Eco-Chain: A Closed-Loop Multimodal AI and Consortium Blockchain Framework for Verifiable Plant Disease Diagnosis and Sustainable Adaptive Agriculture

Eco-Chain: A Closed-Loop Multimodal AI and Consortium Blockchain Framework for Verifiable Plant Disease Diagnosis and Sustainable Adaptive Agriculture

Sania Parkar, Utsav Pandey

University of Mumbai, Mumbai, Maharashtra

Abstract – Global food security is imperiled by a 40% annual loss in crop production due to pests and diseases, costing the global economy over $220 billion annually. While Deep Learning (DL) has achieved

>95% accuracy in image-based disease classification, current solutions suffer from "contextual blindness" and a lack of accountability in treatment outcomes. This paper proposes Eco-Chain, a novel ecosystem integrating Multimodal Vision-Language Transformers (VLT) and a Consortium Blockchain. Unlike static diagnostic tools, Eco-Chain generates time-bound Care Regimens and enforces a closed-loop feedback mechanism. Farmer feedback is analyzed via Natural Language Processing (NLP) to calculate a Regimen Success Score (RSS), recorded on an immutable ledger. This system incentivizes "Proof of Care" through tokenomics, creating a verified, global knowledge base that empowers plants to adapt to climate change, thereby securing human existence.

Keywords: Multimodal Deep Learning, Vision-Language Transformers, Consortium Blockchain, Smart Contracts, Regimen Success Score (RSS), Sustainable Agriculture, Tokenomics.

1. INTRODUCTION

   1. Biological Basis: The Leaf as a Sentinel

      Plants act as the foundation of human existence, processing energy to sustain life. In this physiological hierarchy, leaves are the primary organs for photosynthesis and transpiration. Consequently, they are often the first biological indicators of systemic distress. Whether through chlorosis (yellowing), necrosis (death of tissue), or crisping edges, leaves visually manifest internal pathologies. In woody plants or succulents where leaves are modified or absent, these symptoms translate to the stem or bark. Therefore, accurate diagnosis must prioritize these vegetative organs to catch diseases before they impact reproductive organs like flowers or fruits.

   2. Problem Statement

      Despite the projected need to increase food production by 50% by 2050, we lose nearly half our harvest to pests and diseases. Current State-of-the-Art (SOTA) solutions primarily rely on Computer Vision (CV) to classify diseases from images. However, these systems face four critical failures:

      I. Limited Contextual Understanding: They often fail to incorporate crucial non-visual cues, such as the farmer's textual or vocal descriptions of symptoms, environmental conditions, or historical plant health, leading to potentially incomplete or inaccurate diagnoses.

2. Lack of Actionable Regimens: While diagnosing diseases, many systems provide generic recommendations rather than personalized, time-bound, and verifiable treatment regimens tailored to specific plant species, disease severity, and local conditions.

3. Absence of a Validated Feedback Loop: There is a significant void in systematically collecting, verifying, and integrating

   farmer feedback on the success or failure of recommended treatments. Without this crucial feedback, AI models cannot continuously learn, adapt, and improve their recommendations over time, hindering their ability to address evolving agricultural challenges and concept drifts.

4. Data Silos and Trust Deficit: The valuable data generated (diagnoses, regimens, outcomes) often resides in centralized, proprietary systems, limiting transparency, immutability, and equitable access for broader research and policy-making. This also prevents incentivization for data contributors and verifiable data sharing among stakeholders (e.g., government, research institutions, farmers).

1. Proposed Solution

   We propose a Multimodal Diagnostic & Governance System. By feeding image data and user- narrated symptoms (processed via NLP) into a unified neural network, we generate a high-precision diagnosis and a specific "Care Regimen." This regimen is tracked via a Consortium Blockchain, which locks the data until the user provides feedback. Successful regimens verified by the system are permanently recorded on the ledger, creating a transparent history of "what works" for future research.

   Our proposed solution, Eco-Chain, aims to:

   Enhance Diagnostic Accuracy: Leverage Multimodal VLTs to fuse visual data with linguistic context (symptoms, history). Enforce Accountability: Use Blockchain Smart Contracts to lock a diagnosis and release rewards only upon verified treatment success.

