Functional photonic and optoelectronic materials and devices for climate-resilient smart agriculture: a review

1 Introduction

Global population growth and climate change pose significant challenges to agricultural production and sustainability. By 2050, the global population is projected to reach approximately 9.6 billion (United Nations Department of Economic and Social Affairs and Population Division, 2024). Earlier FAO projections estimated that global agricultural production would need to increase by approximately 60% relative to the 2005–2007 baseline to meet projected demand (Food and Agriculture Organization of the United Nations FAO, 2009). In parallel, agriculture must adapt to climate change while improving the efficiency and sustainability of natural-resource use (Vishnoi and Goel, 2024; Getahun et al., 2024). These challenges highlight the need for innovative technologies that can improve agricultural productivity, resource-use efficiency, and climate resilience while minimizing environmental impacts. Nevertheless, traditional agricultural methods have difficulty adapting to alterations in the climatic conditions in terms of extreme weather events, unpredictable rainfalls, heat stress, and water scarcity. The challenges require new technologies, which would enhance the productivity of agriculture whilst at the same time protecting the environment. Photonics and optoelectronics technology have potential solutions to the current farming industries by allowing crop and soil to be accurately, non-invasively, and real-time monitored (Falcon et al., 2022; Vishnoi and Goel, 2024; Getahun et al., 2024). Photonics is the production, control and measurement of light and optoelectronics is the devices that can transform electrical signals to optical signals and the other way around. These technologies are based on the engagement of the electromagnetic radiations and biological substances to attain precise data concerning the plant state, health, nutrient levels, water strains, as well as environmental protection (Butt et al., 2024; Cavaco et al., 2022). Optical sensing systems offer real-time and spatially detailed data of the state of crops as opposed to conventional monitoring systems using manual observation or destructive sampling. Hyperspectral cameras, fibre optic cameras, thermal imaging systems, among other sensors, can record minor shifts in the physiology of plants that cannot be detected by a human eye. These technologies alongside the use of precision agriculture systems help to facilitate site-specific cultural management methods that maximize the use of inputs and minimize effects on the environment. The recent developments of artificial intelligence (AI), miniaturization, wireless communication, and data analytics have increased the rate at which photonic technologies are being adopted in the agricultural industry (Miller et al., 2025; Bouarroudj et al., 2025). The current smart farming platforms include both optical sensors, IoT networks, and machine learning algorithms to allow automatic surveillance and make predictions. The developments allow more farmers of various sizes to have access to advanced technologies. Several earlier reviews have catalogued individual optical sensing modalities in agriculture or have treated photonics and precision agriculture as parallel but separate subjects (Ram et al., 2024; Brewer et al., 2022; Pineda et al., 2021; Haque et al., 2024). The present review differs in three respects. First, it follows a mechanism to outcome pathway rather than a technology-by-technology inventory: each sensing principle is traced from its underlying physics, through the materials and photonic/optoelectronic components that realise it, to the specific climate-resilience or resource-efficiency outcome it enables. Second, it explicitly benchmarks each technique against technology readiness level (TRL), cost, and field reliability, so that the review can be read as a deployment roadmap rather than a survey of laboratory demonstrations. Third, it links each optical modality directly to a named climate stressor (drought, heat stress, flooding/waterlogging, and shifting pest and disease pressure), which prior reviews of photonic agricultural sensors have not done systematically. Together these three lenses mechanism, maturity, and climate relevance are the review’s distinguishing contribution. The review gives an overview of photonics and optoelectronics technologies applicable to smart farming systems and the contribution of these technologies to the enhancement of climate resilience, resource efficiency, and sustainable crop production.

The role of photonic sensing in climate resilience needs, however, to be understood within the confines of the sensing environment. Climatic stresses do not act in isolation, and drought can be accompanied by heat stress, high temperature affecting disease development, and rain affecting the nutrient balance and soil aeration at the same time. Therefore, the same spectral or thermal reaction can be produced by several factors acting together (Haque et al., 2024; Cho et al., 2024). This becomes a very significant limitation when it comes to single-sensor-based diagnostics. For instance, the increase in canopy temperature can be caused by a water deficit, high evaporative demand, or changes in the plant structure, while the spectral reaction could be caused by lack of nutrients, infections, or normal phenology (Haque et al., 2024; Fei et al., 2023). Climate-resilient sensing systems should incorporate contextual information on weather, soil, and crop conditions and increasingly employ multimodal data fusion to distinguish among multiple interacting stress responses. The significance of photonics technology should, therefore, be measured not only by its capacity to detect stress but also by its capacity to distinguish different stressors (Guer et al., 2024; Fei et al., 2023).

