Sustainable and Efficient agricultural techniques: A case study on AI-Enhanced Greenhouse in India

Introduction

Smart Agriculture in the Era of AI and Digital Trust Agriculture has been the bone marrow of human civilisation, a meaningful structure for societies, economics, and ecosystems for decants. Issues such as climate change, soil degradation, also the food security have driven the need for sustainability ("State Food Agric. 2021," 2021). The importance of Agriculture in Human Civilisation is based on several factors, such as Food Security and nutrition, which ensures a stable food supply, which reduces the need for hunting and foraging. Economic Development-earlier, agricultural economics has raised industrial and technical success, creating diversification (Clark & Thorbecke, 1972) [5] . Social and cultural evolution-traditional practices have enhanced it, rituals and rural past life (Mazoyer & Roudart, 2006) [13] . Environmental impact-has contributed to deforestation and land degradation, but has also encouraged efforts towards sustainable agriculture, aiming to reduce the depleting consequences (Tilman et al., 2011) [26] . Smart farming, is also known as precision agriculture, including modern technology for essential productivity efficiency with sustainability in farming practices. The rapid increase in the global population, coupled with scarcity for natural resources, makes it imperative for countries to embrace application of smart farming with respect to practices for purpose of food security and the conservation of natural resources (Mathew, 2019) [12] . For more elaboration can be explain through this Table 1 andFig

Challenges in Controlled Environment Agriculture and Supply Chains

Many challenges are being faced, but the control measures prevent the damage by referring the integration of innovative engineering, plant science and IIT-field management systems to optimise p lant growth and development. Most relevant systems are: 1. Greenhouses-equipment with automated climate control (heating, cooling, shading, ventilation) and soilless cultivation systems, including hydroponics, cocopeat. 2. Vertical Farmsmulti-layer, full control on surroundings, involving hydroponic or aeroponic systemic body. Ideal for high-density production, but faces challenges. 3. Growth Chambers and Indoor Systems: used primarily for research and propagation but scalable for ultra-high-value products (Soto Arenas & Dressler, 2013) [23] . In India, 60 accuracy approx of the food quality is vanish in the supply chain from the farm to the end consumer. The consumer is actually paying around 35 acuuracy approx. more than what they could pay if the supply chain is optimized. This is due to wastage and multiple margins in the existing supply chain structure. The farmer in India gets 30 accuracy approx of what the consumer pays at the retail outlet. Compare this with the scenario in developed countries, where the farmer gets 70 accuracy approx. of the final retail price, and the wastage is only 4 to 6 accuracy approx. It is easy to understand the gains that could be achieved by doing what they do and leveraging their expertise in the supply chain in India." This paper focuses on improving the competitive advantage of Indian rural market through Supply Chain Management and also highlights some of the most important initiatives our government has taken so far (Bishnoi, 2009) [3] .

Scope and Contribution

In this chapter of the book we explores the use of smart [1] . More clear vision of this can be explained through this Figure 2.

Reinforcement Learning for Microclimate Control

• An agent is required in Reinforcement Learning (RL) to obtain the status of eco-system and must have the most suitable based on the environment's current status. When an agent has made its selection (or taken action), it receives an associated reward for its action taken, along with updated state of the eco-system. By the time, agent is attempting maximise its awards. The optimisation of this problem can be represented using the Markov Decision Process (MDP), a mathematical representation of how decisions are made sequentially (Sutton & Barto, 2018) [24] . • When an individual selects an action using the MDP, the MDP returns the agent a new and updated environment state and the agent's reward for the action taken. In the MDP, these values are also subject to probability, e.g. the frozen lake environment where agent is attempting move over a frozen lake that contains pores in it. Objective of the agent is to traverse the frozen lake in a maze-like manner, avoiding falling through any holes that are not present in the frozen lake. Since the ice of the lake may cause an agent to slide and be unable to achieve the exact direction of their intended goal, whenever an agent selects an action they might not arrive at their desired state. Therefore, there will always be some probability of the agent ending at an undesired position and not achieving their objective..(Sutton & Barto, 2018) [24] . (these are the transition probabilities. P:S×A→S). It stands for-It's observable states that S, and possible actions. A, a reward signal. R providing an immediate reward for choosing an action depending on the current state, transition probabilities. P that describes the probability distribution over states when taking an action when currently being in that specific state a discount factor. Γ that describes how to decrease the weight of future rewards (usually taken from [0,1]). (Morales & Zaragoza, 2011) [15] .

