Smart water management in agriculture: convergence of sensors, imaging, AI, precision irrigation, and robotics systems

1 Introduction

Agriculture sources around 70 to 80% of global freshwater extraction, and this demand is under increasing pressure from climate change, shifting rainfall patterns, and rapid population growth (Abdelhamid et al., 2025). Conventional gravity-based irrigation systems, including basin, border, and furrow flooding, waste nearly 60% of applied water through runoff, deep percolation, and evaporation, while traditional irrigation scheduling based on fixed calendars or visual observations does not explain soil heterogeneity, microclimatic variability, or real-time crop water status (Abdulrasaq et al., 2026; Aghaabbasi and Ganjouei, 2026). Surplus irrigation leads to waterlogging, soil salinization, nutrient depletion, and reduced agricultural productivity (Ahsen et al., 2025), while intensive groundwater extraction, which accounts for approximately 70% of groundwater extraction, causes declining water tables, salinization, and nitrate contamination of aquifers (Colizzi et al., 2025; Dercas et al., 2025). Although pump irrigation systems reduce field-level water losses, they increase energy requirements and operational costs because of continuous pumping demands (Luque-Sánchez et al., 2025). Agricultural water management has evolved from surface gravity irrigation systems with 40–60% efficiency to pressure sprinkler and drip irrigation systems with 70–90% efficiency, followed by automated irrigation using timers and basic soil-moisture sensors and, more recently, smart Internet of Things-enabled predictive systems (Moustafa et al., 2026). The automation level presents a progression from manual irrigation to precision, data-driven irrigation, and fully autonomous closed-loop irrigation systems, where conventional methods respond after soil moisture reaches the wilting stage, whereas predictive systems evaluate future water requirements before crop stress develops (Aghaabbasi and Ganjouei, 2026; Dantas et al., 2025). Smart water management in agriculture significantly relies on sensor and imaging technologies, Internet of Things-based precision irrigation systems, and agricultural robotics for improved crop water monitoring and management. Recent advances in artificial intelligent models include Convolutional Neural Networks used to interpret moisture-stress signals directly from crops and for high-precision evapotranspiration estimation using limited climatic inputs, while Model Predictive Control supports dynamic irrigation-depth optimization under changing environmental conditions (Chen J. et al., 2026; Becker and Zohdi, 2026). Edge-computing architectures perform Artificial Neural Network and Convolutional Neural Network models directly on local microcontrollers for real-time decision-making, and combining Wireless Sensor Networks with closed-loop automated irrigation systems attains water savings ranging from 15% to more than 30% while improving crop productivity (Abdelmoneim et al., 2025). Ground-based sensing technologies include vegetation spectral analysis, phenocams for crop phenology monitoring, SPAD meters for leaf nitrogen measurement, LAI meters for canopy characterization, and infrared thermometers for canopy-temperature measurement linked with stomatal activity (Allu and Mesapam2025). Agricultural imaging systems link thermal and multispectral sensing for crop water-stress measurement, while UAV platforms use infrared thermal cameras operating within the 8–14 μm range and multispectral sensors for foliage-temperature mapping and Crop Water Stress Index estimation (Ahmad et al., 2021).

Artificial intelligence applications are used for spatial soil–water interaction modelling, combining crop and water-balance equations with reinforcement learning for irrigation scheduling and optimization, while Convolutional Neural Networks support water-stress classification from hyperspectral and multispectral imagery and Random Forest models classify noisy sensor data (Madhuri et al., 2026; Gao et al., 2023; Oğuztürk, 2025). Resource-constrained systems use ESP32-based Fuzzy Sugeno controllers for local irrigation and fertigation control, while artificial intelligence-based clustering supports Variable Rate Irrigation through delineation of homogeneous management zones and site-specific water application (Wijayakusuma et al., 2025). Field robotic platforms link UAVs and autonomous ground vehicles for monitoring and site-specific operations, with multi-robot systems coordinating aerial water-stress surveillance with ground interventions, machine vision, GIS documentation, penetrometers, and precision-spraying modules (Zhang, 2023). Agriculture is increasingly challenged by freshwater scarcity, groundwater depletion, climate variability, and inefficient irrigation practices. Despite these advances, conventional irrigation still faces limitations due to uniform water application, limited consideration of spatial variation in soil and crop conditions, fixed or single-parameter scheduling, and water losses through evaporation, runoff, and deep percolation. This review therefore highlights studies covering smart irrigation and water-saving approaches using AI-driven sensors and imaging, Internet of Things-based precision irrigation, Variable Rate Technologies, digital twins, decision-support systems, and UAV–UGV robotics for more responsive crop-water management. The review also reflects the integration of sensors, imaging, IoT, precision irrigation, and agricultural robotics for real-time measurement of soil-water status, crop stress, evapotranspiration, and field variability. The reviewed also addresses water savings, improved water-use efficiency, crop performance, and site-specific irrigation control, while their integration provides broader relevance for efficient and climate-resilient agricultural water use.

