Source-linked AI summary
Monte Carlo-Based Ex-Ante Assessment of the Green Benefits of an AI-Driven Smart Agriculture Platform in Hainan
Zhaoyang Li, Ruijie Zhang, Zhaoji Sun, Lu Zhang
TL;DR
A unified quantitative framework for smart agriculture’s environmental benefits is lacking, especially for integrated platforms under uncertain parameters. This study combines cradle-to-farm-gate accounting with Monte Carlo simulation across three Hainan scenarios, finding substantial median reductions in four indicators and robust overall carbon decline. It also identifies influential parameters and reports a conditional full-adoption scope boundary.
Problem
A unified quantitative framework for assessing integrated smart agriculture platforms’ environmental benefits is still lacking, while existing assessments face substantial parameter uncertainty.
Method
The study builds cradle-to-farm-gate carbon accounting and propagates uncertain platform-intervention parameters by Monte Carlo simulation across weighted mango, winter-vegetable, and rice/nanfan scenarios.
Results
Under full adoption, median reductions are 23.5% for pesticide use, 21.0% for fertilizer, 16.5% for irrigation water, and 21.5% for carbon intensity.
Takeaways & Limitations
The framework supports ex-ante green-value assessment and pilot-observation design by quantifying uncertainty and prioritizing influential transmission parameters.
Takeaways & Limitations
The analysis is a conditional full-adoption ex-ante scenario that does not model learning curves, seasonal differences, or partial adoption, and it does not consider yield effects.
Abstract
from arXiv · showhide
Smart agriculture platforms are widely regarded as key carriers for implementing China's pesticide and fertilizer reduction, water-saving and carbon-reduction agendas, yet a unified quantitative framework for assessing their green value is still lacking. Taking an AI-driven decision platform for tropical agriculture as the object (integrating large-language-model question answering, multimodal pest diagnosis, IoT sensing, satellite remote sensing, and a closed-loop field record system), this study builds a cradle-to-farm-gate agricultural carbon accounting model covering pesticide and fertilizer production, field N2O, irrigation electricity and paddy CH4, translates platform interventions into quantifiable transmission parameters, and propagates parameter uncertainty by Monte Carlo simulation over three Hainan scenarios (mango, winter vegetable, rice/nanfan, area-weighted 40%:30%:30%). Under full adoption, median reductions are 23.5% (90% interval 15.0%-33.2%) for pesticide use, 21.0% (13.8%-28.9%) for fertilizer, 16.5% (10.9%-23.5%) for irrigation water, and 21.5% (16.1%-27.2%) for carbon intensity. Attainment probabilities are high for fertilizer reduction >=15% (90.6%) and clear carbon decline (98.1%), but only about 20% for aggregate water saving >=20%, favoring scenario-specific statements. Sobol first-order indices show soil-test recommendation and organic substitution jointly explain about 83% of the variance of aggregate carbon-intensity reduction. Convergence tests show 10,000 iterations stabilize all statistics; conservative/baseline/optimistic scenario bounds are reported. The framework offers a reproducible, calibration-ready methodology for ex-ante green-value assessment and pilot observation design.
1 INTRODUCTION
China’s agricultural policy agenda elevates pesticide and fertilizer reduction, water saving, and carbon reduction, while smart agriculture platforms are positioned to translate these goals into field practice. Yet unified platform-level assessment remains limited, and existing carbon assessments often rely on uncertain parameters or point estimates.
- Agricultural green development is framed as a key lever for China’s rural revitalization and dual-carbon goals.
- Smart agriculture platforms combine artificial intelligence, IoT, and satellite remote sensing to translate pesticide reduction, fertilizer reduction, water saving, and carbon reduction into field practice.
- A unified quantitative framework for assessing smart agriculture platforms’ environmental benefits is still lacking.
- Existing studies commonly examine individual interventions rather than platforms integrating diagnosis, medication, irrigation, fertilization, and remote sensing.
- Precision-agriculture carbon assessments face substantial parameter uncertainty, while many studies report single point estimates without intervals.
- This study develops a Monte Carlo framework for Hainan tropical agriculture to estimate benefit intervals, target-attainment probabilities, and parameters for pilot-observation prioritization.
2 Materials and Methods
The “Hongqiu Zhinong” platform integrates language-model assistance, multimodal pest diagnosis, IoT sensing, satellite analysis, and closed-loop field records. Its pesticide pathway links diagnosis and medication records to reduced blind, repeated, and excessive applications.
- The “Hongqiu Zhinong” platform integrates an agricultural LLM question-answering assistant, multimodal pest diagnosis, IoT sensors, satellite remote sensing, and closed-loop field records.
