland, including Heilongjiang (HLJ), Jilin (JL), and Liaoning (LN) provinces (Fig. 1). This region involved warm, mid, and cold temperate zones from south to north, respectively, resulting in hot and rainy summers, and cold and dry winters. Annual temperatures during 1981–2010 ranged from − 3.8 to 11.3 °C, precipitation between 370 and 1078 mm, and hours of sunshine varied between 2136 and 2910 h. The rice growing season extended from May to September during most years.

Fig. 1
figure 1

The location of study area

Full size image

We utilized 20 GCMs that were involved in phase five of the Coupled Model Intercomparison Project (CMIP5), i.e., ACCESS1-0, BNU-ESM, CCSM4, CanESM2, CNRM-CM5, CSIRO-Mk3-6-0, GFDL-CM3, GFDL-ESM2G, GFDL-ESM2M, inmcm4, HadGEM2-ES, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC5, MIROC-ESM, MIROC-ESM-CHEM, MPI-ESM-LR, MPI-ESM-MR, MRI-CGCM3, and NorESM1-M (Zhang et al. 2020a). The datasets included daily mean near surface air temperature (Tmean), maximum near surface air temperature (Tmax), minimum near surface air temperature (Tmin), and precipitation (P) during the recorded period (1961–2005) and for future periods (2006–2100) under RCP4.5 and RCP8.5 scenarios. Limited by the coarse resolution for the raw GCMs, a downscaling technique was necessary to generate fine-resolution simulations. The statistical downscaling method we used was the bias correction spatial disaggregation, which was applied to daily outputs (Thrasher et al. 2012; Werner and Cannon 2015). Therefore, Tmean, Tmax, Tmin, and P were reproduced across the study region for 1961–2005 and 2006–2100 under RCP4.5/8.5 for the 20 GCMs at a 0.25° × 0.25° resolution.

Daily observations including Tmean, Tmax, Tmin, P, wind speed, sunshine duration, and actual vapor pressure were recorded during 1961–2005 at 184 meteorological stations (Fig. 1). These records had been strictly controlled by the China Meteorological Data Sharing Service System to maintain internal long-term consistency and homogeneity.

2.2 The calculation of irrigation water requirement

IWR for rice was defined as the deficit of effective precipitation (EP) relative to crop water requirement (CWR) expressed as:

$$ \mathrm{IWR}=\sum \limits_{i=1}^n{\mathrm{CWR}}_i-\sum \limits_{i=1}^n{\mathrm{EP}}_i $$
(1)

where CWRi and EPi are the daily CWR and EP at day i during the calculated period. Daily CWR was quantified as follows:

$$ \mathrm{CWR}={k}_c\times {\mathrm{ET}}_0 $$
(2)

where kc is the crop coefficient and ET0 is the potential evaporation. The value for kc was obtained from prior knowledge (Huang et al. 2015) and assigned 1.05, 1.12, 1.20, 1.20, and 0.75 to May (May), June (Jun), July (Jul), August (Aug), and September (Sep), respectively. To generate reliable outputs from statistical downscaled GCMs, the ET0 was estimated using the Hargreaves model (Hargreaves and Allen 2003) which was a simple and practical method based on daily temperatures as follows:

$$ {\mathrm{ET}}_{0-\mathrm{Har}}=\frac{k}{\lambda }{\left(T\ \max -T\min \right)}^{0.5}\left(T\mathrm{mean}+17.8\right){R}_{\mathrm{a}} $$
(3)

where λ is latent heat of vaporization (2.45 MJ·kg−1) and k is the adjusted coefficient taking 0.0023 as the initial value. Ra is extraterrestrial radiation quantified as:

$$ {R}_{\mathrm{a}}=\frac{24\times 60}{\pi }{G}_{\mathrm{s}\mathrm{c}}{d}_{\mathrm{r}}\left[{\omega}_{\mathrm{s}}\sin \left(\varphi \right)\sin \left(\delta \right)+\cos \left(\varphi \right)\cos \left(\delta \right)\sin \left({\omega}_{\mathrm{s}}\right)\right] $$
(4)

where Gsc is solar constant (0.082 MJ·m−2·min−1); dr is the average distance between sun and earth; ωs is the angle of sunrise (rad); φ is latitude (rad); and δ is magnetic declination of sun (rad). dr, ωs, and δ are expressed as:

$$ {d}_r=1+0.033\cos \left(\frac{2\pi J}{365}\right) $$
(5)
$$ \delta =0.409\left(\frac{2\pi J}{365}-1.39\right) $$
(6)
$$ {\omega}_{\mathrm{s}}=\operatorname{arccos}\left[-\tan \left(\varphi \right)\tan \left(\delta \right)\right] $$
(7)

where J is the Julian day, ranging from 1 (Jan 1) to 365 or 366 (Dec 31).

