River interlinking alters land-atmosphere feedback and changes the Indian summer monsoon

India’s major river basins (Ganga, Brahmaputra, Indus and others) support a largely agrarian society but face severe water stress due to climate change, population growth, pollution, and intensive water use. The Indian summer monsoon (ISM, June–September) supplies about 80% of annual rainfall, yet recent decades show declining mean rainfall and increased extremes, raising flood and drought risks. In response, India has proposed massive river interlinking projects to transfer water from surplus to deficit basins via ~15,000 km of canals and ~3000 reservoirs, aiming to move ~174 bcm annually and expand irrigation and hydropower. Prior experiences highlight potential benefits (e.g., groundwater stabilization) but also ecological and water footprint concerns. Crucially, planning assumes hydrological independence of basins, neglecting potential atmospheric feedbacks. The study poses the hypothesis that inter-basin water transfer, via increased irrigation, can modify land-atmosphere interactions and causally affect rainfall and hydrology in donor and adjacent basins, especially during the ISM.

The paper situates its work within literature documenting: (i) observed ISM changes—declining mean rainfall, rising intensity and variability of extremes; (ii) strong land-atmosphere coupling and precipitation recycling over India, particularly in the Ganga basin and recycled precipitation contributions to neighboring basins; (iii) ENSO’s control on ISM variability; (iv) impacts of aerosols and land-use/land-cover changes on monsoon rainfall via land-surface cooling; and (v) experiences and concerns regarding large-scale inter-basin transfers, including ecological impacts and altered water footprints. While prior studies established local and remote land-atmosphere feedbacks and recycled precipitation, no prior work explicitly examined feedbacks of river interlinking-induced irrigation on ISM spatial patterns and inter-basin hydrometeorological linkages.

Data and variables: The study uses 40 years of daily reanalysis (ERA-5 and MERRA-2; 1980/81–2019/20) over six major Indian basins (Ganga, Godavari, Mahanadi, Krishna, Narmada–Tapi combined, Cauvery). Basin-averaged time series are constructed for near-surface variables: soil moisture to root zone (SM), latent heat flux (LH), sensible heat flux (SH), precipitation (P), incoming shortwave radiation (SR), temperature (T), wind speed (WS, from U/V), and relative humidity (R, at 850 hPa). Time series are anomaly-transformed and tested stationary (ADF p<0.01). ENSO classification uses ONI (Oct–Dec) thresholds ±1 for El Niño/La Niña.

Causal discovery: Three approaches are applied: pairwise Granger causality (GC) with VAR models (max lag 10, AIC selection), Transfer Entropy (TE; binned with m=11, lags 1–10, daily Δt=1; significance by shuffled surrogates at 99%), and PCMCI causal network learning. PCMCI employs a two-stage approach: a PC stage (α=0.2) to identify parents (reduce conditioning dimensionality), followed by MCI tests (α=0.05) across lags up to 10; both linear (ParCorr) and nonlinear (CMI-knn) conditional independence tests are used. Inter-basin land–land networks use PCMCI with ParCorr on 40-year continuous daily data. Land–atmosphere networks use PCMCI with CMI-knn run separately for each monsoon season (122 days per year) to capture nonlinearities and seasonality. Links are deemed robust if significant (p<0.05) in more than 20 of 40 years.

Regional climate modeling: A modified WRF-ARW v3 coupled with CLM4 land model is used with an India-specific irrigation module representing flood irrigation and groundwater withdrawals. Domain: 59.5°E–107°E, 3.7°S–41.5°N; 25 km grid, 30 vertical levels; ERA-Interim provides IC/LBCs. Two experiment sets simulate 22 monsoons (1991–2012; 15 May–31 Oct) separately per year: (1) Control (CTL) with prescribed contemporary irrigation from agricultural census and reconstructed gridded withdrawals; (2) Irrigation (IRR) representing interlinking by increasing irrigated crop PFTs to 80% of grid area in targeted beneficiary regions, approximating ~30 Mha additional irrigated area (∼20 Mha Himalayan-fed, ∼10 Mha peninsular-fed) per official DPRs. Irrigation water amounts: ~600 mm for normal crops (~4 mm/day) and ~1450 mm for paddy (~12 mm/day) during Kharif. Spin-up: 16 days; sensitivity tests (15–60 days) show negligible differences. Physics: BMJ convection, Lin microphysics, YSU PBL, RRTM longwave, Dudhia shortwave; identical across CTL/IRR except irrigation. Analyses focus on ISM months, with emphasis on September due to minimal significant JJA changes and its hydrological importance for post-monsoon water availability.

Causal links under perturbation: TE is applied to IRR–CTL differences to trace causal connections from LH changes in irrigated regions to P changes in downwind drying regions, assessing consistency across all years and separately for El Niño/La Niña. Robustness: Networks are cross-validated with MERRA-2 and WRF-CTL outputs to assess reproducibility of inferred pathways.