2 Review methodology

This review used a structured literature search and critical narrative synthesis to identify and evaluate recent advances and synthesis to discover, categorize and critically evaluate current advances in photonics and optoelectronics in climate resilient smart agriculture. The literature search was centered mainly on peer-reviewed publications that examined optical sensing techniques, photonic and optoelectronic devices, multispectral and hyperspectral imagery, thermal imaging, chlorophyll fluorescence, fibre optic sensors, LiDAR, laser technologies, Internet of Things (IoT), Artificial Intelligence (AI), machine learning (ML), sensor fusion, and digital decision support systems for agriculture. Recently published literature from 2021 to 2026 was mainly taken into consideration to cover current advances in technology; some previous but seminal literature was selected whenever it was needed to understand the principles and technologies involved. The literature was classified based on four dimensions of analysis: (i) sensing principle and photonic/optoelectronic technology used, (ii) agricultural information or stressors sensed, (iii) deployment and operation conditions, and (iv) impact in climate resilience and resource efficiency (Page et al., 2021a; Page et al., 2021b). Studies were identified as relevant especially when they showcased sensing at field scale, measurement of crops or soil in quantifiable terms, integration into decision support system, or deployment of sensors in practice. Studies focused purely on non-farming applications were ruled out from this review unless they could provide an applicable sensing principle for agricultural use. The review did not consider any technology as an isolated sensing technique but synthesized all evidence from mechanisms to outcome. This comprised the process through which the interaction between light and plant/soil constituents creates measurable signals, which then get processed to yield agronomic indicators useful for decision-making for climate-adapted agriculture (Ram et al., 2024; Cheng et al., 2025). In addition to that, the balance between spectral richness, spatial and temporal resolution, cost, computational needs, environment responsiveness, scalability, calibration needs and technology maturity was considered especially. The review also looked into the increasing intersection of optical sensors with IoT communication, edge computing and artificial intelligence/machine learning. Where feasible, technologies were compared based on their sensing capacity, practicality at field level, technological maturity and their possible role in managing water, nutrients and stress on crops.

3 Fundamental principles of photonics and optoelectronic sensing

Photonics finds use in agriculture on the principle of interaction of electromagnetic radiation and plant tissues. The absorption, reflections, and transmission of light are not equal between plants of different wavelengths based on their biochemical makeup, structural properties, and physiological state (Butt et al., 2024; Cavaco et al., 2022).

3.1 Spectral signatures and vegetation indices

There are typical spectral patterns in healthy vegetation. The chlorophyll pigments have a high level of absorption in the blue and red matter of the visible spectrum and the leaf cellular structure causes the plant tissues to reflect a lot of the near-infrared (NIR) wavelengths. Such spectral properties are the foundation of vegetation indices like the Normalized Difference Vegetation Index (NDVI) is widely used to characterize vegetation greenness, canopy vigour and biomass-related variation, and may show correlations with chlorophyll and nitrogen status under appropriate crop and environmental conditions (Silva et al., 2023; Shah et al., 2021). Red-edge vegetation indices such as the Normalized Difference Red Edge (NDRE) use reflectance from a red-edge band and a near-infrared band and can provide greater sensitivity to chlorophyll and canopy nitrogen-related variation than NDVI in some crop conditions (Zhou et al., 2021).

3.2 Hyperspectral and multispectral sensing

Multispectral sensors are sensors that record reflectance over a few spectral bands, usually three to ten wavelengths. Instead, hyperspectral sensors record data in hundreds or thousands of small bands of spectral data. Hyperspectral imaging is finding a use as a valuable instrument of crop phenotyping, nutrient analysis and disease detection (Ram et al., 2024; Cheng et al., 2025; Guer et al., 2024) and has offered non-destructive and very detailed data about plant physiology (Xia et al., 2024).

3.3 Detection of water stress using thermal imaging

The development of thermal imaging can detect any changes in plant temperature due to changes in transpiration rate. In case of water stress in plants, the stomata close and this minimizes transpiration which causes an increase in the temperature of the leaves (Brewer et al., 2022; Cho et al., 2024). These changes in temperature can be captured with a thermal camera and can be used to identify water stress early and optimize irrigation by identifying when it occurs.