It was previously determined that the value of k in the Hargreaves model varied by locality implying that it was necessary to first calibrate k (Hargreaves and Allen 2003; Feng et al. 2017). To address this, ET0 derived from the Penman-Monteith model was accepted as the reference (Donohue et al. 2010; Zhao and Dai 2017).

$$ {\mathrm{PET}}_{\mathrm{PM}}=\frac{0.408\varDelta \left({R}_{\mathrm{n}}-G\right)+\lambda \frac{900}{T\mathrm{mean}+273}{\mathcal{U}}_2\left({e}_{\mathrm{s}}-{e}_{\mathrm{a}}\right)}{\Delta +\gamma \left(1+0.34{u}_2\right)} $$
(8)

where Rn is net radiation (MJ·m−2·day−1); G is soil heat flux density (MJ·m−2·day−1); es is saturation vapor pressure (kPa); ea is actual vapor pressure (kPa); Δ is slope of the saturation vapor pressure function (kPa·°C−1); γ is psychometric constant (kPa·°C−1); and u2 is wind speed at 2 m height (m·s−1). In the routine observation from meteorological station, wind speed was measured at 10 m height. u2 was thus calculated by the following equation:

$$ {\mathcal{U}}_2={\mathcal{U}}_{10}\frac{4.87}{1n\left(67.8z-5.42\right)} $$
(9)

where u10 is the measured wind speed at 10 m height (m·s−1) and z is the measured height, which is 10 m here.

Given that ET0 from the Penman-Monteith model was available, the value of k in the Hargreaves model was calibrated as follows: (1) ET0 from Penman-Monteith during May–Sep in 1976–2005 was obtained from observations and separately calculated for each station; (2) ET0 from Penman-Monteith was interpolated to gridded values at a resolution of 0.25° × 0.25° same to downscaled GCMs; (3) gridded ET0 from Hargreaves estimated from an ensemble of 20 GCMs was compared with gridded ET0 from Penman-Monteith to adjust the coefficient k. Based on this adjusted k value, the ET0 from Hargreaves during May–Sep was recalculated during 1976–2005. The normalized root mean squared error (NRMSE) was recognized as the metric to validate the performance of ET0 from Hargreaves when comparing with the ET0 from Penman-Monteith.

As another important factor determining IWR, EP is the fraction of the total precipitation available for rice. It was quantified as previously described (Döll and Siebert 2002) as:

$$ \mathrm{EP}=\left\{\begin{array}{c}\frac{P\times \left(4.17-0.2P\right)}{4.17},\\ {}4.17+0.1\times P,\end{array}\right\},{\displaystyle \begin{array}{c}P<8.3\\ {}P\ge 8.3\end{array}} $$
(10)

where P is daily total precipitation (mm·day−1). The daily EP values were summed during a single period.

2.3 Spatial and temporal analysis method

It had been proved that the reproduced simulations from statistical downscaling GCMs were consistent and agreed with observations from previous studies (Zhang et al. 2020a). The products of 20 downscaled models were merged to an ensemble dataset using the multi-model arithmetic mean ensemble method, and performed better than individual model and reduced uncertainties (Li et al. 2015; Wang et al. 2017). Here, three variables (i.e., tmean, tmax-tmin, EP) used to calculate IWR were evaluated across the northeastern China mainland, as represented in Fig. S1. Comparisons of regional tman, tmax-tmin, and EP in 1976–2005 estimated from downscaled models ensemble and observation elucidated that the variability between them was consistent and the difference was in a small bias, implying the ensemble outputs were reliable to be used in this study. Moreover, it was clear that tmean and EP were generally detected in an increasing trend, but tmax-tmin was in a non-significant trend. The trend of the three variables could contribute to explain the potential changes of IWP in the spatial and temporal patterns under RCP4.5 and RCP8.5 scenarios. The historical period of 1976–2005 (P0) from GCMs was taken as baseline, and future periods were divided into two periods, namely the mid-term period of 2036–3065 (P1) and late-term period of 2070–2099 (P2). To quantify the IWR trends, the Thiel-Sen method (Sen 1968; Thiel 1950) was adopted at grid and provincial scales. In addition, the coefficient of variation (CV) was used to express IWR stability that was defined as the degree of deviation from a standard value over a certain period.

