📄 agriculture

Differential Climate Sensitivity of Maize and Wheat in Nepal: Evidence from Panel Econometrics and Machine Learning

This study utilizes panel econometrics and machine learning on 30 years of district-level data to demonstrate that wheat in Nepal is significantly more sensitive to temperature increases and drought stress than maize, with a 1°C rise reducing wheat yields by 8.7% compared to 4.2% for maize.

Original authors: Lokendra Khatri

Published 2026-06-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Lokendra Khatri

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine Nepal as a giant, diverse garden where two very different plants are trying to survive: Maize (corn) and Wheat. This garden is getting hotter and drier, and the author of this paper, Lokendra Khatri, wanted to figure out which plant is more likely to throw a tantrum and stop growing when the weather gets weird.

To do this, he didn't just guess; he looked at 30 years of data (from 1993 to 2022) across 77 different "neighborhoods" (districts) in Nepal. He used two powerful tools:

The "Accountant" (Econometrics): A strict math method that calculates exactly how much yield drops for every degree of heat.
The "Super-Computer" (Machine Learning): An AI that learns patterns to predict the future, acting like a weather detective.

Here is the simple breakdown of what he found:

1. The Two Plants Have Different Personalities

Think of Maize as a tough, sun-loving hiker. It is a "C4" plant, which is a fancy way of saying it has a built-in air conditioner that helps it handle high heat (up to about 38°C). It grows during the rainy summer monsoon.
Think of Wheat as a delicate, cool-weather flower. It is a "C3" plant, which means it prefers cooler temperatures (15–25°C). It grows in the dry winter months. If it gets too hot, its "pollen" (its way of making seeds) basically melts, and it stops producing grain.

2. The Big Discovery: Wheat is the "Canary in the Coal Mine"

The study found that Wheat is significantly more sensitive to climate change than Maize.

The Heat Test: If the temperature goes up by just 1°C (about 2°F), Wheat loses 8.7% of its harvest, while Maize only loses 4.2%.
- Analogy: Imagine two people walking in the sun. One (Wheat) starts sweating and stumbling after 10 minutes. The other (Maize) keeps walking for another 10 minutes before getting tired. Wheat gives up much faster.
The Drought Test: When the soil gets dry (measured by something called SPEI, which is like a "thirst meter"), Wheat's yield drops by 6.1%, while Maize drops by 3.8%.
The Verdict: Wheat is roughly twice as fragile as Maize when it comes to heat and drought.

3. Why Does This Happen?

The paper explains that Wheat is in a double bind:

Wrong Season: It grows in the winter (Rabi season), which is supposed to be cool. But in Nepal's lowlands (the Terai), winter is getting hotter.
Wrong Location: Wheat is mostly grown in the hot, flat plains (Terai). Maize is grown higher up in the hills, where it's naturally cooler.
The "Silking" Problem: Maize is sensitive to drought, but only during a specific short time when it's making kernels (silking). Wheat, however, is sensitive to heat and drought throughout its entire life cycle, especially when it's trying to set its seeds.

4. The AI "Crystal Ball"

The author used a machine learning model (XGBoost) to predict future harvests.

The AI was very good at guessing what would happen to Wheat (84% accurate) and pretty good at guessing Maize (79% accurate).
The AI confirmed the math: Temperature and drought are the main villains for Wheat, while Maize's problems are a mix of rain and heat.

5. What Should We Do? (The Paper's Recommendations)

Based on these findings, the paper suggests specific "first aid" for the garden:

Change the Seeds: We need to plant Wheat varieties that are built to handle heat, like upgrading a car's engine to run in a desert.
Water Where It Hurts: Since Wheat is the most vulnerable, irrigation (watering systems) should be prioritized for the Wheat-growing plains.
Early Warnings: Use the AI models to predict bad weather seasons early, so farmers know to prepare.
Insurance: Create a special insurance policy for Wheat that pays out automatically if the weather gets too hot or dry, based on the "thirst meter" (SPEI) numbers.

Summary

In short, while both crops are struggling with a changing climate in Nepal, Wheat is taking a much harder hit. It's like a house with a weak roof in a storm, while Maize is a house with a slightly weaker roof. The storm is getting worse, and the paper argues we need to reinforce Wheat's roof immediately to keep Nepal's food supply safe.

Technical Summary: Differential Climate Sensitivity of Maize and Wheat in Nepal

Problem Statement
Climate variability poses an escalating threat to Nepal's smallholder cereal agriculture, which supports 65% of the labor force and contributes approximately 27% to the GDP. While global literature establishes that rising temperatures and precipitation shifts reduce cereal yields, there is a lack of rigorous, comparative quantification of how these impacts differ between Nepal's two primary cereals: maize (Zea mays) and wheat (Triticum aestivum). Existing studies in Nepal often analyze single crops, rely on simplified time-series regressions lacking district-level fixed effects, or use crop simulation models calibrated on sparse data. Furthermore, no prior research has integrated machine learning with panel econometrics to compare the differential climate sensitivity of these crops across Nepal's diverse agro-ecological zones.