Greenhouse Microclimate Dynamics Environmental Parameters Affecting Crop Growth

By depending on the equipment fitted along with control means, greenhouse microclimate control options can be achieved at different levels. First control level consists of controlling (automatically) start-up and shut-down of the greenhouse microclimate mode with some form of emergency protection. Second control level can be defined by controlling by stabilizing all parameters of the greenhouse microclimate at the required levels by automatically controlling the operational performance of the various microclimate elements (Saba et al., 2021) [19] . The third control level is pertinent to the most efficient means of operating the greenhouse microclimate based upon the criteria of consuming the least possible energy for heating and electric power consumption (Shamshiri & Wan Ismail, 2014) [21] .

The dynamic equation of the Energy Balance have generated using all aerodynamic data to determine the air flow regime of the greenhouse. Due to the influence of gravity and wind, air is flowing through the openings in the walls of the greenhouse via filter action and through vent openings. The equations that define the microclimate conditions of the greenhouse contain boundary conditions that were established experimentally but do not fully align with the experimental data. (Shamshiri et al., 2016) [22] . Therefore, the Microclimate conditions can only be approximated by the use of this model because it doesn't take consideration to the change in the parameters of Microclimate by the volume of greenhouse. However, based upon the values from this model, tasks for the design of Microclimate conditions can be calculated and may be used to anticipate the interactive effect of the different Microclimate parameters on Quality Indicators. Dynamic conditions were created using the model, which determined the adequacy of the Model's ability to simulate dynamic conditions in the Greenhouse based on the analysis of the Conditions that were created using the Fisher statistic method. The Maximum Deviation is 11 per cent (Kadirov et al., 2023) [9] . Similarly explained in Table 3. To measure grape size (berry) and color (uniformity) automatically across blocks of grapes in the vineyard 2. To reduce reliance on manual sampling from the field 3. To provide data to make informed harvest and thinning decisions

Working Solution

High-resolution RGB (color) cameras were mounted on utility vehicles (e.g., ATV, truck) to take images of grape clusters during the veraision and pre-harvest periods. Images were taken using available daylight and had proper geotags from GPS.

A computer vision (CV) workflow was developed

1. Preprocess the images to correct for lighting discrepancies 2. Use convolutional neural networks (CNNs) to identify grape clusters 3. Use segmentation techniques to separate individual berries in an image 4. 4.Use the features of berry in order to quantify diameter and "intensity of color" as it would represent the concentration of anthocyanins.

The CV-based metrics/analyses were generated and represented within Geographic Information System (GIS) mapping throughout the vineyard block locations (Mohimont et al., 2022) [14] .

Results

Berry sizes estimated to be over 90% accurate versus manual calipers (within +/-2 mm). Maps of variability in berry colour showed how a range of berry colours existed between vines that were growing together but were not represented in normal sampling. The managers of the vineyard were able to identify areas of low uniformity that needed specialised management of the vine canopy

Impact

The system allowed for targeted harvesting and intervention, allowing for the improvement of grape uniformity and reduced cost of labour. Also, the CV method provided a means of scaling the system to continuously monitor grape quality throughout the growing cycle. To understand the varies CVs with their Aspects see table 5.

Key Takeaways

• CV systems provide quantitative data to replace visual judgements about grape quality • Spatial maps of grape quality provide the ability to utilise precision management at the sub-block level • The ability to identify variability at an early stage will result in efficiencies in the time required to make decisions. [18] . Traceability of products through use of blockchain technique provides a solution which increases transparency and accountability within complex agricultural supply chains by providing a secure, immutable along with decentralized record of all whole transactions that are completed, along with product data; within the entire length of agricultural supply chain from farm to fork (Hasan & Habib, 2022) [8] .

Objectives

This project focuses on constructing and developing an integrated blockchain traceability system supported with machine learning (ML) models to accomplish:

• To provide transparency in tracking produce throughout the supply chain (farm, transportation, cold storage, and retail). • To predict post-harvest losses through real-time data analytics. • To enable stakeholders (farmers, aggregators, cold chain providers and retailers) to make timely and proactive decisions to reduce waste and improve the efficiency of supply chains.

Traceability Infrastructure

A blockchain-based traceability platform was created by researchers and practitioners who utilized the core capabilities of blockchain (decentralized, immutable, and transparent) to log every transaction and condition change of agricultural produce.