3 Results

3.1 Calibration and validation of the Hargreaves model

The Hargreaves model was calibrated by the calibration of k using ET0 from Hargreaves compared with ET0 from Penman-Monteith. The calibrated k values were spatially distinct and decreased from west to east over the study area with a magnitude of 0.0021–0.0024 in southwestern HLJ and western JL. ET0 was then reobtained from the Hargreaves model with the use of calibrated k and validated by comparing it with ET0 from Penman-Monteith. The NRMSE values for the two ET0 values were < 10% in the majority of the study area with only 10–14% in scattered locals indicating that Hargreaves was an acceptable method to estimate ET0 (Fig. 2).

Fig. 2
figure 2

Calibrated k and NRMSE for the Hargreaves model

Full size image

3.2 Potential spatial-temporal IWR alterations

The availability of reliable P and ET0 data allowed reliable estimation of IWR for rice. During May–Sep there were obvious spatial differences. In baseline, the highest IWR (> 500.00 mm) was located in southwestern HLJ and western JL. From these areas, IWR decreased with the lowest magnitude (< 100.00 mm) in southeastern LJ and eastern LN. IWR in HLJ varied in a decreasing-to-increasing from west to east. There was a similar spatial pattern of IWR in future periods with only small differences between them and baseline. Overall, IWR decreased in 2036–2065 under RCP4.5 but increased in other periods, with larger magnitude under RCP8.5 than RCP4.5 (Fig. 3).

Fig. 3
figure 3

Potential change of IWR in different periods during May–Sep (P0, P1, and P2 represent 1976–2005, 2036–2065, and 2070–2099, respectively; unit is mm)

Full size image

Potential alterations for IWR were also examined for growth time slices (Fig. S2–S6). It was similar that IWR decreased from western areas (southwestern HLJ and western JL) to southeastern and northeastern areas in different months from May to Sep. This spatial pattern was consistent for both baseline and future periods, and the changes between them were small. A relative positive change of IWR occurred in the latter periods and a larger value was found for RCP8.5 relative to RCP4.5. IWR in Sep was generally < 50.00 mm, much smaller than all other months in which IWR was consistently < 30.00 mm in the southeastern areas (eastern LJ and eastern LN). In western areas (southwestern HLJ and western JL), IWR was > 100.00 mm. Overall, comparisons of IWR grouped by month indicated that IWR was ranked as Jun > May > Jul > Aug > Sep.

The baseline IWR indicated increases during May–Sep from the southwestern areas to northeastern areas, and the tendency ranged from < − 0.8 to 0.4–0.8 mm·a−1. Although the spatial patterns differed for the monthly IWR values, in general they increased from south to north (Fig. 4). In the future periods, IWR generally tended to increase during May–Sep in the respective ranges of 0.20–0.80 and 0.40–1.00 mm·a−1 under RCP4.5 and RCP8.5. Additionally, a trend of 0–0.60 mm·a−1 in monthly IWR was expected and greater for RCP8.5 than RCP4.5.

Fig. 4
figure 4

Tendency of IWR in different periods (P0 represents 1976–2005; unit is mm·a−1)