   Incentivize Care: Calculation of the Regimen Success Score (RSS) from user feedback, which triggers token rewards based on "Proof of Care" and creates a trusted, immutable record of proven treatment methods by rewarding farmers for healing their crops, not just reporting sickness.

2. The Biological Sentinel: Abrupt Evolution and Pathogen Adaptation Building on the physiological cues identified in Section I.I, it is critical to recognize that climate change is fundamentally reshaping the dynamics of host-pathogen interactions, leading to what

   researchers define as "ecosystem drift."I Rising atmospheric levels and shifting precipitation patterns create conducive microclimates for the rapid multiplication of pathogens.2

   Pathogenic Range Shifts: Fungal pathogens are no longer geographically localized. Taxa such as Alternaria and Fusarium are predicted to increase significantly under global warming, drifting into previously disease-free temperate and boreal biomes.

   Abrupt Evolutionary Responses: Research indicates that some annual plants, such as Brassica rapa, advance flowering dates by nearly 8.6 days in response to multiyear droughts. While this

   allows for immediate survival, it often leads to a synchronization failure between the plant and its pollinators, permanently severing ecological dependencies.3

   Virulence Acceleration: Warmer temperatures accelerate the life cycles of bacteria and fungi, facilitating the acquisition of new phenotypes via horizontal gene transfer or mutation. This

   necessitates the Eco-Chain's "Proof of Care" model to track whether established resistance genes in host plants are being broken down by these newly resilient strains.

   1. Phenotype-to-Genotype (P2G) Feedback Loop

      To transition from managing plant survival to facilitating long-term adaptation, the Eco-Chain

      integrates a Phenotype-to-Genotype (P2G) feedback loop. By recording "Verified Success" outcomes for specific plant varieties under extreme climatic stress, the system acts as a repository for adaptive phenotypes. This metadata should be cross-referenced with genomic markers to identify specific

      resilience genes in successful plants, providing a data pipeline for researchers to utilize marker- assisted breeding or CRISPR- based gene editing to secure future food supplies.

      1. REVIEW OF LITERATURE

         Current research can be categorized into three isolated silos, which our paper aims to bridge.

         1. The Superiority of Multimodal Models in Agriculture

            Research consistently demonstrates that combining data modalities significantly outperforms single- source learning. A 2024 comparative study revealed that while RGB-only models achieved -87.5% accuracy in complex field conditions, multimodal models combining RGB images with contextual or spectral data reached 96.8% accuracy. The integration of Vision Transformers (ViT) with Large Language Models (LLMs) allows for "visual question answering," enabling the system to "reason" about a disease based on multiple inputs rather than just "recognizing" a pattern in an image.

         2. Integrated Agricultural Systems and their Limitations

            Platforms like "AgriAI" have attempted to merge detection with virtual assistance and feedback mechanisms. While these systems offer a user-friendly interface for reporting, they lack a rigorous, tamper-proof validation mechanism. The feedback remains subjective and is stored in centralized databases, making it vulnerable to manipulation and unsuitable for high-stakes government or research applications.

         3. Blockchain for Data Integrity Beyond Supply Chains

            While the blockchain in agriculture market is projected to grow significantly, current implementations focus heavily on supply chain traceability tracking a product from farm to fork. The application of blockchain for intellectual provenance tracking the efficacy of agronomic advice remains an untapped frontier. Our research adapts the "Digital Twin" concept from precision health, where a patient's treatment path is secured and tracked on a blockchain, and applies this rigor to plant pathology.

            Research Gap: No existing academic architecture connects Multimodal AI diagnosis with a Smart Contract-enforced Regimen Verification Model to create a decentralized, self-correcting agricultural knowledge base. This is the unique contribution of Eco- Chain.

            2.4 Review of Digital Traceability and Tokenomics Models

            Traceability in agriculture traditionally focused on tracking commodities (e.g., seed batch number, harvest date). However, Eco- Chain introduces the concept of Care Traceability tracking the efficacy of the intervention. The proposed Proof of Care (POC) Tokenomics is a novel variant of the Proof-of- Engagement (PoE) model. Instead of rewarding simple clicks or login time, PoC rewards the verified positive change in the real world. The token, AgriToken, functions as a utility token, redeemable for subsidized inputs (seeds, fertilizer) or government grants.