A secure and ready-to-access shared ledger with stakeholders has been established, allowing for full end-to-end visibility (Lv et al., 2023) [11] .

The records related to the agricultural produce would include the following • The source of the farm, type of crop, and time of harvest.

• Cold storage conditions (e.g.: temperature and humidity).

• Transport route and time.

• Payments for the warehousing for entry and exit of produce.

Data will be collected through IoT sensors for all conditions and for those without some type of sensor through manual means, and will be uploaded using smart contracts to the blockchain (Zhang et al., 2023) [28] .

Machine Learning Model for Post-Harvest Losses

With real-time traceability data (e.g., cold storage conditions and transport time) documented in the blockchain, machine learning models will be applied to traceability data to detect patterns and indicators of the risk associated with post-harvest loss events.

For example

Outlier detection would be used to identify those cold storage or transport conditions where anomalies have occurred in the past that were historically associated with spoilage. Supervised ML models, like random forests and logistic regression, will be trained to forecast the probability of loss, using historical data associated with spoilage, extremes in temperature, high humidity, and duration of handling, as training inputs.

The machine learning algorithms will use historical blockchain records linked to real loss occurrences and allow for forecast indications of risk prior to spoilage (Zhang et al., 2023) [28] .

Analysis and Findings Increased Transparency & Trust

Blockchain provides an unchangeable record of the condition and movement of product throughout the entire supply chain reducing dispute and malicious alteration of data. With the ability for all parties in the supply chain to verify:(Reddy et al., 2025) [18] .

• History and Origin of the Goods Scores" along with alerts when specific conditions are present that indicate an increased likelihood of post-harvest loss due to: • Sustained High Temperatures During Transport • Humidity Levels Outside Desired/Optimal Range Pilot test results have demonstrated that the use of ML to support risk assessment have resulted in lower rates of spoilage due to proactive intervention by the parties, however the percentage of reduction varies by each unique situation. The level of visibility for all supply chain participants has also provided logistics providers with the information necessary to identify and address bottlenecks in order to improve processes. To understand more accurately see

Background

The environmental conditions of modern greenhouse agriculture such as temperature, humidity, CO₂, light, and ventilation etc done carefully controlled if the growers want to maximize the growth of the plants, use their resources efficiently, and maximize the quality of their crop. Fixed set-point controls and classical feedback control approaches are not designed to perform well with rapidly changing environments. Additionally, these techniques typically require a substantial amount of manual calibration and specialized knowledge from the growers. The inability to create a fixed control solution for the environment in the greenhouse becomes extremely obvious when working in a greenhouse that exhibits nonlinear dynamic behaviour and experiences unpredictable and/or unmeasured weather effects, variability in the plants growing inside of the greenhouse, and extremely complex interaction among the various control systems (Li et al., 2025) [10] .

To solve these problems, researchers and practitioners are beginning to utilize reinforcement learning (RL), which is a subfield of the artificial intelligence (AI) field of research. Reinforcement learning, an agent who learns to execute a particular control strategy by learning through trial and error as they interact with their environment. This is important for greenhouses because an agent can learn how to adaptively optimize multiple greenhouse performance goals, including crop performance, energy required to produce a crop, and total energy consumption through a single control strategy, and does not require a definitive mathematical model of the greenhouse system in order to be successful.(Li et al., 2025) [10] .

Objectives

The major objectives of utilizing reinforcement learning within greenhouse microclimate management include:

• Adaptability: To enable the automatic adjustment of environmental parameters according to the changing internal and external conditions. • Resource efficiency: To utilize energy and water resources more efficiently than static control methods through the development of an understanding of when and how to operate devices such as heating/cooling systems, fans, vents, and irrigation. • Optimizing Crop Growth: To maintain optimal conditions for plant physiology and yield objectives. • Human in the Loop: To incorporate grower experience and expertise into the learning process of improving the efficiency of the process (Platero-Horcajadas et al., 2024) [16] .

Methodology Systems Approach

An RL enabled greenhouse typically includes the following components:

• Data/Field Sensors & IoT devices which collect continuous real-time data of various temperature, humidity, CO₂, light, and other microclimate characteristics. • Actuators (heating, cooling, ventilation, shading and irrigation) that will be controlled based on feedback from the agent. • An RL Agent that interacts with either a simulated or realworld greenhouse environment in order to identify the optimal control strategies. • A RL Reward Function has been developed to encourage the agent to satisfy multiple objectives such as saving energy, stabilizing the microclimate, and promoting healthy plant growth indicators (Platero-Horcajadas et al., 2024) [16] .