• Causal connectivity: PCMCI on ERA-5 reveals statistically significant (typically 95–99% confidence) inter-basin causal links mediated via land–atmosphere–land pathways, indicating basins are not hydrologically independent. Some basins behave as moisture ‘donors’ (e.g., Ganga with many outgoing links), while others (e.g., Cauvery) as ‘recipients’ with many incoming links. Both positive and negative feedbacks occur between land variables of different basins.
• Land–atmosphere pathways: Within-basin links from LH to P (recycling) are evident in Godavari, Krishna, and Mahanadi; in others (Cauvery, Narmada–Tapi), LH links to R and T but not P, implying exported moisture. Cross-basin atmospheric links trace pathways such as LH_G → P_G → SR_M′ → LH_M, confirming inter-basin influence via atmospheric transport and thermodynamics.
• September rainfall impacts under interlinking: WRF-CLM4 IRR experiments show significant spatial reorganization of September precipitation. Median reductions reach up to about 12% (east-central India, Odisha region R4), 10% (east peninsular India, Andhra Pradesh R2; arid western Rajasthan R5), 9% (Gujarat R3), and 8% (central India). Moderate declines (~6.4%) occur in the western Himalayan foothills (Uttarakhand) and east-central India. Some areas experience median increases: up to ~12% in east India (Bihar, Jharkhand, eastern Uttar Pradesh, R8) and up to ~10% over parts of the Deccan plateau (Maharashtra, Telangana, R7). Despite heavy positive tails in distributions (sporadic large increases), central tendency indicates widespread drying in already water-stressed regions.
• Temperature and soil moisture responses: Regions with reduced September precipitation exhibit increases in daily maximum temperature up to ~1°C and declines in root-zone soil moisture of ~15 mm. While irrigated grids show higher LH, neighboring non-irrigated regions often dry due to feedbacks.
• ENSO modulation: Drying due to interlinking is more widespread and pronounced during La Niña years than El Niño years; however, central India can see improved rain in El Niño years. Arid western India tends to dry under both ENSO phases.
• Causal attribution under perturbation: TE analysis on IRR–CTL differences shows consistent causal links from increased LH in irrigated regions (southern peninsula A, western India B, Ganga region C) to reduced P in downwind drying regions (central-eastern India P1, central India P2, western India P3, western Himalaya P4). The LH_C → P1 link appears in all 22 years, highlighting the Ganga basin as a strong feedback hotspot.

The findings validate the hypothesis that enhanced irrigation from proposed inter-basin transfers perturbs land–atmosphere coupling, which then reorganizes atmospheric moisture pathways and rainfall across basins. This challenges the planning assumption of hydrological independence among basins. Reduced September rainfall in several donor and recipient regions implies post-monsoon river flows may decline, increasing water stress, altering water-demand projections, and potentially undermining the interlinking’s objectives. The study underscores that inter-basin transfers can induce both local recycling changes and remote teleconnections via temperature-driven circulation shifts and moisture advection, with ENSO modulating the magnitude and extent of impacts. These insights are critical for water-resources planning, agricultural risk management, and climate adaptation strategies, advocating for inclusion of land-atmosphere feedbacks and climate variability in design and operation of large hydrological infrastructure.

This work demonstrates that Indian river basins are interconnected through land–atmosphere feedbacks, and that river interlinking—by increasing irrigation—can reduce September rainfall by up to about 12% in already water-stressed regions, particularly during La Niña years. Using causal discovery on reanalysis and targeted regional climate modeling, the study provides mechanistic pathways linking irrigation-induced LH changes to downwind precipitation responses. The results imply that interlinking plans must be reassessed to incorporate atmospheric feedbacks, as these can alter monsoon spatial patterns, post-monsoon flows, and basin-scale water balances. Future research should: (i) evaluate feedbacks including altered land-to-ocean runoff impacts on monsoon; (ii) incorporate inter-catchment groundwater exchanges; (iii) explore sensitivity to alternative irrigation practices, operation rules, and crop calendars; (iv) assess robustness under future climate and land-use scenarios; and (v) integrate socio-economic risk assessments to guide adaptive, climate-informed inter-basin water management.

• Causal inference assumptions: PCMCI assumes stationarity, causal sufficiency, acyclicity, no contemporaneous links, and faithfulness; violations (e.g., noise dependencies) may yield spurious links. While key land and atmospheric variables are included, unobserved drivers may remain.
• External drivers: Low-frequency climate modes (e.g., ENSO) are addressed by year-wise seasonal analyses, but residual confounding cannot be fully excluded.
• Model structural uncertainty: WRF-CLM4 physics choices and irrigation parameterizations influence results; although evaluated and consistent with prior studies, alternative schemes could yield different sensitivities.
• Scope of processes: Feedbacks from reduced land-to-ocean runoff, and inter-catchment groundwater flows are not modeled; these could amplify or modulate monsoon responses.
• Temporal focus: Impacts are assessed for ISM seasons (emphasis on September) over 1991–2012; results may vary under different climates or land-use trajectories.
• Spatial idealization: Interlinking implementation is approximated by increasing irrigated fractions to match DPR targets; actual project phasing, operations, and water allocation dynamics are simplified.