2 Review methodology

A structured narrative review was conducted to show the development and integration of technologies for smart agricultural water management, with emphasis on irrigation scheduling, site-specific water application, water-use efficiency, and climate-resilient irrigation systems. The literature search was conducted using Scopus, Web of Science, ScienceDirect, and Google Scholar across 141 publications with combinations of keywords related to smart water management, precision irrigation, crop water requirement, soil moisture, plant water stress, evapotranspiration, sensors, imaging, remote sensing, Internet of Things, artificial intelligence, machine learning, deep learning, wireless sensor networks, unmanned aerial vehicles, unmanned ground vehicles, variable-rate irrigation, digital twins, decision-support systems, robotics, irrigation scheduling, and water-use efficiency. Additional searches combined technology-related terms with specific water-management parameters and crop applications to identify studies relevant to crop, soil, weather, sensing, modelling, decision-making, and irrigation management. The literature was selected based on its relevance to agricultural water management and its contribution to water-status assessment and autonomous management while also studies with limitation in these models for scope of improvement in existing ones. The selected literature was organized into thematic areas covering relevant areas from conventional analytical methods to advanced technologies for smart water management. Information was extracted according to the technology used, target water-management parameter, operating principle, application, reported benefits, current bottlenecks, data availability and quality, and future improvement areas. The review aimed to provide an integrated understanding of how crop, soil, and weather parameters can be synergistically linked with real time sensing, imaging, data communication, artificial intelligence-based analysis, decision support systems, variable-rate application, automation, and robotics, while identifying limitations in data availability, data quality, model development, validation, and generalization that influence crop-specific and site-specific water management. Thus, the review contributes a connected framework linking water-status assessment and crop water requirement estimation with prediction, decision-making, irrigation application, monitoring, and adaptive feedback, while identifying current research gaps and future directions for climate-resilient agricultural water management.