            This mechanism directly solves the "last mile" problem of data integrity: the economic incentive (AgriToken) is tied directly to the Regimen Success Score (RSS), ensuring farmers become active participants in the verification process. This transforms the farmer from a passive data source into a compensated ecological auditor.

      2. REQUIREMENT ANALYSIS

         1. Functional Requirements

            I. Input Module: Must accept high-resolution leaf/stem images and audio/text descriptions (support for multi- language via Machine Translation).

            1. Analysis Engine: Must correlate visual lesions with textual descriptors (e.g., "dry," "sticky") to predict disease and generate

               a JSON-structured Regimen (Nutrients, Timestamp for recovery).

            2. Governance Module: A permissioned blockchain (Consortium) where the Government acts as the Certificate Authority (CA).

            3. Incentive Layer: A smart contract that distributes utility tokens upon verification of a high "Regimen Success Score."

         2. Non-Functional Requirements

            Scalability: The blockchain must handle high transaction throughput (using Hyperledger Fabric or similar).

            Usability: The interface must be accessible to farmers with varying literacy levels (voice-first design).

            Eco-Efficiency: The consensus mechanism should be low-energy (here, Proof of Care, not Proof of Work).

            To illustrate the system's immersive flow, we present the user journey of "Riya," a tomato farmer.

            Step 1: The Distress Signal (Multimodal Input)

            Riya notices her tomato leaves are curling and yellowing. She opens the Eco-Chain app.

            Visual: She snaps a high-res photo of the leaf.

            Vocal Context: She presses the mic button and says (in her local dialect), "The leaves started curling two days after the heavy rains, and the soil feels very sticky."

            Processing: The system's Translation Module converts audio to text. The VLT Engine fuses the image features (curling) with the text embedding ("heavy rains" + "sticky soil").

            Step 2: The Diagnosis & Smart Regimen

            A visual-only model might guess "Yellow Leaf Curl Virus." However, Eco-Chain's multimodal engine detects the "heavy rain" context and correctly diagnoses "Root Asphyxiation due to Waterlogging" (a physiological disorder, not a virus).

            Output: The system generates a JSON-structured Smart Regimen:

            Action: "Stop irrigation immediately. Aerate soil by hoeing 2 inches deep."

            Nutrient: "Apply Foliar Calcium spray to reduce stress."

            ETA: "Visible recovery expected in 5 days."

            Step 3: The Blockchain Lock

            A Smart Contract is instantiated on the Consortium Blockchain.

            Record: Hash(User:Riya, Diagnosis:Root_Asphyxiation, Regimen_ID: 887, Status: PENDING)

            Timer: The contract locks this record for 5 days.

            Step 4: The Proof of Care (Feedback Loop)

            On Day 5, the app notifies Riya. She uploads a new photo of the recovering plant and says, "The leaves are straightening out, and the green color is back."

            Validation: The system runs Sentiment Analysis on her voice note and Change Detection on the image.

            Scoring: It calculates a Regimen Success Score (RSS) of 9.2/I0.

            Step 5: Settlement & Incentivization

            The Smart Contract executes:

            Status Update: PENDING — VERIFIED_SUCCESS

            Reward: 50 AgriTokens are transferred to Riya's wallet.

            Global Sync: The anonymized success data is written to the global ledger for research institutes to analyze.

         3. Environmental Footprint: The "Green AI" Arithmetic

            To satisfy the "Eco-Efficiency" requirement, the Eco-Chain framework adopts a holistic Software Carbon Intensity (SCI) metric, which accounts for both operational energy and embodied carbon (emissions from device manufacturing). Embodied carbon constitutes a massive upfront environmental cost, particularly for the billions of edge devices required to scale agricultural AI.

            Where O is operational energy, I is carbon intensity, M is embodied carbon, and R is the functional unit (per diagnosis).