3.4 Chlorophyll fluorescence

Chlorophyll-fluorescence methods give direct data on photosynthetic efficiency. When light energy is taken by the plants a small part of it is re-emitted back as fluorescence. Fluorescence intensity change is indicative of a perturbation of photosynthetic mechanisms by stress factors like nutrient deficiency, infection by pathogens, or stresses like environmental stress (Janeeshma et al., 2022; Moustaka and Moustakas, 2023).

3.5 Fiber optic sensing

Fibre-optic sensors can measure physical and environmental variables such as strain, temperature and humidity using changes in optical intensity, phase, spectral characteristics or wavelength. In particular, fibre Bragg grating (FBG) sensors detect strain or temperature through shifts in the Bragg wavelengt (Li et al., 2024).

Although Table 1 summarizes the sensing principles, capabilities, advantages and limitations of major photonic technologies, their practical suitability varies considerably with the intended application, environmental conditions, data-processing requirements and deployment scale. Therefore, a critical comparison of their operational strengths, limitations and deployment suitability is presented in Table 2 to highlight the trade-offs that influence technology selection for climate-resilient agriculture.

TABLE 1. Comparative evaluation of major photonic/optoelectronic sensing techniques in agriculture.

TABLE 2. Critical evaluation and deployment suitability of photonic sensing technologies for climate-resilient agriculture.

4 Critical comparison of sensing modalities

There is no single photonic sensing technology that is ideal for every agricultural sensing requirement. In selecting a sensing method, there is always a trade-off between spectral information, spatial resolution, temporal frequency, sensing range, cost, computational load and robustness in the field (Getahun et al., 2024; Ram et al., 2024; Cheng et al., 2025; Miao et al., 2024). While the multispectral sensing methods are relatively cost-effective and relatively mature in their computation, they have fewer spectral bands which limit their capability to detect subtle biochemical changes. Hyperspectral sensing offers more spectral information and hence, it can detect more subtle biochemical and physiological changes that cannot be detected by the multispectral method, but the former is more expensive, has larger amounts of data and hence, requires high computation (Ram et al., 2024; Cheng et al., 2025; Xia et al., 2024). Thermal sensing can offer a better estimation of canopy temperature and plant water state than multispectral and hyperspectral methods, but its measurements are very dependent on atmospheric conditions, sunlight, canopy geometry and solar incidence angle (Pineda et al., 2021). Chlorophyll fluorescence sensing can provide very sensitive information on any disturbances in photosynthesis but the relatively low signal strength, the short effective sensing distance and low throughput limit its application (Brewer et al., 2022; Pineda et al., 2021; Haque et al., 2024; Zhou et al., 2021). Also, fibre Bragg gratings and other optical fibre sensors are characterized by high sensitivity, non-stop data collection on the plant level, and electromagnetic immunity, yet their point sensing requires the installation of a dense array of sensors (Lo Presti et al., 2021; Lo Presti et al., 2022; Venketeswaran et al., 2022). LiDAR can provide three-dimensional data on canopy structure, heights and biomass and work in diverse lighting conditions, but the cost of equipment, processing data, and its integration are still serious limitations (Farhan et al., 2024; Debnath et al., 2023). On the other hand, RGB computer vision is cheaper and more technologically advanced and is thus applicable for plant counting and phenological measurements and weed detection; however, it does not possess direct biochemical specificity (Fuentes-Peñailillo et al., 2023; Ninomiya, 2022; Yaseen and Long, 2024). It follows that the choice of technologies should be based on specific applications, as opposed to accuracy alone: low-cost RGB or multispectral imaging techniques will do well for regular farm-level monitoring, while hyperspectral, fluorescent or LiDAR imaging should be used where more physiological or structural information justifies higher cost and complexity of processing (Ram et al., 2024; Janeeshma et al., 2022; Farhan et al., 2024; Guer et al., 2024). The major photonic technologies and their climate-resilience outcomes are summarized in Table 3.

TABLE 3. Mapping of photonic/optoelectronic technologies to climate-resilience outcomes.

5 Significant photonics and optoelectronics techniques in agriculture

5.1 Spectroscopy systems and optical sensors

5.1.1 Optical sensors

Photonics based precision agriculture involves optical sensors. They allow the monitoring of the environmental conditions and plant health in a non-destructive manner (Cheng et al., 2025; Miao et al., 2024).