3 Conventional approaches for water management

Conventional water management systems generally rely on scheduled water application and manual field information. Traditional systems are mainly gravity based and pumped driven, using field slope and natural water flow for distribution, with basin, border, and furrow flooding achieving approximately 40%, 40–60%, and 50–70% efficiency, respectively (Lakhiar et al., 2024; Anjum et al., 2023). These systems generally treat fields as homogeneous units and apply water uniformly without accounting for soil or topographic variability, resulting in variable soil-moisture conditions (Elhoseny et al., 2026; Zhang, 2023). Fixed irrigation schedules may not reflect weather, crop growth stages, or actual crop water requirements (Dantas et al., 2025; Kaushik and Singh, 2025). Open-channel irrigation systems use gravity-fed networks with transmission losses and limited farm-level coordination (El Hachimi et al., 2025). Conventional monitoring remains largely qualitative, based on soil cracking, leaf curling, and manual field examination, with limited continuous information on soil moisture, crop water status, and environmental conditions (Lakhiar et al., 2024; Abdulrasaq et al., 2026), while irrigation schedules respond poorly to changes in temperature, relative humidity, and atmospheric demand (Ashraf et al., 2025). Surface irrigation can lose 30–50% of applied water and up to 60% in some regions through runoff, deep percolation, and evaporation (Abdulrasaq et al., 2026). Fixed schedules may reduce soil moisture below the wilting point or above field capacity, causing plant water stress or waterlogging (Quimbita et al., 2022), while approximately 70% of groundwater resources are allocated to irrigation activities (Sg and Selvaraj, 2026). Surplus and unmanaged irrigation can promote waterlogging, soil compaction, and salinization, while poor synchronization with crop growth stages can affect plant growth, photosynthetic activity, yield, and quality (Aghaabbasi and Ganjouei, 2026; Sg and Selvaraj, 2026; Abdulrasaq et al., 2026). Under prolonged drought, farmers may shift toward crops with lower water requirements (Luque-Sánchez et al., 2025), while groundwater depletion has been reported at 4 cm year⁻¹ in Punjab, India, 1.5 m year⁻¹ in California’s Central Valley, and 6 cm year⁻¹ in the North China Plain, with 52% of global cropland projected to experience extreme water stress by 2070 (Dutta et al., 2026). These conditions have supported the transition toward intelligent irrigation, where predefined moisture limits can sensitize responses to crop water stress (Aghaabbasi and Ganjouei, 2026). Reduced cost of Internet of Things device, wireless sensor networks, low-power communication, and cellular connectivity support continuous data collection and automated irrigation control (Adamo et al., 2025; Abdelmoneim et al., 2025). Conventional water management systems generally use scheduled irrigation based on crop water requirement, ET₀, crop coefficients (Kc), and soil-water availability. Fixed irrigation thresholds such as management allowable depletion (MAD) are used to determine irrigation timing. These systems provide a practical basis for water application using established soil, crop, and weather information, but do not inherently account for real-time spatial and temporal variability.

4 Crop water requirement and plant stress detection

Crop water requirement (CWR) represents the water needed by a crop to support evapotranspiration and growth under given crop, soil, and weather conditions. Accurate CWR measurement is essential for irrigation water productivity, while inaccurate estimation can cause under-irrigation, reduced growth and yield, or over-irrigation, resulting in salinization, waterlogging, and nutrient leaching (Abdulrasaq et al., 2026). Crop water demand varies with crop species, growth stage, soil, and environmental factors; maize requires approximately 4,800 m³ ha⁻¹ per growing season compared with about 4,000 m³ ha⁻¹ for wheat (Nasar et al., 2026). Soil evaporation dominates early growth, whereas transpiration becomes the main component with canopy development (Aghaabbasi and Ganjouei, 2026). Crop water requirements are commonly estimated using crop coefficients (K_c), ranging from 0.3–0.5 initially to 1.0–1.2 during vegetative and reproductive stages, while soybean is more sensitive to water stress during flowering and pod development (Dantas et al., 2025). Soil properties regulate water availability and irrigation frequency; clay-rich soils retain more water, whereas sandy soils drain rapidly (Capone et al., 2026). Total Available Water (TAW) ranges from 170 to 225 mm m⁻¹ in silty loam to 55–75 mm m⁻¹ in sandy soils, requiring more frequent irrigation in coarse-textured soils (Ding et al., 2025). Soil moisture is bounded by Field Capacity (FC) and Permanent Wilting Point (PWP), with Management Allowable Depletion (MAD) commonly around 50% of TAW (Oyo et al., 2026). Soil salinity, expressed as electrical conductivity (EC), increases osmotic potential, restricts root water uptake, and affects evapotranspiration and irrigation requirements (Bounajra et al., 2024). Weather conditions and evapotranspiration are primary factors of crop water demand. Reference evapotranspiration (ET₀) is governed by air temperature, relative humidity, wind speed, and solar radiation, while precipitation replenishes soil moisture, but insufficient or surplus rainfall can cause drought stress or waterlogging (Fuentes-Peñailillo et al., 2025; Batra et al., 2026). Increasing ET₀ from 4.5 to 6.2 mm day⁻¹ increased irrigation water requirements by 31%, whereas heavy rainfall reduced them by 45% (Popescu et al., 2025). Dual crop-coefficient models separate basal crop coefficient (K_cb) for transpiration and soil evaporation coefficient (K_e) for soil evaporation (Bounajra et al., 2024), while standardized FAO-56 crop coefficients can overestimate crop water requirements by 25.5% during tillering and 20.4% during maturity under specific regional conditions (Sarr et al., 2026). Above 30 °C with low atmospheric humidity, transpiration remains high for leaf cooling while carbon assimilation is limited (Song et al., 2026). Precision irrigation can provide farm-level water savings of up to 30%, although basin-level conservation may increase irrigated areas and regional water depletion (Dantas et al., 2025). Crop water requirement represents the water needed to maintain evapotranspiration and crop growth under given soil, crop, and weather conditions. It varies with growth stage, ET₀, Kc, soil-water availability, and environmental conditions, forming the basis for irrigation scheduling and water-use efficiency.