            Table 1: Environmental Cost Benchmarks for Agricultural AI

            Activity	Resource Consumption	Carbon Equivalent (CO2e)
            Training BERT-large	-650,000 kWh	284,000 kg
            Training GPT-3	I,287 MWh	552 tons
            Single Median Prompt	0.24 Wh (Inference)	< 0.I g
            Global Data Centers	3% Global Electricity	-Total emissions of Brazil

            Sustainable Mitigation Strategies:

            • Temporal Scheduling: Scheduling training tasks during high renewable energy generation periods can cut emissions by half.8

            • Hardware-Aware Co-Design: Utilizing the CATransformers framework to co-optimize model and hardware design can reduce total carbon by up to 30% without sacrificing performance.

         4. Socio-Technical Integration: Farmer Learning Hubs

            To bridge the digital divide and ensure system usability, Eco-Chain incorporates "Learning Hubs" that treat traditional farmer wisdom as a valuable asset rather than a deficit.7

            UI/UX for Low Literacy: Abstract iconography is replaced with literal, functional illustrations that depict farm equipment exactly as it appears in the field.9 Interfaces are designed in grayscale first to ensure readability under high-glare sunlight where color distinction is often lost.9

            Voice-First Interaction: On-device Automatic Speech Recognition (ASR) allows farmers to report symptoms in their native dialects, which significantly increases engagement among those with limited textual literacy.I0

            The TOT Model: Training of Trainers (TOT) programs increase digital tool uptake by 20% by empowering local community leaders to serve as intermediaries between digital insights and cultural practices.

      3. DESIGN AND PLANNING (SYSTEM ARCHITECTURE)

         1. The Three-Layer Integrated Architecture (Sensing, Thinking, Acting)

            The Eco-Chain operates as a true closed-loop feedback system, ensuring the AI is constantly learning from the real-world outcomes verified on the blockchain.

            1. Sensing Layer (Mobile Client)

               This layer captures the raw, multimodal input. The farmer uses the mobile app to capture an image of the symptomatic plant and simultaneously records a voice note describing the issue. A lightweight Whisper-Lite model performs on-device ASR to convert the speech to text, ensuring privacy and minimizing network bandwidth reliance for the initial input. The image, audio, and text are packaged and sent to the Thinking Layer.

            2. Thinking Layer (Multimodal AI Engine)

               This is the central decision-making unit. It receives the multimodal input package and processes it through the Agri-VLT model. To move beyond "contextual blindness," the model employs a Cross- Attention Fusion mechanism where visual and linguistic tokens interact dynamically.

               In this architecture, Q (Query) represents the linguistic embedding (e.g., "sticky soil"), while K (Key) and V (Value) represent the visual feature patches of the symptomatic leaf. This allows the model to "attend" specifically to image regions that align with the narrated symptoms.

            3. Acting Layer (Consortium Blockchain)

            This layer manages trust, immutability, and incentives. The JSON Regimen's hash is recorded on the Hyperledger Fabric ledger (Chaincode: EcoCare). At the end of the TTR, the farmer submits feedback, which is processed by the Validation Engine (back in the Thinking Layer) to generate the RSS. This RSS is the critical input that the Smart Contract uses to execute the core logic of the Proof of Care model:

            If RSS > = Threshold: Transaction is validated, status set to VERIFIED_SUCCESS, and AgriTokens are rewarded.

            If RSS < Threshold: Transaction is validated, status set to VERIFIED_FAILURE, providing crucial negative reinforcement data for the AI model's retraining dataset.

            Table 2: Architectural layers of Eco-chain

            Layer	Component	Tech Stack	Function
            Sensing	Mobile Client	React Native, Whisper	Captures multimodal input; on-device ASR.
            Thinking	AI Engine	CLIP + LLaMA-3	Aligns V-L embeddings for diagnosis & JSON regimen.
            Acting	Consortium Blockchain	Hyperledger Fabric	Permissioned ledger; PoC consensus for incentives.
            Storage	Data Persistence	Amazon S3 I IPFS	Scalable redundant storage for high- volume media.

         2. Data Flow and Multimodal Fusion Pipeline

         The core data process follows a specific lifecycle: I. Initiation ( ):

         Client — S3: Image and Audio are uploaded to cloud storage.

         Client — API Gateway: A transaction request containing farmerID, timestamp, and the mediaHashes is sent. API Gateway — VLT Engine: Input is processed to generate the JSON Regimen.