5.1.2 Multispectral sensing

Multispectral sensors are used to detect reflectance of the product at specific wavelengths to determine chlorophyll concentration, nitrogen, and vitality (Ram et al., 2024; Cheng et al., 2025). Commercial systems like crop sensors operate on spectral bands within the red and near infrared spectrum to compute vegetation indices on a real time basis.

5.1.3 Hyperspectral imaging

Hyperspectral imaging systems give specific biochemical details because hundreds of small spectral bands are studied. They are common in crop phenotyping, disease and nutrient determination.

5.1.4 Laser-induced breakdown spectroscopy

Another sophisticated optical method is laser-induced breakdown spectroscopy (LIBS) that is applied to analyse the nutrients of soil in situ in a very rapid method (Andrews et al., 2024). A powerful burst of laser lights generates a plasma on the soil surface, and they display the spectrum of emission that indicates the elemental structure.

5.1.5 Thermal imaging

Thermal cameras and infrared thermometers measure canopy or leaf surface temperature, which can be used as an indicator of transpiration and plant water status when interpreted together with environmental conditions (Karmakar et al., 2024).

5.2 Imaging systems

The latest imaging systems would be used to integrate the spatial, spectral and temporal data to investigate the crop in a comprehensive manner. The RGB cameras are used to capture typical colour images and are commonly utilized in computer vision systems like counting plants, examining the growth stage and the detection of weeds (Yaseen and Long, 2024; Debnath et al., 2023). Miniature images on satellites, drones, or on-ground vehicles can be used to monitor crops of large areas using multispectral and hyperspectral imaging platforms (Venketeswaran et al., 2022). Thermal imaging Systems determine field crop health and variability through the detection of temperature differences across the canopies of crops. They have been effectively utilized in the orchards and vineyards to detect water stress and disease infection. Chlorophyll fluorescence imaging gives information on photosynthetic activity and plant stress.

5.3 Fiber optic sensors

Fiber optic sensors are a novel method of monitoring the plant. The sensors are lightweight, flexible, and resistant to electromagnetic interference, and thus can be used in the long term in the field due to their ability to be attached to stems or leaves to trace the growth of plants, temperature, humidity, and mechanical strain (Li et al., 2024). These sensors have very precise measurements of plant physiological activities and microclimatic fluctuations Emerging studies are also exploring the application of fibre optic sensors to measure chemical and biological activity in plant tissues and nutrient transport and sap flow presented in Figure 1 (Lo Presti et al., 2021; Lo Presti et al., 2022; Tosi, 2018).

5.4 LED and laser-based systems

LED and laser technologies have applications in both controlled-environment crop production and crop-management systems (Li et al., 2025). LED systems are widely used in greenhouses and vertical farms to control light intensity, spectral composition and photoperiod. Laser-based weed-control systems are fundamentally different from UV-C pathogen-control approaches. In laser weeding, a focused laser beam is directed at the weed and delivers sufficient energy to damage plant tissues, thereby suppressing or killing the target weed without relying on chemical herbicides (Karmakar et al., 2024; Yaseen and Long, 2024).

5.5 LiDAR and time-of-flight sensors

LiDAR (Light Detection and Ranging) is a type of measurement used to determine the distance of objects by a calculation of time (Farhan et al., 2024; Debnath et al., 2023) that the laser pulses require to reflect. The sensors produce three-dimensional images of crop structures. LiDAR technology enhances the measurement of crop height, canopy structure, leaf area index and biomass accurately. LiDAR sensors can be quickly used to map crop fields mounted on drones or ground vehicles, providing the same structural information, whereas Time-of-Flight sensors can be used to provide this kind of data in smaller and less expensive instruments, and thus are applicable in IoT-based smart farming systems (Farhan et al., 2024; Miao et al., 2024).