Plant Water Stress occurs when water availability becomes insufficient to meet crop water demand, causing changes in plant water potential, stomatal conductance, transpiration, canopy temperature, photosynthesis, and growth (Cudjoe et al., 2026; Ma et al., 2026). It develops with declining soil-water availability and can be assessed using soil, physiological, thermal, and spectral indicators, often before visible stress symptoms appear (Chen et al., 2025; Fevgas et al., 2025). Timely detection is important for crop productivity, water-use efficiency, and irrigation precision because water deficits restrict photosynthesis, nutrient uptake, and cell expansion, reducing biomass, yield, and crop quality (Chen N. et al., 2026), while in specialty crops they can also alter flavor- and aroma-related secondary metabolites and increase susceptibility to pests and pathogens (Sharma V. et al., 2025). Under soil-moisture deficit, declining leaf water potential (LWP) and increased vapor pressure deficit (VPD) trigger abscisic acid-mediated signalling, causing stomatal closing and reduced transpiration rate (Tᵣ) (Cudjoe et al., 2026). Reduced transpiration suppresses evaporative cooling, shifts energy balance from latent heat flux (LHF) to sensible heat flux (SHF), and increases canopy temperature (T_c) (Cudjoe et al., 2026), while reduced cellular turgor limits leaf-cell expansion, leaf area index (LAI), net photosynthetic rate (Pₙ), and biomass increase (Yang X. et al., 2025). Water-stress severity can be classified as minimal/no stress, moderate stress, or severe stress, corresponding to optimal soil moisture, mild drought, and extreme drought (Safdar et al., 2023). At the whole-plant level, stress increases daytime stem shrinkage and delays nocturnal rehydration because root water uptake cannot match daytime transpiration (Ke et al., 2026). Cellular dehydration promotes osmotic adjustment through proline and sugars, which lower leaf osmotic potential and maintain turgor (Le et al., 2023), while chlorophyll-a, chlorophyll-b, and total chlorophyll decrease linearly with root-zone water deficit (R² ≥ 0.97), altering light absorption and spectral reflectance (Salah Ud Din et al., 2025). Water deficits also generate reactive oxygen species (ROS), activating enzymatic antioxidants such as superoxide dismutase, catalase, and peroxidase, together with phenols, flavonoids, and glutathione (Moyankova et al., 2023). Additional responses include leaf curling, canopy thinning, reduced leaf area, and structural root-system elongation for deeper soil-water access (Cudjoe et al., 2026). Suppressed transpiration cooling increases canopy temperature, with water-stressed canopies exceeding 2 °C above air temperature (ΔT ≥ 2 °C), which can be normalized using the Crop Water Stress Index (CWSI) (Haq et al., 2025). The CWSI normalizes canopy-to-air temperature differences against wet and dry references to generate values from 0 to 1 (Sharma V. et al., 2025). Sun-Induced Chlorophyll Fluorescence (SIF), emitted between 650 and 800 nm and representing 1.0 to 3.5% of shortwave solar energy, provides a direct pre-visual indicator of photosynthetic function and short-term stress changes (Ahmad et al., 2021). Direct physiological indicators such as stomatal conductance (Gₛ), transpiration rate (Tᵣ), and leaf or stem water potential (Ψ) provide additional water-status information, although conventional measurements are labour-intensive and spatially localized (Ma et al., 2026; Sharma V. et al., 2025). Plant Water Stress: Plant water stress develops when soil-water availability becomes insufficient to meet crop water demand, affecting Ψ, Gₛ, Tᵣ, canopy temperature, photosynthesis, and growth. It can be detected using soil, physiological, thermal, spectral, and electrophysiological indicators, often before visible wilting, supporting timely irrigation management.