Reinforcement Learning Framework

A greenhouse is one of the example of a Reinforcement Learning (RL) environment representation of MDP (Markov Decision Process). It states that environment are a set of microclimate measures (e.g., air temperature, moisture level, etc.) and could also include indicators of how well the plants are growing. The actions available to the RL agent would be changing the settings of the actuators (e.g., turning on or off heaters, etc.). The RL agent would receive rewards for maintaining the desired environment, for not using too much energy, and for not wasting other resources (e.g., fertilizer, water, etc.). Policies are learned and refined through continuous interactions with the agent (Cao et al., 2022) [4] . Another method of incorporating expert knowledge into reinforcement learning is through "interactive" reinforcement learning which uses the expertise of growers as a human-in-theloop to influence policies; to guide exploration; and to decrease training time. Comparative results from using interactive schemes versus strictly autonomous RL agents suggest that there would be greater performance improvement when combining less than perfect input from human experts than when using only an autonomous RL agent under conditions of uncertainty such as climate (Xiao et al., 2026) [27] .

Results & Insights Improved Environmental Stability

Greenhouses managed by real-time learning systems had microclimate variables such as Temperature, Humidity, and Moisture closer to optimal ranges than those controlled by traditional methods. The systems were more responsive to outside influences and unexpected variables (Platero-Horcajadas et al., 2024) [16] .

Improvement in Resource Use

Using real-time learning systems allowed growers to utilize control methods for actuators, such as heaters or fans, before the actuation was necessary. This proactive approach resulted in energy / water savings while not compromising plant growth (Platero-Horcajadas et al., 2024) [16] .

Combining Grower Knowledge and AI

The performance and reliability of Real-Time Learning systems that combined expert grower information were significantly better than the performance and reliability of the non-combined systems. This indicates that hybrid human and AI systems are of value in agricultural practices (Xiao et al., 2026) [27] .

Conclusion

The case study provides evidence that using reinforcement learning and a combination of IOT sensor networks with human input will provide a viable method of controlling microclimate conditions in greenhouses. Reinforcement learning-enhanced systems are capable of improving the adaptive capabilities of greenhouses, increasing resource efficiency and promoting scalable automation. This represents a major advance towards AI-Enhanced Greenhouses that can intelligently regulate their environment with a minimum amount of human interaction.

Future Directions and Research Opportunities

Advancements of artificial intelligence, Reinforcement Learning, and Blockchain Technology will create an irreversible transition point in current agricultural practices. The application of these three technologies within Controlled Environment Agriculture (CEA) and agri-food supply chains has already demonstrated a number of measurable results, but there is still a large opportunity to improve these systems through greater independence, scalability and sustainability. The remainder of this section identifies key future research directions that can further strengthen the role of AI-enabled greenhouses and intelligent supply chains to build on existing food systems towards more resilient food systems. In order to transition from isolated technological innovations to all-encompassing, intelligent agri-food ecosystems, future research on the integration of autonomous greenhouses and supply chains, privacy-preserving analytics and policycompliant AI deployment will be crucial.