Full size image

3.3 Regional change of IWR

When gridded IWR was integrated to the province value, provincial IWR dynamic was explored. In LN, IWR in baseline was 317.34, 85.74, 82.73, 61.65, 60.89, and 26.32 mm, for May–Sep, May, Jun, Jul, Aug, and Sep, respectively. Relative to baseline, IWR during May–Sep in 2036–2065 and 2070–2099 was changed by − 4.63 and 1.46% under RCP4.5, and − 0.30 and 3.09% under RCP8.5, respectively (Table 1). A general negative change of IWR in the future periods was detected in Jul, Aug, and Sep, while a positive change was recorded in May and June. Regarding JL, IWR was 288.27, 77.62, 74.00, 61.50, 55.92, and 19.23 mm in baseline, for May–Sep, May, Jun, Jul, Aug, and Sep, respectively. A positive change of IWC during May–Sep was identified for 2070–2099 under RCP4.5, as well as 2036–2065 and 2070–2099 under RCP8.5. IWR for each month was generally positive in 2070–2099 under RCP4.5 and RCP8.5. For HLJ, IWR in baseline was 298.27, 76.04, 82.97, 75.06, 54.39, and 9.81 mm, for May–Sep, May, Jun, Jul, Aug, and Sep, respectively. Similar to JL, IWR for HLJ during May–Sep and individual month was generally positive in 2070–2099 under RCP4.5, as well as 2036–2065 and 2070–2099 under RCP8.5. Overall, change of IWR in 2070–2099 relative to 1976–2005 was larger than that in 2035–2065 relative to 1976–2005.

Table 1 IWR in baseline and percent change in future periods relative to baseline in provinces
Full size table

Comparisons of IWR by province indicated a positive trend for May–Sep, May, and Jun (Fig. 5). In addition, this tendency was greater under RCP8.5 relative to RCP4.5. IWR was characterized by fluctuation within periods and CV for May–Sep under RCP8.5 was generally larger than for RCP4.5 (Fig. 6). The CV in May, Jun, and Jul was smaller than that in Aug and Sep. A smaller CV in Jun, Jul, Aug, and May–Sep was demonstrated in HLJ, relative to LN and JL. Overall, the IWR changes were largest in HLJ, followed by JL and LN. Moreover, the change of IWR was larger under RCP8.5 and this value in May–Sep in 2070–2099 was more than 2 times greater than that calculated under RCP4.5.

Fig. 5
figure 5

Yearly variation of IWR in provinces (black, blue, and red scatters represent baseline, RCP4.5, and RCP8.5, respectively)

Full size image
Fig. 6
figure 6

Coefficient of variance for IWR in provinces (black, blue, and red scatters represent baseline, RCP4.5, and RCP8.5, respectively; P0, P1, and P2 represent 1976–2005, 2036–2065, and 2070–2099, respectively)

Full size image

4 Discussion

Increases in temperatures and altered patterns of precipitation inevitably affected future water and irrigation requirements for rice. Climate simulations for these elements were therefore necessary for projecting the dynamic of water and irrigation requirement for rice. The ET0 taken from the Hargreaves model was a good reproduction of temperatures derived from downscaling GCM. However, it was necessary to calibrate the Hargreaves model according to locality (Hargreaves and Allen 2003). Following calibration, ET0 from Hargreaves was consistent with reference values and with previous publications (Feng et al. 2017; Zhang et al. 2020b). This confirmed that the calibrated Hargreaves model outperformed raw model results. Crop models have been used to estimate yields and biomass as well as ET (Wang et al. 2014, 2017) and have taken the bias between simulation and observation or reference into account as inevitable. For example, comparisons between simulated and measured ET indicated a substantial 6 to 20% discrepancy (Boudhina et al. 2019). Therefore, our Hargreaves model containing < 10% NRMSE was acceptable and could be used to obtain reliable ET0, and CWR values for rice.

Based on reliable EP and CWR values, IWR could be dynamically explored for baseline and future periods. It was found similar spatial patterns of IWR during rice growing periods that decreased from western areas (southwestern HLJ and western JL) to southeastern and northeastern areas, which was consistent with previous studies that identified IWR decreased in these same areas (Huang et al. 2015). In addition, there was a special pattern of decrease-to-increase from west to center to east in HLJ that was also consistent with a previous study (Nie et al. 2019). Over time, a general increasing trend for IWR was detected across the study region that displayed different magnitudes and more significant under RCP8.5 relative to RCP4.5. The predictions indicated IWR increased across NEC at rates of 0.01–0.10 mm·day−1 in 2046–2065 with greater levels under high over low emission (Zhu et al. 2015). This reinforced our data for IWR spatial-temporal changes although the particular details could vary between studies. For example, western areas had been exposed to higher magnitudes and increased IWR trends indicating that the water deficit for rice would be more severe. However, previous investigations reported a 5-m decline of the shallow groundwater level and a large area of cone of depression in the central and western regions of JL (Fan et al. 2009). The ongoing demand for IWR and declining water resources implied that effective irrigation schedules should be developed for maintaining rice production.