5 Technological advances in smart water management

Modern technologies are enabled in the current field water management for irrigation practices. These systems use AI-driven sensors and imaging, IoT-based precision irrigation, variable-rate technologies, digital twins, decision-support systems, and agricultural robotics for real-time monitoring and management of soil, crop, weather, and irrigation conditions (Aghaabbasi and Ganjouei, 2026; Moustafa et al., 2026). These technologies support crop water-demand estimation, water-stress assessment, adaptive irrigation scheduling, site-specific water application, and closed-loop control across heterogeneous agricultural fields (Wang et al., 2025; Sarr et al., 2026). Smart water management in agriculture integrates sensing, imaging, AI/ML/DL, IoT, VRT, decision-support systems, and robotics for site-specific irrigation management. Soil and plant sensing using capacitive/FDR/TDR sensors, tensiometers, electrophysiological sensors, and water-potential measurements provides soil- and plant-water status, while RGB, multispectral, hyperspectral, thermal, and LiDAR imaging provides spatial information on crop stress, evapotranspiration, and water status. These data are processed using classification, and optimization models for water-demand prediction, ET estimation, stress detection, and irrigation scheduling. IoT and edge–cloud systems connect sensors, communication networks, and automated valves for real-time monitoring and closed-loop irrigation as shown in Figure 1, respectively and how these technologies work together were shown in a closed loop precision irrigation system for smart water management as presented in Figure 2, respectively. The resulting information supports map-based and sensor-based VRT for site-specific water application and is integrated with digital twins and decision-support systems, i.e., Irri_DesK, IrriTwin, GreenLab, etc. for irrigation prediction and adaptive scheduling. UAV, UGV, and AGV robotics further support field-scale mapping, localized sensing, leak detection, and irrigation intervention, linking sensing, imaging, data processing, decision-making and finally to site-specific irrigation (Aghaabbasi and Ganjouei, 2026; Wang et al., 2025b; Sarr et al., 2026). Modern technologies integrate AI-driven sensing and imaging, IoT-based precision irrigation, VRT, digital twins, DSS, and agricultural robotics for real-time water management. Learning models support crop water-demand prediction, ET estimation, stress detection, and irrigation scheduling from sensor and imaging data. IoT, UAV/UGV systems, and edge–cloud platforms enable continuous monitoring, automated control, and localized intervention. Together, these technologies support adaptive, site-specific irrigation and improved water-use efficiency.

6 Intelligent sensing systems for smart water management

6.1 Soil sensors

Soil sensors are widely used for irrigation scheduling and root-zone water management. Capacitive and Frequency Domain Reflectometry (FDR) probes estimate soil-water content from dielectric permittivity, exploiting differences between water (80), dry soil (3–5), and air (1), whereas Time-Domain Reflectometry (TDR) determines volumetric water content (VWC) from electromagnetic pulse travel time (Kumar et al., 2026b).

Tensiometers measure soil matric potential, neutron probes measure moisture through hydrogen-neutron interactions, and gypsum blocks estimate matric potential from electrical resistance (Patel et al., 2022). In semi-arid sweet corn, capacitive sensing triggered irrigation at ≤33.1% VWC and termination at 43.5% or 34.8% VWC, increasing grain and stalk yields by 12.05 and 14.97% and saving up to 24.5% water compared with conventional crop evapotranspiration (ETc)-based scheduling (Sharma H. et al., 2025). In olive orchards, sensors at 15, 35, and 50 cm maintained 24.4–46.7% soil moisture, approximately 2.8 m³ h⁻¹ flow, and prevented deep percolation (Meriç, 2025). Multi-depth sensing defines refill limits and active-root-zone moisture boundaries, reducing root hypoxia and moisture-deficit effects (Ke et al., 2026). Sensor networks are also arranged in 2 × 5 grids at 38–40 m spacing to capture field heterogeneity (Zhang et al., 2024; Qin et al., 2025) and monitor VWC, soil matric potential, temperature, electrical conductivity (EC), pH, and NPK (Gopal et al., 2025). TDR provides approximately ±2% VWC accuracy, calibrated capacitive sensors achieve R² = 0.94–0.99, tensiometers operate at 0–85 kPa with ±2 kPa accuracy, and neutron probes profile soil moisture to 3–4 m (Gopal et al., 2025; Arafa et al., 2024; Patel et al., 2022). Key limitations include corrosion and salinity effects in resistive probes, saline-condition calibration requirements for capacitive sensors, tensiometer cavitation above approximately 85 kPa, gypsum-block degradation, and radioactive safety concerns for neutron probes (Zhang et al., 2024; Patel et al., 2022). Soil sensors provide continuous measurements of volumetric water content (VWC), soil-water potential, temperature, EC, pH, and nutrients for irrigation management. Capacitive/FDR/TDR sensors estimate water content from dielectric properties, while tensiometers and gypsum blocks assess soil-water potential or resistance. Multi-depth sensing improves assessment of root-zone water availability, supporting irrigation triggering and site-specific water application.

