El Nino Southern Oscillation-Driven Variability in Vegetation Indices and Carbon Fluxes over Rajasthan, India: A Non-parametric District-level Analysis (2000-2024)

Technical Summary: ENSO-Driven Variability in Vegetation and Carbon Fluxes over Rajasthan (2000–2024)

Problem Statement
This study addresses the need to quantify how El Niño Southern Oscillation (ENSO) events and monsoon rainfall influence kharif-season vegetation conditions and terrestrial carbon fluxes across the diverse aridity gradients of Rajasthan, India. While the dominance of rainfall in driving vegetation is established, the specific magnitude of ENSO impacts on Gross Primary Productivity (GPP) and Net Primary Productivity (NPP) at the district level, independent of rainfall variability, remains less characterized. Furthermore, there is a gap in understanding how these climatic drivers affect carbon assimilation relative to canopy greenness (NDVI), particularly in hyper-arid and arid zones where interannual variability is high.

Methodology
The research employs a fully non-parametric statistical framework analyzing 25 years of data (2000–2024) across all 33 districts of Rajasthan.

• Data Sources: The study integrates four MODIS Collection 6 products—MOD13Q1 (250 m NDVI/EVI), MOD17A2H (8-day GPP), MOD17A3HGF (annual NPP), and MCD12Q1 (Land Use/Land Cover)—with IMD 0.25° gridded rainfall data and NOAA ONI indices.
• Preprocessing: A dynamic agricultural mask (IGBP classes 12 and 14) derived from MCD12Q1 is applied annually to isolate agricultural pixels, preventing the attenuation of ENSO signals by natural dryland vegetation.
• Statistical Framework:
  • Trend Detection: Mann-Kendall (MK) tests with Sen slope estimators identify long-term trends.
  • Correlation Analysis: Kendall tau and Spearman rho assess bivariate associations. Partial Spearman correlations control for rainfall (RF) to isolate independent ONI effects on GPP.
  • Modeling: District-level predictive equations are derived using an iterative Theil-Sen regression approach. The model structure is: $GPP = \alpha + \beta_{RF} \cdot (RF - RF_{med}) + \beta_{ONI} \cdot (ONI - ONI_{med})$.
  • Validation: Model skill is quantified via Leave-One-Out Cross-Validation (LOOCV) yielding Mean Absolute Percentage Error (MAPE), RMSE, and systematic bias.
  • Forecasting: 2026 outlooks utilize 1000-replicate bootstrap resampling to generate 90% confidence intervals for district-level projections.

Key Contributions

1. District-Level Attribution: This is the first study to provide a fully non-parametric, district-level attribution of ENSO signals in MODIS vegetation and carbon flux products for the entirety of Rajasthan.
2. Decoupling Rainfall and ENSO: The study successfully isolates independent ENSO effects on GPP after controlling for rainfall, identifying specific "ONI-sensitive" districts in the arid and semi-arid west.
3. GPP:NDVI Discrepancy: It quantifies a ~2:1 suppression ratio of GPP relative to NDVI during El Niño events, highlighting that carbon assimilation declines faster than canopy greenness due to stomatal closure.
4. Operational Forecasting: The development of Theil-Sen prediction models with quantified uncertainty (LOO-MAPE) provides a framework for generating actionable seasonal agricultural early warnings.

Results

• ENSO Impacts: El Niño events suppress state-mean kharif NDVI by 3.6% and GPP by 7.0% on average. The impact is most severe in the hyper-arid Jaisalmer district, with deficits reaching 14.3% for NDVI and 21.6% for GPP. La Niña events produce symmetric surpluses of 3–18% in GPP.
• Correlations: Rainfall-NDVI correlations are universally significant across all 33 districts (mean $\rho$ = 0.935). However, independent ONI effects on GPP are significant only in six arid and semi-arid western districts (Bikaner, Churu, Ganganagar, Hanumangarh, Jaisalmer, and Jodhpur), with partial correlations ranging from -0.42 to -0.62. In the sub-humid east, ENSO influence is primarily mediated through rainfall.
• Long-term Trends: Significant positive Mann-Kendall trends in NDVI and GPP are detected in 16 of 33 districts. The highest greening rates occur in the sub-humid southeast (mean +5.4 g C m⁻² yr⁻¹ for GPP), attributed to agricultural intensification and CO₂ fertilization.
• Model Performance: The Theil-Sen models achieve a mean LOO-MAPE of 6.2% for NDVI and 17.4% for GPP. Performance is best in the sub-humid southeast (<12% MAPE) and poorest in hyper-arid Jaisalmer (31.5%) and arid Churu (41.4%) due to high interannual noise and detection floor issues.
• 2026 Forecast: Based on an ONI of -0.1 (FMA 2026), the state-mean kharif GPP is forecast at 249.5 g C m⁻² (anomaly -3.2%). Thirty of 33 districts are classified as NEAR NORMAL, while Jaisalmer, Churu, and Jhunjhunu are forecast as BELOW NORMAL.

Significance and Claims
The paper claims that its findings offer critical insights for satellite-based carbon monitoring and agricultural planning. Specifically, it asserts that NDVI-based yield models calibrated in average or wet years will systematically overestimate kharif production during El Niño events if the disproportionate suppression of carbon flux (GPP) relative to greenness (NDVI) is not accounted for.

The study positions its Theil-Sen equations as a simple correction framework. By integrating pre-season ONI forecasts with district-median rainfall projections, planners can adjust GPP estimates weeks before harvest. The authors conclude that the identified "ONI-sensitive" districts provide a basis for differentiated crop advisory strategies that incorporate Sea Surface Temperature (SST) outlooks alongside rainfall forecasts. The work is presented as operationally ready for seasonal agricultural planning, input supply management, and drought declaration at the sub-district level, provided that real-time rainfall forecasts are propagated through the established equations.

Note: The "Introduction" section of the provided text contains a recommendation regarding the "Force Account approach" for school infrastructure in Tanzania, which appears to be an editorial error or a copy-paste artifact unrelated to the scientific content of the Rajasthan study. This summary excludes that unrelated content as it does not pertain to the study's problem, methodology, or findings.