References

1. Al-Amlih A, Shehadeh D, Al-Emari S, Sakha MA. IoTbased greenhouse environment monitoring and AI-driven crop optimization. Proceedings of the 1st International Conference on Computational Intelligence Approaches and Applications (ICCIAA); 2025. doi:10.1109/ICCIAA65327.2025.11013701
2. Bishnoi V. Supply chain management: a key to success in Indian rural markets. 2009.
3. Cao X, Yao Y, Li L, Zhang W, An Z, Zhang Z, et al. iGrow: a smart agriculture solution to autonomous greenhouse control. Proceedings of the AAAI Conference on Artificial Intelligence. 2022;36:11837-11845. doi:10.1609/aaai.v36i11.21440
4. Clark C, Thorbecke E. The role of agriculture in economic development. Econometrica. 1972;40(3):591-??. doi:10.2307/1913193
5. Cornelissen C, De Coninck S, Willekens A, Leroux S, Simoens P. In-field mapping of grape yield and quality with illumination-invariant deep learning. IEEE Internet of Things Journal. 2025. doi:10.1109/JIOT.2025.3617805
6. Godfray HCJ, Beddington JR, Crute IR, Haddad L, Lawrence D, Muir JF, et al. Food security: the challenge of feeding 9 billion people. Science. 2010;327(5967).
7. Hasan I, Habib M. Blockchain technology to ensure traceability of agriculture supply chain management. International Supply Chain Technology Journal. 2022;8(9). doi:10.20545/isctj.v08.i09.01
8. Kadirov Y, Samadov A, Goziev O. Equation of dynamics of greenhouse microclimate parameters. E3S Web of Conferences. 2023;390. doi:10.1051/e3sconf/202339004019
9. Li K, Shi J, Hu C, Xue W. The intelligentization process of agricultural greenhouse: a review of control strategies and modeling techniques. Agriculture (Switzerland). 2025;15(20). doi:10.3390/agriculture15202135
10. Lv G, Song C, Xu P, Qi Z, Song H, Liu Y. Blockchainbased traceability for agricultural products: a systematic literature review. Agriculture (Switzerland). 2023;13(9). doi:10.3390/agriculture13091757
11. Mathew J. Smart farming in agriculture sector. Data Driven Investor. 2019.
12. Mazoyer M, Roudart L. A history of world agriculture: from the Neolithic age to the current crisis. 2006.
13. Mohimont L, Alin F, Rondeau M, Gaveau N, Steffenel LA. Computer vision and deep learning for precision viticulture. Agronomy. 2022;12(10). doi:10.3390/agronomy12102463
14. Morales EF, Zaragoza JH. An introduction to reinforcement learning. In: Decision theory models for applications in artificial intelligence. 2011:63-80. doi:10.4018/978-160960-165-2.ch004
15. Platero-Horcajadas M, Pardo-Piña S, Cámara-Zapata JM, Brenes-Carranza JA, Ferrández-Pastor FJ. Enhancing greenhouse efficiency: integrating IoT and reinforcement learning for optimized climate control. Sensors. 2024;24(24). doi:10.3390/s24248109
16. Pothen Z, Nuske S. Automated assessment and mapping of grape quality through image-based color analysis. IFACPapersOnLine. 2016;49(16):72-78. doi:10.1016/j.ifacol.2016.10.014
17. Reddy BR, Vahoniya DR, Bagda B, Damor AB. Postharvest losses in mango supply chains in India: causes and preventive measures. Journal of Experimental Agriculture International. 2025;47(10):293-302.
18. Saba D, Sahli Y, Maouedj R, Hadidi A, Medjahed MB. Contribution to the realization of a smart and sustainable home. Studies in Computational Intelligence. 2021;912:261-290. doi:10.1007/978-3-030-51920-9_14
19. Seng KP, Ang LM, Schmidtke LM, Rogiers SY. Computer vision and machine learning for viticulture technology. IEEE Access. 2018;6:67494-67510. doi:10.1109/ACCESS.2018.2875862
20. Shamshiri AJ, Zakaria AJ, Wan Ismail WI, Man HC, Yamin M. Evaluation of the reduced state-variable TOMGRO model using boundary data. ASABE Annual International Meeting. 2016. doi:10.13031/aim.20162454205
21. Shamshiri R, Wan Ismail WI. A comprehensive comparison between wave propagation and heat distribution via analytical solutions and computer simulations. Research Journal of Applied Sciences, Engineering and Technology. 2014;7(1):111-122. doi:10.19026/rjaset.7.228
22. Soto Arenas MA, Dressler RL. A revision of the Mexican and Central American species of Vanilla with characterization of their ITS region. Lankesteriana. 2013. doi:10.15517/lank.v0i0.12065
23. Sutton RS, Barto AG. Reinforcement learning. 2nd ed. 2018.
24. FAO. The state of food and agriculture 2021. 2021. doi:10.4060/cb4476en
25. Tilman D, Balzer C, Hill J, Befort BL. Global food demand and sustainable intensification of agriculture. Proceedings of the National Academy of Sciences. 2011;108:2026020264.
26. Xiao M, Lan J, Yu J, Ma W, Xie Q, Sun C. Grower-in-theloop interactive reinforcement learning for greenhouse climate control. Computers and Electronics in Agriculture. 2026;242. doi:10.1016/j.compag.2025.111312
27. Zhang Y, Wu X, Ge H, Jiang Y, Sun Z, Ji X, et al. A blockchain-based traceability model for grain and oil food supply chain. Foods. 2023;12(17). doi:10.3390/foods12173235