Divergent effectiveness of irrigation in enhancing food security in droughts under future climates with various emission scenarios

The study addresses how effective irrigation is in mitigating drought-induced wheat yield losses under future climate change scenarios in China. While droughts are projected to intensify and expand due to climate change, prior assessments often used meteorological or soil-moisture-based drought indices that could not isolate irrigation’s role. The authors adopt the Modified Palmer Drought Severity Index (MPDSI), which includes irrigation in its water balance, to examine the differential impacts on irrigated versus non-irrigated systems. Focusing on wheat—the most vulnerable staple crop in China—the research aims to quantify irrigation’s effectiveness across varying drought intensities and Representative Concentration Pathways (RCP2.6, RCP6.0, RCP8.5), thereby informing water resource allocation and food security strategies under climate change.

Prior work documents substantial impacts of droughts and extremes on crop yields (e.g., 10% average cereal yield reduction and >40% interannual variability). Traditional statistical approaches linking drought indices to yield are sensitive to the chosen index and often fail to separate irrigation’s mitigating role. Agricultural drought indices based on soil moisture conflate effects of drought, irrigation, and management. Studies in North China Plain indicate irrigation can halve drought-induced yield loss, and supplementary irrigation post-drought can cut yield reduction by >80% and crop failure by >50%. However, the effectiveness of irrigation under future climate scenarios has been insufficiently examined. This study builds on these gaps using MPDSI to explicitly incorporate irrigation and compares multiple modeling approaches to assess yield impacts under future climates.

Study area and crop: China, focusing on wheat across nine agricultural subregions (NASR, LP, NCP, HHHP, QTP, SBSR, YGP, SC, MYP). Scenarios and climate data: Outputs from four ISIMIP GCMs (GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5) were used for historical (1860–2004) and future (2005–2099) RCP2.6, RCP6.0, and RCP8.5 scenarios. Drought characterization: The Modified Palmer Drought Severity Index (MPDSI) incorporates irrigation into the water balance and uses adaptively estimated empirical parameters for regional conditions. Future irrigation was assumed constant due to uncertainty in human activities. MPDSI values were classified into drought levels (slight to extreme) and drought events were identified using Run Theory on grids. For each event, three descriptors were derived: intensity (cumulative MPDSI), duration (consecutive months with at least slight drought), and affected area (fraction of territory meeting event criteria). Trend analysis: Modified Mann–Kendall (MMK) test, accounting for autocorrelation via pre-whitening when lag-1 autocorrelation >0.1; significance at 95% and 99% thresholds. Yield modeling: Three complementary approaches were used. (1) Multiple Linear Regression (MLR): Explanatory variables were drought-affected area, intensity, and duration; response was detrended short-term climate yield. Detrending used the Hodrick–Prescott filter to separate trend yield (technology/management), climate yield, and noise. (2) Deep Learning (DL): Long Short-Term Memory (LSTM) networks were trained on time-series inputs including meteorological data, soil data, mean MPDSI, and wheat yield. Inputs were normalized; backpropagation optimized parameters to minimize loss against historical yields. (3) EPIC crop model: Process-based simulation of wheat growth and water stress under different drought conditions and management; parameters were tuned (trial-and-error) based on prior studies. Two EPIC scenarios were designed to evaluate drought impacts consistent with MPDSI-characterized intensities. Cross-method synthesis: Results from MLR, DL, and EPIC were compared to reduce uncertainty; figures report yield changes by drought intensity and region for historical and future scenarios. Irrigation effectiveness analysis: Using mapped irrigated areas, yield changes under drought were computed for irrigated zones and for comparable neighboring non-irrigated areas to assess irrigation’s mitigating effect across intensities and scenarios. Data sources: ISIMIP climate, Soil Available Water Content from CRensed, national wheat yield statistics, and irrigation data from the National Agricultural Weather Station (via correspondence).