- Intelligent diagnosis reduces the share of blind spraying, while medication records and pre-harvest reminders target repeated and over-dosage applications.
2 MATERIALS AND METHODS
Platform interventions are represented as transmission parameters across mango, winter-vegetable, and rice/nanfan scenarios. Water, fertilizer, pesticide, and carbon pathways connect platform functions to reductions relative to conventional management.
- Fertilizer reduction combines soil-testing prescription advice, crop-stage topdressing, and organic substitution.
- Water saving uses sensor-driven drip or integrated water-fertilizer management for mango and vegetables, plus smart irrigation scheduling for rice.
- Carbon reduction transmits pesticide, fertilizer, and irrigation reductions to production and field emissions, with intermittent irrigation also mitigating paddy CH4.
- The scenarios comprise Hainan mango orchards, winter vegetable bases represented by cowpea rotation, and rice/nanfan breeding fields.
- The aggregate indicator weights mango, vegetable, and rice scenarios at 40%:30%:30%.
- Reduction rates are modeled relative to conventional management, with adoption rate α and intervention-specific coefficients transmitting platform effects.
2 MATERIALS AND METHODS
The study uses cradle-to-farm-gate carbon accounting and Monte Carlo uncertainty propagation across five emissions components and three Hainan scenarios. Sensitivity, convergence, endpoint scenarios, and reproducible computation support calibration and interpretation.
- The cradle-to-farm-gate model accounts for pesticide production, fertilizer production, field N2O, irrigation electricity, paddy CH4, and machinery carbon.
- Carbon after intervention is computed by applying reduction rates to pesticide, fertilizer, irrigation, and CH4 components while retaining machinery emissions.
- Paddy CH4 is included only for rice, field N2O is estimated as 1% of applied nitrogen, and irrigation electricity uses a 0.5366 kg CO2/kWh grid factor.
- Model inputs combine official methodologies, peer-reviewed literature, and expert judgment, with triangular distributions used for several scenario-specific baselines.
- Scenario-specific baselines are intended for calibration and replacement with pilot data.
- Monte Carlo simulation uses 10,000 iterations, Sobol first-order indices assess sensitivity, and convergence and endpoint analyses provide additional checks.
3 Results and Analysis
Monte Carlo results quantify four green-benefit indicators, target attainment, sensitivity drivers, convergence, and scenario-specific uncertainty for the platform. The baseline simulation indicates substantial pesticide, fertilizer, water, and carbon-intensity reductions, with water performance varying most by scenario.
- Probability intervals: 23.5% median pesticide reduction and 21.0% median fertilizer reduction were estimated from 10,000 Monte Carlo iterations.The 90% intervals were 15.0%–33.2% for pesticide reduction and 13.8%–28.9% for fertilizer reduction.
- Probability intervals: 16.5% aggregate irrigation water saving and 21.5% aggregate carbon-intensity reduction were estimated.The corresponding 90% intervals were 10.9%–23.5% and 16.1%–27.2%.
- Target attainment: 90.6% attainment probability was estimated for fertilizer reduction ≥15%, compared with 19.9% for aggregate water saving ≥20%.Pesticide reduction ≥20% had 73.8% attainment probability, while carbon-intensity reduction ≥15% had 98.1%.
- Scenario differences: 19.7% median water saving for mango drip irrigation contrasted with 11.2% for rice, supporting scenario-specific target setting.The low aggregate water-saving attainment mainly reflected rice’s limited saving potential and the adoption discount.
- Sensitivity and convergence: 0.527 and 0.303 were the Sobol first-order indices for soil-test reduction and organic substitution, jointly explaining about 83% of carbon-reduction variance.The remaining parameters jointly contributed less than 5%, and the total first-order effect was near 1.0.
- Sensitivity and convergence: 10,000 iterations produced negligible statistical error after convergence beyond 1,000 samples.The median carbon-intensity reduction stabilized around 21.5%, with P5/P95 near 16.2%/27.1%.
4 DISCUSSION
Scenario endpoints show positive carbon-intensity reduction across the modeled range, while lower adoption and parameter values can place other indicators below design targets. Simulated medians are generally consistent with literature or official ranges, with irrigation water saving as the exception noted by the authors.
- Scenario analysis: 11.3%–37.5% pesticide reduction, 6.5%–37.5% fertilizer reduction, 4.5%–38.0% water saving, and 6.9%–38.8% carbon reduction span the three scenarios.These are the conservative–baseline–optimistic endpoint ranges reported for the four indicators.
- Scenario analysis: 6.9% carbon-intensity reduction remains positive under the most unfavorable parameter combination.The authors describe the directional conclusion of clear carbon decline as robust.