5.6 Enabling materials

The performance of the sensing modalities described above is set largely by the photonic materials used at the detector and light-delivery stage. Silicon CMOS and CCD arrays remain the workhorse detectors for RGB and visible-band multispectral cameras because of their low cost and manufacturing maturity (Cheng et al., 2025; Silva et al., 2023). Fibre Bragg gratings (FBGs) represent an important class of fibre-optic sensing technology investigated for plant growth and microclimate monitoring. In FBG sensors, changes in strain or temperature produce shifts in the Bragg wavelength that can be measured optically (Lo Presti et al., 2021; Lo Presti et al., 2022). Plasmonic materials, including nanostructured noble-metal surfaces, are being investigated for surface-plasmon-resonance sensing and other label-free optical biosensing applications. In agriculture, emerging studies have explored their potential for detecting plant pathogens and other target analytes, although broader applications such as pesticide-residue and mycotoxin detection require analyte-specific validation (Ren et al., 2026). Metamaterials and metasurfaces can provide engineered control of electromagnetic-wave propagation, spectral filtering and light–matter interactions. Their potential for miniaturized agricultural optical sensors is an emerging research direction, but practical field deployment remains to be demonstrated more extensively (Butt et al., 2024). Finally, conventional silica and polymer optical fibres provide the low-loss transmission backbone linking distributed field sensors to central data-logging or IoT gateway units. Continued cost reduction in agricultural photonics is therefore closely tied to progress in wafer-scale In GaAs fabrication, low-cost FBG inscription, and printable plasmonic/meta surface manufacturing, rather than to sensing concepts alone (Butt et al., 2024; Venketeswaran et al., 2022; Ren et al., 2026).

6 Applications for crop and soil monitoring

6.1 Crop stress and health detection

Photonics technologies can be used to detect plant stress at the earliest possible stage before visible symptoms develop. Hyperspectral sensors detect the minor spectral variations due to physiological disturbances, whereas thermal imaging identifies temperature variations that you will find due to water stress (Brewer et al., 2022; Cheng et al., 2025). These technologies can monitor the status of crops in real-time and provide the farmer to implement corrective measures before losses in yields presented in Figure 2 (Zhou et al., 2021; Karmakar et al., 2024).

6.2 Nutrient management

Nutrient sensors are used to optimize the fertilizer application by detecting the nutrient status of crops (Shah et al., 2021; Mitra et al., 2023). The indexes of vegetation, like NDVI and NDRE, give a clue on the nitrogen concentration, making it possible to administer fertilizers on the site (Zhou et al., 2021). Accurate fertilization technologies will reduce fertilizer application by 15–30 percent without loss of crop losses. LIBS can provide rapid in-situ elemental analysis of soil and plant material. However, matrix effects, calibration requirements and the influence of moisture and sample composition can affect quantitative accuracy; therefore, laboratory-based reference measurements remain important for calibration and validation (Andrews et al., 2024).

6.3 Water management

Sustainable agriculture requires proper water management. IoT sensors and optical sensors can be used to schedule irrigation automatically according to the real-time needs of crops in terms of water (Brewer et al., 2022). Thermal imaging systems monitor water stress behaviour at a tender stage, hence accurate irrigation choices. Research has revealed that sensor-based irrigation systems have the potential of saving of water 12–15 percent and enhancing water-use efficiency (Miller et al., 2025; Bouarroudj et al., 2025).

6.4 Yield prediction and crop phenotyping

Multispectral, hyperspectral, LiDAR and RGB-based phenotyping systems can provide structural and spectral traits relevant to yield prediction. Reported predictive performance varies substantially among crops, datasets, environments and modelling approaches; therefore, performance should be reported using appropriate metrics such as R², RMSE or MAE rather than a generic percentage accuracy (Farhan et al., 2024; Fuentes-Peñailillo et al., 2023; Debnath et al., 2023; Xu and Li, 2022).

6.5 Pest and disease management

It is important to ensure that the pests and diseases are detected at an early stage in order to reduce losses in crop. Hyperspectral imaging has shown potential for detecting spectral changes associated with some pathogen infections before or around the onset of visible symptoms, although detection performance depends on pathogen, host genotype, disease stage, environmental conditions and sensor configuration (Reis Pereira et al., 2023; Guer et al., 2024). Computer vision cameras based on RGB can automatically track insect traps and determine pest populations. Laser-based weed-control systems can selectively damage weed plants and thereby reduce reliance on herbicide applications in suitable cropping systems. Their practical effectiveness depends on weed detection accuracy, plant growth stage, crop–weed separation, operating speed and energy requirements (Yaseen and Long, 2024; Vijayakumar et al., 2025).