         VLT Engine — Chaincode: The Regimen Hash and TTR are written to the ledger in a transaction signed by the Thinking Node. Chaincode State: RegimenRecord.status = PENDING.

         2.Feedback (TTR + ):

         Client — S3: New Feedback Image and Audio are uploaded.

         Client — API Gateway: Feedback request containing the recordID, feedbackHashes, and farmerID is submitted.

         API Gateway — Validation Engine: The engine fetches the initial and final media, calculates the RSS (as detailed in Section 4.3).

         Validation Engine — Chaincode: The RSS is passed to the Smart Contract function submitFeedback(recordID, rssScore). Chaincode State: RegimenRecord.status is updated to VERIFIED_SUCCESS or VERIFIED_FAILURE, and token reward logic is executed.

         1. The Regimen Success Score (RSS) Algorithm

            The RSS is a weighted, composite score designed to quantify the ecological success of the prescribed Regimen by comparing the initial and final state of the crop, weighted by the certainty of the farmer's feedback.

            The RSS is calculated as:

            where:

            4.3. Edge Operability: The SUQ-3 Model Compression Pipeline

            To enable the "Sensing Layer" to function on low-power mobile devices, the Agri-VLT must be optimized through the SUQ-3 framework (Structured, Unstructured Pruning, and Quantization).

            I.Structured Pruning: Removes entire blocks of redundant weights using an sparsity pattern. In Gated Recurrent Unit (GRU) models used for environmental time-series, this stage reduces parameter counts from 7,344 to 2,I60 weights without sacrificing reliability.

            1. Unstructured Pruning: Following the structured stage, individual low-magnitude weights are eliminated, refining the model further to approximately 26% of its original size.I3

            2. Quantization-Aware Training (QAT): Model weights are converted from 32-bit floating-point to 8-bit or 5-bit integers, reducing energy consumption to just I.I8 per inference-a 74.4%

            reduction.

         2. Remote Connectivity: Hybrid LoRaWAN and Satellite Architectures

            Agricultural ecosystems under threat from climate drift are often located in "no-signal" zones.I5

            LoRaWAN for Local Sensing: Provides a range of 5-I5 km with ultra-low power consumption, enabling multi-year battery life for in-field soil and leaf sensors.

            Satellite IoT Backhaul: In regions lacking cellular infrastructure, systems like the MF 400 IoT Satellite Bridge act as a gateway. This bridge provides network connectivity for up to I00 distributed LoRaWAN sensors via a single solar-powered satellite terminal, ensuring "Proof of Care" data is synchronized with the blockchain regardless of location.I7

         3. Blockchain Governance: CRPBFT and Slashing Mechanisms

         We employ a Consortium Blockchain to ensure scalability and controlled governance. To handle high-frequency agricultural data and energy constraints, the system utilizes the CRPBFT (Clustering and Reputation-based PBFT) consensus algorithm. CRPBFT achieves a 73% latency reduction and 92% energy savings compared to traditional Proof-of-Work mechanisms.

         To mitigate the risk of false reporting (e.g., uploading stock photos of healthy plants), the Eco-Chain incorporates an Evolutionary Game Theory model.3 A Slashing Mechanism is implemented where a farmer's reputation and token stake are burned if the Validation Engine detects high-entropy discrepancies or metadata mismatches in the feedback media. A "Nash Equilibrium" is reached when the cost of fraud exceeds the utility of the token reward, ensuring honest participation.

         Table 3: Ledger State Schema:

         Field Name	Type	Description
         recordID	string	Unique identifier (UUID).
         farmerID	string	Authenticated ID via MSP.
         regimenHash	string	SHA-256 hash of JSON output.
         reviewDeadline	uint64	Unix timestamp for recovery check.
         rssScore	uint8	Success score (0-100).
         status	enum	PENDING, VERIFIED_SUCCESS, VERIFIED_FAILURE.

         4.7 Smart Contract (Chaincode) Design and Ledger Schema

         The Smart Contract, implemented as a Chaincode in GoLang, governs the entire Proof of Care (PoC) lifecycle. Key Chaincode Functions:

         I.createRegimen(): Triggered by the Thinking Layer to lock a diagnosis.