- Scenario analysis: Lower adoption and parameter bounds place pesticide, fertilizer, and water indicators clearly below design targets.The discussion identifies adoption rate and field-level implementation as key conditions.
- External plausibility: All medians fall inside literature ranges except irrigation water saving, which is below the official upper bound because adoption discount is included.The authors present this comparison as supporting the plausibility of the model settings.
- External plausibility: 0.235 pesticide reduction, 0.210 fertilizer reduction, 0.197 mango irrigation-water saving, and 0.300 paddy CH4 reduction are the reported study medians.The comparison table pairs these medians with literature or official ranges.
4 Discussion
The framework differs from earlier single-technology assessments by representing all platform modules as quantifiable transmission parameters. It combines life-cycle assessment and Monte Carlo uncertainty propagation without requiring extensive field-experiment data, fitting early-stage assessment.
- Methodological novelty: The framework extends earlier single-technology assessments to a whole-platform representation of green benefits.This is identified as the key difference from earlier assessments.
- Methodological novelty: All platform modules are encoded as quantifiable transmission parameters, enabling whole-platform benefit intervals.The modules include question answering, diagnosis, sensing, remote sensing, and closed-loop records.
- Methodological basis: Monte Carlo uncertainty propagation is combined with the mature life-cycle-assessment paradigm.The approach is described as avoiding hard dependence on field-experiment data.
- Methodological basis: Avoiding hard dependence on field-experiment data suits early-stage assessment.The statement concerns the framework’s intended assessment setting.
5 CONCLUSIONS
The framework decomposes uncertainty across parameters, transmission structures, and adoption scenarios, identifying local calibration priorities and scenario-specific limits on interpreting green benefits.
- Uncertainty analysis: Three uncertainty classes—parameter, structural, and scenario uncertainty—are represented with triangular distributions, Sobol decomposition, and three-scenario analysis.Grade-C parameters include blind-spray share ranges of 0.15–0.50.
- Uncertainty analysis: Wide Grade-C parameter ranges are the main source of pesticide-result uncertainty, so pilot observation should first collect local Hainan data to tighten them.
- Scenario interpretation: Aggregate water saving of 20% has only about 20% attainment probability, favoring scenario-disaggregated reporting.Mango drip and vegetable water-fertilizer scenarios show median saving of 17%–20%, whereas the rice scenario is limited; an aggregate 10%–25% interval is more robust.
- Scenario interpretation: Paddy intermittent irrigation offers additional CH4 mitigation potential in Hainan’s double-cropped rice and nanfan fields, contributing about 14% of carbon-intensity-reduction variance.The study characterizes this as a low-cost, high-benefit green lever.
- Limitations and outlook: The analysis is a conditional full-adoption ex-ante scenario that omits learning curves, seasonal differences, partial adoption, and yield effects.Not accounting for yield-scaled intensity is described as conservative, while carbon baselines and paddy CH4 factors require local validation.
- Limitations and outlook: Future work will observe the four indicators in pilot plots, update parameter posteriors using Bayesian methods, and extend accounting to post-harvest and ecological product-value pathways.
5 Conclusions
Under full-adoption simulation, the platform produces median reductions across pesticide, fertilizer, irrigation-water, and carbon-intensity metrics. Fertilizer and carbon targets are highly attainable, while aggregate water saving is less robust; sensitivity analysis identifies fertilizer-related parameters as key observation priorities.
- Conclusions: 23.5% pesticide reduction, 21.0% fertilizer reduction, 16.5% aggregate irrigation water saving, and 21.5% carbon-intensity reduction are the median full-adoption benefits.The 90% intervals are 15.0%–33.2% for pesticide reduction, 13.8%–28.9% for fertilizer reduction, and 10.9%–23.5% for aggregate irrigation water saving.
- Conclusions: About 20% attainment probability for aggregate water saving ≥20% supports mango/vegetable scenario reporting rather than a single aggregate claim.
- Conclusions: Soil-test rate and organic-substitution coefficient contribute about 83% of aggregate carbon-intensity-reduction variance.
- Conclusions: Pilot observation should prioritize fertilizer-input and paddy water-layer indicators.
- Conclusions: Monte Carlo analysis with convergence and three-scenario checks provides a reproducible, calibration-ready reference for assessing smart-agriculture platform green value.
Data and Code Availability
The paper provides scripts and sample data for reproducing its results, using a fixed random seed.
- Data and Code Availability: All simulation scripts and sample data are available in the delivery directories scripts/ and output/.
- Data and Code Availability: A fixed random seed of 20260906 supports reproducibility of the reported results.