The potential alterations in IWR were the result of the use of CWR and EP data (Nie et al. 2019). To explain these changes, we separately extracted contributions from EP and CWR (Tables 2 and 3). It was clear that positive CWR and EP values were predicted for the future relative to baseline. Recent studies had also suggested that increased temperatures and altered rainfall patterns were likely to cause a significant increase in IWR (Ye et al. 2015). Following climate change, rising temperatures might be the primary reason for the increase in CWR. This had been confirmed by studies that employed the simple water balance model using the CMIP5 datasets as inputs (Wada et al. 2013). Due to increasing temperatures, elevated CWR values were identified for different types of rice across different cultivated regions under future scenarios (Wang et al. 2014, 2017; Ding et al. 2017). Under future climate scenarios relative to the past decades, temperature increases during the growing season for rice resulted in higher CWR (Cong et al. 2011). By using crop models to simulate future rice growth, an increasing IWR was also detected that was due to higher CWR caused by increased temperature (Boonwichai et al. 2018). Relative to baseline, the percent change of CWR in 2070–2099 was larger than that in 2021–2040, and the difference was larger than that change of EP. In addition, CWR and EP changes were greater under RCP8.5 than RCP4.5, and were attributed to IWR changes, consistent with previous reports (Boonwichai et al. 2018). The latter highlighted more severe temperature rises thereby increasing evaporation as well as CWR although EP was still expected to increase in the future.

Table 2 EP in baseline and percent change in future periods relative to baseline in provinces
Full size table
Table 3 CWR in baseline and percent change in future periods relative to baseline in provinces
Full size table

The consequences of IWR-related results presented above were derived using a single rice variety (Wang et al. 2017). This was applied with the use of a developing growing phenology to calculate phenological period and further assess the impacts of climate change on IWR for rice (Ding et al. 2017). However, the growing period for rice would most likely change with altered climate conditions and the average growth duration for rice would be shortened under warming conditions. For example, the rice growing season in 2010–2050 would be reduced by 4.7 and 5.8 days under the SRES A2 and B2 scenarios, respectively (Zhang et al. 2015). Thus, strategies such as taking mid-late-maturing rice cultivars to replace early-maturing ones and breeding new rice cultivars with higher thermal requirements have been proposed to lessen the impact on yields (Wu et al. 2014; Zhang et al. 2017). Our findings for the use of a stabilized time period of May–Sep for rice growth might be an alternative way as thermal requirements for rice varieties would vary with climate alterations.

IWR projections under climate change were plagued by some inevitable uncertainties. Firstly, although downscaling and ensemble methods reduce bias relative to individual GCMs, inherited uncertainties that originated from emission scenarios and parameterization schemes would not be totally removed (Ning et al. 2012). In addition, more precise GCM simulations lead to decreased uncertainty and more reliable potential IWR predictions. Methods accessing IWR were primarily based on simple assumptions and ignored some uncertainties such as assuming that the rice crop coefficient would be stable same to an inherent assumption in crop simulation models. However, this would not be appropriate here since rice varieties would be altered with changing climate conditions. Up to now there was no satisfactory way to solve this problem and thus the assumption was generally accepted.

5 Conclusion

Exploring the potential impact of climate change on IWR is of great importance for the sustainability of water and rice production. Fine-resolution temperature and precipitation datasets from 20 downscaled GCMs were successfully applied in the Hargreaves model to estimate ET and to systematically assess the potential IWR dynamic for rice across NEC. IWR predictions decreased from southwestern HLJ and western JL to the southeastern and northeastern areas for total growing times in baseline and future periods. In particular, IWR ranked by months was Jun > May > Jul > Aug > Sep. IWR displayed an increasing trend from east to west across the study region and was larger under RCP8.5 relative to RCP4.5. The provincial IWR indicated a positive change in 2070–2099 under RCP4.5 as well as in 2036–2065 and 2070–2099 under RCP8.5. The IWR changes were ranked as HLJ > JL > LN. The change of IWR in May–Sep in 2070–2099 under RCP8.5 relative to baseline was more than 2 times greater than that under RCP4.5. These current and potential changes of IWR suggested that special adaptive strategies should be implemented in different areas across NEC.