         1. submitFeedback(): Validates the time-to-review () and updates the status.

         2. issueToken(): Executes the utility token reward upon successful recovery verification.

         1. CONCLUSION & FUTURE ENHANCEMENTS

            Eco-Chain offers a profound architectural advancement by closing the diagnostic loop in agricultural technology. By moving beyond isolated image classification to verified treatment outcomes recorded on an immutable ledger, we create a dynamic, self- correcting knowledge base that continuously adapts to new pathogens and climatic stresses. This is the ultimate expression of the end goal: to help plants help us in return.

            1. Advanced Modules: Yield Prediction and Resource Optimization

               Future iterations of Eco-Chain will expand into "AgriTransformers" that fuse multispectral satellite imagery and weather time- series data to predict crop yields with scores exceeding 0.9I. Hybrid CNN-LSTM architectures will also enable the prediction of harvest quality, ensuring that as plants evolve under climate stress, their nutritional value is maintained.I8 Variable Rate Technology (VRT) integration will further optimize the application of water and fertilizer, reducing resource waste by up to 35%.20

            2. Guidelines for Government: Climate Adaptation Playbooks

               Governments should leverage the real-time heatmaps generated by the consortium blockchain to coordinate national adaptation strategies.I0

               Adaptation Agents: AI-driven agents can model crop-specific risks and simulate the ROI of interventions, with some measures delivering up to I50% ROI for smallholders.22

               Data Commons: Policymakers must establish governance frameworks that ensure farmer data ownership while promoting open-data initiatives for researchers.

         2. LIMITATIONS & CHALLENGES

            1. Technical and Data Constraints

               Despite high accuracy, model performance remains vulnerable in "edge cases" such as mixed

               infections or rare pathogens. Data quality is also dependent on lighting conditions and optics, which can lead to misclassification. Furthermore, blockchain scalability remains a bottleneck; while permissioned frameworks can handle 3,000 transactions per second (TPS), a global swarm of

               requests could overwhelm storage targets.

            2. Ethical and Adoption Barriers

               Algorithmic bias presents a significant risk: AI systems optimized for monoculture may lack the nuanced understanding needed for intercropping. To mitigate this "Plant Blindness," we propose a Federated Continual Learning framework. Instead of a one- size-fits-all model, local "micro-models" are trained on the edge for specific regional crops, with only updated weights synced to the global ledger to preserve privacy and adapt to local niches.

         3. PROOF OF CONCEPT: PILOT IMPLEMENTATION

            1. Experimental Design

               A pilot study was conducted in Maharashtra, India from October to December 2025. The study focused on 50 smallholder tomato farmers using Android smartphones and LoRaWAN gateways. To ensure statistical validity, a Control Group was established using traditional manual extension services.

               Table 4: Pilot Study Performance Metrics

               Metric	Eco-Chain	Traditional Method
               Diagnostic Accuracy	94.20%	78.50%
               mAP (mean Average Precision)	0.92	0.74
               Dice Coefficient (Lesions)	0.88	NIA
               Avg. Time to Diagnosis	3.2 minutes	45 minutes (expert visit)
               Cost per Diagnosis	12 (~$0.15)	250 (~$3)
               Farmer Satisfaction	4.3I5	3.1I5
               Token Redemption Rate	68%	NIA

            2. Lessons Learned and Iterations

               Voice Preference: Voice-first inputs increased engagement by 40% among low-literacy participants.

               Media Quality: I5% of initial images required re-capture. As a result, an on-device image quality checker was implemented.

               Offline Capability: Connectivity outages necessitated an "Offline Mode" that queues diagnoses and syncs them once LoRaWAN or 4G becomes available.

         4. EXTENDED FUTURE RESEARCH DIRECTIONS

            1. Emerging Horizons: QML and Neuromorphic Sensing

               6G and Ultra-Latency: Future networks will enable real-time 3D leaf scanning for ultra-precise analysis.

               Quantum Machine Learning (QML): Hybrid quantum-classical models (using variational quantum circuits) demonstrate superior accuracy up to 95.2% in early disease recognition with lighter parameterization.