- Drought evolution: Future droughts intensify relative to the historical period across China. Under RCP6.0, the drought-affected area increases to ~30% (vs. ~10–20% historically). Under RCP2.6, drought duration commonly lengthens to 10–25 months with affected areas 30–60% and more severe events. RCP8.5 exhibits more frequent extreme droughts and fewer mild/moderate events, with strong drying especially in NCP and southeastern QTP. - Historical impacts: During 1860–2004, wheat yield reductions during droughts were generally within 15%, with regional variability; mild droughts often had little effect or small increases in some regions. - RCP2.6: Wheat yield losses increase relative to history. Notable losses include up to 12% (HHHP), 25% (MYP), 25% (LP), 50% (NCP), 35% (SBSR), and 51% (YGP); SC shifts from slight gains historically to losses up to 28%. - RCP6.0: Losses are generally less than RCP2.6 but greater than historical. Examples: MYP maximum loss ~15% (vs. 25% under RCP2.6), SC ~14% (vs. 28% under RCP2.6); persistent losses in LP (~16%), NCP and QTP up to ~25%; SBSR and YGP largest losses ~21% and ~27%, respectively. - RCP8.5: Severe widespread declines. Maximum losses reach ~80% in NCP and YGP; 60–70% in LP, HHHP, MYP, SC, SBSR; ~35% in QTP; ~17% in NASR. Abstract-level national projection notes drought-induced wheat yield loss of ~32–49% under RCP8.5. - Irrigation effectiveness (spatial comparison): Irrigation substantially mitigates drought losses historically and under RCP2.6/RCP6.0 but not under RCP8.5. • Historical: Non-irrigated areas saw up to ~15% reduction during droughts, whereas irrigated areas experienced ~8–17% increases; irrigation benefit ~23–32% relative difference. • RCP6.0: Irrigated areas had 0–10% decreases vs. 21–44% decreases in nearby non-irrigated areas (about 2–4× larger losses without irrigation). • RCP2.6: Both irrigated and non-irrigated areas lose yield, but irrigated losses are ~10–15% lower. • RCP8.5: Losses are large in both irrigated and non-irrigated areas, with little discernible difference—indicating limited irrigation effectiveness. - Irrigation vs. drought intensity: In irrigated areas, mild and moderate droughts still yielded gains (~+7% and +1%), whereas non-irrigated areas had losses (~−3% and −13%). Under severe droughts, irrigation reduced losses by ~15% compared to non-irrigated areas. Under extreme droughts, irrigation provided only ~1–3% difference, indicating limited buffering capacity.

By incorporating irrigation into drought characterization (MPDSI) and comparing irrigated versus non-irrigated areas, the study demonstrates that irrigation substantially enhances resilience of wheat yields to mild and moderate droughts and under low-to-medium emissions scenarios (RCP2.6, RCP6.0). However, under a high-emission future (RCP8.5), with more frequent and intense extreme droughts, irrigation’s mitigating effect diminishes to near-indiscernible levels. This indicates that while irrigation remains a critical adaptation to stabilize yields in many contexts, it cannot fully offset the impacts of severe climate change. The findings emphasize that climate change mitigation (limiting emissions) is essential for maintaining the effectiveness of irrigation-based adaptation strategies and for safeguarding food security. They also suggest that water resource management must account for increasing drought extent and duration, especially where potential evapotranspiration rises, and that adaptation portfolios should extend beyond irrigation alone under high-emission scenarios.

The paper quantifies the changing effectiveness of irrigation in mitigating drought-induced wheat yield losses across China under historical and future climates. It shows that irrigation is effective for mild to severe droughts and under RCP2.6 and RCP6.0, but its benefits largely vanish under RCP8.5, where extreme droughts dominate and widespread yield losses occur. The main contributions are: (1) explicit incorporation of irrigation in drought assessment via MPDSI, (2) multi-method triangulation of yield impacts (MLR, LSTM, EPIC), and (3) spatial comparison of irrigated versus non-irrigated areas across intensities and scenarios. Policy implications include prioritizing climate mitigation to preserve irrigation’s efficacy, and adapting water management to increasing drought risk. Future research should relax the assumption of constant future irrigation by exploring dynamic irrigation scenarios, water availability constraints, and cost-effectiveness; expand to additional crops and regions; and integrate broader adaptation measures (cultivars, agronomy) with irrigation strategies.

- Future irrigation assumed constant, which may not reflect evolving water availability, policy, infrastructure, or farmer behavior. - Results depend on four ISIMIP GCMs and three RCP scenarios; model and scenario uncertainties remain despite multi-method yield assessment. - Focus limited to China and wheat; generalizability to other crops/regions may vary. - MPDSI parameterization and data inputs (e.g., irrigation dataset obtained via personal communication) may introduce uncertainties. - EPIC parameter tuning and DL model training depend on available historical data; structural and data limitations may affect projections. - Extreme drought impacts under RCP8.5 may involve compounding factors (e.g., heat stress) not fully captured by irrigation-focused mitigation.