7 Data integration: IoT, AI/ML and smart farming system

Photonics technologies achieve greater effectiveness when integrated with IoT platforms and smart farming systems. IoT networks collect real-time information from optical and environmental sensors and transmit it for processing, while artificial intelligence (AI) and machine learning (ML) algorithms convert these data into predictive models and management recommendations (Miller et al., 2025; Bouarroudj et al., 2025). The integration of photonic and optoelectronic technologies with AI, IoT, and autonomous decision-support systems enables a connected framework for climate-resilient agriculture, as illustrated in Figure 3. Smart farming systems combine sensor observations, weather information and agronomic models to support irrigation, fertilization, pest management and harvesting decisions (Miao et al., 2024; Muhammed et al., 2024). Autonomous robotic platforms equipped with optical sensors further enable crop monitoring, weed detection and precision harvesting. Multimodal sensor fusion involves combining measurements from various sensing systems to get a more holistic perspective on crop, soil and environmental conditions. For instance, multispectral imagery is used for assessing canopy vigour and pigment state; thermal imagery is for detecting canopy temperature variations and water stress; LiDAR is for assessing canopy structure; and soil moisture and weather sensors give contextual information (Karmakar et al., 2024; Fei et al., 2023). Fusion can be performed at data, feature or decision level. At data level, fusion involves aligning measurements, at feature level, fusion involves integrating extracted features like vegetation indices, canopy temperature and plant height, and at decision level, it involves merging of independent model output using statistical and machine learning techniques. Multimodal fusion has the potential to improve discrimination among drought, nutrient deficiency and disease-related responses that may appear similar in individual sensing modalities (Fei et al., 2023; Allu and Mesapam, 2025). In addition, LiDAR based structural information fused with hyperspectral measurements will help discriminate canopy structural responses from biochemical response. Nevertheless, multimodal sensor fusion poses challenges related to synchronization of sensors, registration of data, calibration, resolution differences, missing values and computation requirements. Thus, future smart farming systems should focus on data architecture and fusion methods and not just increase sensor numbers.

7.1 AI/ML technical depth

Optical sensors generate data whose dimensionality and volume make automated analysis essential. Hyperspectral cubes are first reduced using principal component analysis (PCA), band-selection algorithms, or partial least squares (PLS), since adjacent spectral bands are strongly correlated and only a small subset carries agronomically relevant variance (Ram et al., 2024; Cheng et al., 2025; Guer et al., 2024). Classical machine-learning classifiers/regressors - partial least squares discriminant analysis (PLS-DA), support vector machines (SVM), random forest, and gradient-boosted trees (e.g., XGBoost) - remain widely used for nutrient, stress, and disease classification from vegetation indices or reduced spectral features and are preferred where labelled training data are limited (Guer et al., 2024; U.S. Department of Energy, 2026). Deep learning architectures are increasingly applied directly to raw or minimally processed data: one-dimensional convolutional neural networks (1D-CNNs) operate on individual spectral signatures for disease/nutrient classification, while two- and three-dimensional CNNs jointly exploit the spatial and spectral dimensions of hyperspectral image cubes for tasks such as pixel-wise disease mapping (Guer et al., 2024). Recurrent and other temporal modelling approaches can be used to analyse repeated observations collected across the growing season, although their suitability depends on the availability and quality of sufficiently long time-series datasets (Cho et al., 2024). More recently, attention-based and transformer architectures, including vision transformers adapted for hyperspectral classification, have shown improved performance where long-range spectral-spatial dependencies matter, though at higher computational cost (Ram et al., 2024; Cheng et al., 2025; Guer et al., 2024). Because labelled agricultural datasets are typically small relative to those used in computer vision generally, transfer learning pretraining on large general-purpose or remote-sensing datasets and fine-tuning on crop-specific data is now a common strategy for reducing the data burden. Finally, model deployment increasingly targets edge devices (drones, IoT nodes) rather than cloud servers, which motivates model compression, quantization, and lightweight (‘TinyML’) architectures capable of running inference locally under tight power and latency budgets, a point returned to in Section 9.2 on edge AI (Miller et al., 2025; Muhammed et al., 2024). Higher model complexity does not always mean higher agricultural utility for AI/ML algorithms. While deep learning and transformers can work well on some datasets with regard to their ability to predict something, their generalization for different varieties, soil types, climate conditions, lighting conditions and crop growth stages is still an open question. This implies that models based on the data from one geographical area or season can show a considerable drop in performance when used in another environment (Fei et al., 2023). In addition, hyperspectral high-dimensional datasets are often redundant, and in case of a limited number of observations, they can lead to the model’s overfitting (Guer et al., 2024). Therefore, in addition to the classification accuracy or R2 metric, model evaluation must include an analysis of its performance on independent datasets, its applicability to other locations or seasons, uncertainty analysis, and comparison with baseline models (Ram et al., 2024; Upadhyay et al., 2025; Guer et al., 2024; Cho et al., 2024). In addition, interpretability is crucial for agriculture-related decision-making, since farmers and agronomists must understand why the model recommends watering, fertilizing or protecting crops from some threats (Cho et al., 2024; Muhammed et al., 2024).