Methodology
The study constructs a balanced 30-year district-level panel (1993–2022) covering 77 districts, utilizing yield data from the Ministry of Agriculture and Livestock Development (MoALD) and climate data from NASA POWER (MERRA-2 reanalysis). The analysis employs a dual-framework approach:

Panel Econometrics: Fixed Effects (FE) regressions with cluster-robust and Driscoll-Kraay standard errors were applied to control for time-invariant district heterogeneity (e.g., soil quality, infrastructure) and cross-sectional dependence. The model estimates the marginal effects of growing-season mean temperature, total precipitation, the Standardized Precipitation Evapotranspiration Index (SPEI), heat stress days, and intra-seasonal precipitation variability on log-transformed yields. Robustness was verified through lagged climate variables, first-differenced models, and instrumental variable (IV) estimation using lagged SPEI anomalies.
Machine Learning (ML): XGBoost and Random Forest models were trained to predict yields, with hyperparameters tuned via 10-fold cross-validation. Model performance was evaluated on a temporal hold-out test set (2018–2022) using $R^2$ , RMSE, MAE, and MAPE. SHAP (Shapley Additive exPlanations) values were utilized to determine global and local variable importance and to identify non-linear thresholds in climate-yield relationships.

Key Contributions
This research represents the first integrated application of panel econometrics and machine learning to compare maize and wheat climate sensitivity in Nepal. Its specific contributions include:

Comparative Quantification: Providing simultaneous, district-level estimates of climate elasticity for both crops, revealing distinct vulnerability profiles.
Unified Framework: Integrating the SPEI drought index alongside temperature and precipitation metrics within a single econometric specification to capture complex moisture-temperature interactions.
Non-Linear Benchmarking: Utilizing XGBoost and SHAP analysis to capture non-linear climate-yield relationships and identify physiological thresholds that parametric models might miss.
Physiological Validation: Grounding statistical findings in C3 (wheat) versus C4 (maize) photosynthetic physiology to explain differential sensitivities.

Results
The analysis yields three primary findings:

Differential Thermal Sensitivity: Wheat is significantly more sensitive to temperature increases than maize. A 1°C rise in growing-season temperature reduces wheat yields by 8.7% ( $p < 0.001$ ), compared to a 4.2% reduction for maize. This difference is statistically significant and consistent across all robustness checks.
Drought and Variability Impacts: A one-unit decline in SPEI lowers wheat yields by 6.1% versus 3.8% for maize. Similarly, heat stress days and intra-seasonal precipitation variability exert a larger negative penalty on wheat. Wheat yields show higher inter-annual instability (CV = 0.23) compared to maize (CV = 0.19).
Predictive Performance and Drivers: XGBoost models achieved high predictive accuracy ( $R^2 = 0.84$ for wheat; $R^2 = 0.79$ for maize) on the hold-out test set. SHAP analysis confirms that wheat vulnerability is concentrated in temperature and drought metrics, whereas maize vulnerability is more distributed across rainfall volume and heat stress. Non-linear thresholds identified via SHAP indicate wheat yields decline sharply above ~22°C, while maize benefits from precipitation only above ~600 mm per season.

Significance and Claims
The paper claims that wheat is significantly more climate-sensitive than maize in the Nepalese context, a finding grounded in wheat's C3 physiology, lower critical temperature thresholds, and concentration in the high-temperature Terai plains. The study asserts that this differential sensitivity necessitates targeted policy interventions rather than uniform agricultural strategies.

Specifically, the authors argue that:

Policy Prioritization: Adaptation efforts must prioritize climate-resilient wheat varieties (e.g., heat-tolerant lines from CIMMYT) and targeted irrigation investments in the Terai, where thermal penalties are most severe.
Operational Forecasting: The high predictive accuracy of the ML models demonstrates the technical feasibility of operational, district-level yield forecasting using seasonal climate inputs, supporting early warning systems.
Risk Management: The estimated climate-yield elasticities provide actuarially credible inputs for designing index-based crop insurance products triggered by SPEI or temperature thresholds.
Climate Finance: The concentration of food security risk in wheat justifies dedicated adaptation finance streams within Nepal's Nationally Determined Contributions (NDCs) and Green Climate Fund allocations.

The study concludes that while both crops face climate risks, the unmitigated decline in wheat productivity in the Terai constitutes a critical national vulnerability that requires immediate, evidence-based intervention.