               Neuromorphic Computing: Energy-efficient AI chips inspired by biological neurons could reduce the sensing layer's carbon fotprint by an additional factor of ten.20

            2. Interdisciplinary Expansion

               Researchers plan to expand Eco-Chain beyond pathology to include Livestock Disease Monitoring and Aquaculture Health Management. Collaborative efforts with soil microbiologists will integrate "Soil Health Digital Twins" to model how subsurface ecosystems drift alongside the flora.3

         5. EXTENDED DISCUSSION & IMPLICATIONS

            1. Economic and Policy Impact

               Eco-Chain represents a shift from "reactive" to "proactive" governance. Cost-benefit analyses suggest that smallholders can save

               <8,500 per season through reduced crop loss. For governments, every <I invested in digital infrastructure for AI-blockchain systems yields <I2 in saved subsidy costs by preventing large-scale epidemics.2I

            2. Contribution to UN SDGs

               The framework directly supports SDG 2 (Zero Hunger) by securing yields, SDG I3 (Climate Action) through adaptation monitoring, and SDG I7 (Partnerships) by breaking data silos between research

               institutions and farmers.

         6. APPENDICES

            1. Sample JSON "Smart Regimen"

            2. Smart Contract Simplified Pseudocode

         7. REFERENCES

[I.2, 2.4, 3.I, 4.I, 5.I] (From original report)

30 Range shift patterns and pathogen evolution review (2025).

4 Environmental impact of LLMs and Green AI metrics (2025). SUQ-3 Compression framework for sustainable edge AI (2025). I7 Satellite-supported IoT for remote agriculture (2025).

Digital literacy and Training of Trainers models (2025). Evolutionary responses of Brassica rapa to drought (PNAS).

WORKS CITED

1. https://pmc.ncbi.nlm.nih.gov/articles/PMCI0I53038/#:-:text=Climate%20change%20further%2 0increases%20outbreak,plant%20diseases%20in%20new%20areas.

2. Potential impact of climate change on plant-pathogen interactions – ResearchGate, accessed on February 9, 2026, https://www.researchgate.net/publication/227I72257_Potential_impact_of_climate_change_on_ plant-pathogen_interactions

3. Rapid evolution of flowering time by an annual plant in response to a climate fluctuation, accessed on February 9, 2026, https://www.pnas.org/doi/I0.I073/pnas.0608379I04

[ ]The green AI arithmetic: Are farms saving more carbon than they spend" _ illuminem, accessed on February 9, 2026, https://illuminem.com/illuminemvoices/the-green-ai-arithmetic-are-farms- saving-more-carbon-than-they-spend
[ ]The Climate Advantage: How AI Can Help Governments Drive Better Climate Action, accessed on February 9, 2026, https://www.worldgovernmentssummit.org/observer/reports/detail/the- climate-advantage–how-ai-can-help-governments-drive-better-climate-action
[6]The Carbon Cost of Conversation, Sustainability in the Age of Language Models – arXiv, accessed on February 9, 2026, https://arxiv.org/pdf/2507.200I8
[?]accessed on February 9, 2026, https://illuminem.com/illuminemvoices/the-green-ai-arithmetic- are-farms-saving-more-carbon-than-they- spend#:-:text=By%20scheduling%20non%2Durgent%20tasks,carbon%20emissions%20by%20n early%20half.
[ ]Dose blockchain-based agri-food sup p ly chain guarantee the initial …, accessed on February 9, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMCI03I00I3/
[9] AI Driven Climate Adaptation Strategies for Agriculture — Scenario – Prism — Sustainability Directory, accessed on February 9, 2026, https:// prism.sustainability-directory.com/scenario/ai- driven-climate-adaptation-strategies-for-agriculture-2/
[1O] Research.pdf

1. Introducing the AgriTech User Experience (UX) Design Guidebook – GSMA, accessed on February 9, 2026, https://www.gsma.com/solutions-and- impact/connectivity-for-good/mobile- for-development/blog/introducing-the-agritech-user-experience-ux-design-guidebook/

2. Certificate in Digital and Climate-Smart Agriculture – CCSA – Cosmopolitan University Abuja, accessed on February 9, 2026, https://ccsa.cosmopolitan.edu.ng/trainings/certificate-programme- in-digital-and-climate-resilient-agriculture-dcra