8 Deployment readiness: challenges, TRL, and economics feasibility

Photonics technologies have a lot of potential, but they have several challenges that inhibit their overall adoption. The high initial prices of powerful sensors like hyperspectral cameras and LiDAR are a significant obstacle to small scale farmers. There is also a problem of data integration and interoperability when multiple sensing technology can be used in the same system. Furthermore, interpreting spectral data is a specialized task that cannot be easily found in rural farming communities. Sensors can also be influenced by environmental variability like cloud cover and the state of the lighting (Miller et al., 2025; Bouarroudj et al., 2025; Fei et al., 2023). The small-scale availability of IoT connectivity within rural environments and lack of infrastructure to validate these technologies and in areas where they are required further limits the large-scale adoption of these technologies (Vishnoi and Goel, 2024; Miao et al., 2024). Technology readiness level (TRL, on the standard 1–9 scale from basic principles observed to full commercial deployment) is a useful common yardstick for comparing how close each sensing modality is to routine farm use (Upadhyay et al., 2025; Guer et al., 2024; U.S. Department of Energy, 2026). Technology readiness levels and deployment barriers are summarized in Table 4. RGB imaging, GPS-referenced multispectral crop sensors (e.g., commercial active optical sensors), and basic LiDAR-derived canopy metrics are at TRL 8–9, with established commercial products and vendor support networks. UAV-mounted multispectral and thermal systems sit at TRL 7–8, commercially available but still requiring specialist operation. Hyperspectral imaging, LIBS soil analysis, and laser-based weeding are at an intermediate TRL 5–7: early commercial units exist (including start-ups offering laser weeding as a service), but cost, throughput, and calibration requirements still limit adoption to well-resourced operations. Fiber optic/FBG plant-wearable sensors and quantum-enhanced sensing remain largely at TRL 3–5, confined to research trials and pilot deployments (Carbon Robotics, 2026; Harvested Robotics, 2026; Kyron Robotics, 2026). Beyond TRL, three practical factors govern real-world adoption. First, reliability and maintenance: optical sensors calibrated in controlled conditions often drift under field dust, humidity, and temperature extremes, and require either frequent recalibration or robust environmental packaging, both of which add cost and technical burden that smallholder operations may not be equipped to absorb. Second, scalability: many demonstrated systems are validated on small research plots, and the interoperability problem noted in Section 6 (data integration across sensor brands and formats) becomes more acute at commercial farm scale. LiDAR systems means payback periods can extend beyond a single growing season for smaller operations, favouring sensor-as-a-service, cooperative equipment-sharing, and subsidized pilot programs as transitional adoption pathways rather than direct capital purchase by individual smallholders (Colaço and Bramley, 2018; González-Briones et al., 2019).

TABLE 4. Indicative TRL range for representative agricultural applications.

TRL values are indicative estimates for representative agricultural implementations rather than intrinsic maturity levels of the sensing modality as a whole. Actual TRL depends on the specific hardware, application, validation environment and deployment context.

8.1 Research gaps and unresolved challenges

Despite remarkable progress in the field of photonic and optoelectronic sensing technologies, there are some research questions, which make it impossible to transfer the results of laboratory research straight to practical use in agriculture. The first issue is that many experiments are made under controlled environmental conditions or on small experimental plots, while actual agricultural lands are highly variable in terms of illumination conditions, canopy structure, soil background, atmospheric conditions and management practices. The second challenge is that the models, based on seasonal, cultivar- or geographical-specific data, may lack transferability to other agroecosystems. The third question related to the need to calibrate and standardize the sensors to ensure comparability between measurements made with different instruments and platforms. Moreover, multimodal systems produce heterogeneous data, which requires synchronization, registration and fusion. The fourth issue is the fact that high-performing artificial intelligence models may require extensive labelled datasets, which are costly to obtain in agriculture. The final challenge is the fact that technological efficiency does not automatically imply economic viability. The costs of sensors, UAV platforms, connectivity, maintenance, data processing and expert skills should be considered alongside the economic utility of the decision, supported by the technology.

9 Future trends

The future trends in photonics-based agriculture revolve around accuracy of sensors, cost reduction, and automation. New technologies include bio-photonic feedbacks, wearable plant sensors, superior spectroscopy, and deep learning data analysis. Precision agriculture technologies will become available to farmers as miniaturization of optical devices and mass production of sensors will make them more affordable. The combination of satellite, drone, and ground-based sensing platforms is bound to allow the multi-scale monitoring of agricultural landscapes. The studies on the biodegradable sensors and quantum sensing technologies will also help to expand the agricultural monitoring further by leading to the creation of fully automated smart farming systems that can detect the stress of crop, forecast harvest, and manage resources in real time.

9.1 Quantum sensing

Quantum-enhanced optical measurement approaches, including squeezed-light techniques and advanced photon-counting detectors, are being investigated for improving the sensitivity of weak optical signals. Such approaches may eventually benefit applications such as chlorophyll-fluorescence sensing; however, their use in agricultural field environments remains largely at the research stage, and practical advantages over mature classical sensing systems have yet to be established.

9.2 Edge AI

Transmitting raw hyperspectral or high-resolution imagery from the field to a central server is often impractical given rural connectivity constraints. Edge AI running compressed inference models directly on drones, cameras, or IoT sensor nodes reduces bandwidth demand and enables near-real-time decisions (e.g., triggering targeted spraying) without continuous connectivity. This requires co-design of the sensing hardware, the compressed neural network, and the power budget of battery- or solar-powered field nodes.

9.3 Digital twins for agriculture

Optical sensing systems can provide spatially and temporally resolved observations that may contribute to the data streams used to update agricultural digital-twin models. Coupling such models with weather forecasts and climate projections is an emerging research direction for evaluating management strategies under potential drought, heat and other climate-stress scenarios. The reliability of these applications depends on sensor quality, data assimilation, model calibration and validation, and the representativeness of the underlying field data.

9.4 Other emerging directions

• Federated learning across farms/regions, allowing shared model improvement without centralising raw farm data.
• Biodegradable and fully wearable plant sensors that reduce field waste and enable dense, low-cost sensor networks.
• Multi-platform data fusion (Satellite + UAV + Ground) for consistent multi-scale monitoring across an agroecological region.
• Integration of sensor provenance data with blockchain-based traceability systems for supply-chain and carbon-credit verification.

10 Conclusion

Photonics and optoelectronics furnish the technological basis for climate resilient smart agriculture by enabling non-destructive, spatially resolved and increasingly real time measurement of crop, soil and environmental parameters. Nevertheless, the evidence reviewed suggests that no single sensing modality can deliver a holistic representation of the state of crops in a field environment. Multispectral and RGB sensors provide favourable trade-offs of cost, maturity and scalability, while hyperspectral, thermal, fluorescent, LIDAR and fiber-optic sensors provide more specialized data but have additional challenges regarding cost, calibration, environmental effects, spatial coverage or computation power. The important challenge is thus not in making sensors more sensitive, but rather in bringing together different sensor modalities in combination with IoT communication, AI/ML and weather and agronomic modelling. The fusion of different sensors improves stress discrimination and reliability of decision-making, even though interoperability, synchronization, data standardization, transferability and computational needs are among the most difficult challenges. On the other hand, high prediction accuracy achieved through controlled or one-site experiments is not necessarily transferable to the field and economically feasible. Thus, future studies should focus on independent validation over several seasons and at multiple locations, uncertainty quantification, benchmarking, edge computing, cost-effectiveness analysis and sensor architectures. The success of the transition towards climate resilient precision agriculture is contingent upon balancing sensing capability and cost, reliability and interpretability and conversion of heterogeneous sensor readings to farm management decisions.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Sreed Sharma Kanakkillam, University of Ljubljana, Slovenia

Reviewed by: Ravi Ray Chaudhari, Amity University, Gwalior, India

Domna Kotsifaki, Duke Kunshan University, China