3. Evolutionary context for understanding and manipulating plant responses to past, present and future atmospheric [CO2] – PMC – NIH, accessed on February 9, 2026, https:IIpmc.ncbi.nlm.nih.govIarticlesIPMC3248707I

4. SUQ-3: A Three Stage Coarse-to-Fine Compression Framework for …, accessed on February 9, 2026, https:IIwww.mdpi.comI2071-1050I17I12I5230

5. Remote IoT in NoSignal Areas: Satellite vs Cellular Connectivity Explained – Myriota, accessed

6. on February 9, 2026, https:IImyriota.comIremote-iot-satellite-vs-cellular-connectivityI 16. AI-Driven Climate Smart Agriculture: Enhancing Resilience and Productivity in a Changing

7. Climate – ResearchGate, accessed on February 9, 2026, https:IIwww.researchgate.netIpublicationI389561072_AI-

   Driven_Climate_Smart_Agriculture_Enhancing_Resilience_and_Productivity_in_a_Changing_C limate

8. Satellite-based IoT solutions for agriculture – Network Innovations, accessed on February 9, 2026, https:IIwww.networkinnovations.comIsolutionIresourcesIsatellite-supported-iot-solutions

9. Yield and Quality Prediction of Crops using a Deep Learning-Based Multimodal Data Integration Framework I European Journal of Electrical Engineering and Computer Science, accessed on February 9, 2026, https:IIwww.eu-opensci.orgIindex.phpIejeceIarticleIviewI19757

10. Digital Climate Smart Agriculture Mercycorps Agrifin, accessed on February 9, 2026, https:IImercycorpsagrifin.orgIdigital-climate-smart- agricultureI

11. AI-Driven Future Farming: Achieving Climate-Smart and Sustainable Agriculture – MDPI, accessed on February 9, 2026, https:IIwww.mdpi.comI2624-7402I7I3I89

12. Digital Literacy Training to Promote Diffusion of Digital Agricultural Tools to Smallholder Farmers – CGSpace, accessed on February 9, 2026, https:IIcgspace.cgiar.orgIserverIapiIcoreIbitstreamsIa95d7429-19f8-4a76-8152- 06c499fce4a0Icontent

13. 5 AI-Powered Adaptation Solution Examples – ClimateAi, accessed on February 9, 2026, https:IIclimate.aiIblogI5-ai-powered-adaptation- solution-examplesI

14. arXiv:2503.06973v1 [cs.CV] 10 Mar 2025, accessed on February 9, 2026, https:IIarxiv.orgIpdfI2503.06973

15. [2508.10817] Mobile-Friendly Deep Learning for Plant Disease Detection: A Lightweight CNN Benchmark Across 101 Classes of 33 Crops – arXiv, accessed on February 9, 2026, https:IIarxiv.orgIabsI2508.10817

16. Projects – IC3, accessed on February 9, 2026, https:IIwww.initc3.orgIprojects

17. LoRaWAN in Smart Agriculture A Comprehensive Guide, accessed on February 9, 2026, https:IIblog.contractize.appIlorawan-in-smart- agriculture-a-comprehensive-guideI

18. Rural Connectivity Solutions: Top 5 Agri Technologies – Farmonaut, accessed on February 9, 2026, https:IIfarmonaut.comIprecision-farmingIrural- connectivity-solutions-top-5-agri- technologies

19. PDF) Lightweight Deep Learning Models For Edge DevicesA Survey – ResearchGate, accessed on February 9, 2026, https:IIwww.researchgate.netIpublicationI388967279_Lightweight_Deep_Learning_Models_For

20. _Edge_Devices-A_Suvey

21. ML-EcoLyzer: Quantifying the Environmental Cost of Machine Learning Inference Across Frameworks and Hardware – arXiv, accessed on February 9, 2026, https:IIarxiv.orgIhtmlI2511.06694v1

22. Climate change impacts on plant pathogens, food security and paths …, accessed on February 9, 2026, https:IIpmc.ncbi.nlm.nih.govIarticlesIPMC10153038I