6.2 Plant-based and crop physiology sensors

Plant-based sensing technologies directly measure crop physiological responses to water stress. Infrared Thermometry (IRT) detects canopy thermal radiation, where reduced transpiration from stomatal closing increases canopy temperature above ambient conditions (Gopal et al., 2025). Sap-flow sensors measure xylem water movement, dendrometers measure stem and fruit diameter fluctuations, and Leaf Patch Clamp Pressure (LPCP) sensors monitor leaf turgor pressure (Gopal et al., 2025; Singh and Singh, 2026). Electrophysiological (EP) sensors detect rapid plant electrical responses to moisture fluctuations within 1–5 min; Convolutional Neural Network (CNN) classification of irrigation levels achieved 83% accuracy with 10–20% lower water use without tissue damage (Chen et al., 2025). The PlantRing sensor monitored stem and fruit diameter micro-variations for deficit irrigation in tomato, increasing fruit soluble solids by 14.7% over well-watered treatments (Ke et al., 2026). Integration of thermal-dissipation sap-flow probes with Internet of Things systems reduced vineyard water use by 20–30%, while continuous stem water potential (Ψstem) monitoring in olive and almond improved WUE by 25–35% without reducing yield (Ke et al., 2026). Thermal sensors are positioned above or beside crop canopies, whereas sap-flow probes require insertion into active xylem tissue (Gopal et al., 2025; Singh and Singh, 2026). These systems monitor canopy temperature (T_c), Crop Water Stress Index (CWSI), actual crop transpiration (Tact), stem growth, trunk shrinkage, leaf turgor, and stem water potential (Ke et al., 2026). Thermal sensors operate from −10 to 60 °C with ±0.5 °C accuracy, and plant-based indicators can detect water stress 3–7 days before visible symptoms with strong correlation to stem water potential (r = 0.89) (Gopal et al., 2025; Nxumalo et al., 2026). Conventional pressure chambers and porometers remain destructive and labor-intensive, supporting the use of direct physiological monitoring for early water-stress detection (Ke et al., 2026). Plant-based and crop physiology sensors provide direct assessment of canopy temperature, sap flow, stem/fruit diameter, leaf turgor, and Ψ_stem for evaluating plant water status. Thermal sensors detect increased canopy temperature and CWSI associated with stomatal closure, while sap-flow and dendrometric measurements indicate plant water use and dehydration. Electrophysiological (EP) sensors provide rapid water-stress signals at 1–5 min intervals, supporting early stress detection and irrigation control.

6.3 Environmental and meteorological sensors

Environmental and meteorological sensors provide atmospheric inputs for evapotranspiration estimation and climate-responsive irrigation scheduling, including air temperature, relative humidity, wind speed and direction, solar radiation, precipitation, and atmospheric pressure measured using NTC thermistors, capacitive polymer-film sensors, anemometers, pyranometers, rain gauges, and piezoresistive MEMS devices, typically integrated into automatic weather stations (AWS) and localized microclimate systems (Singh and Singh, 2026; Srivastav et al., 2026). In lettuce, integrating canopy temperature-derived crop water stress index (CWSI), soil moisture, and vapor-pressure deficit (VPD) within a fuzzy-logic model reduced water consumption by 59.61%, electrical-energy use by 67.35%, and increased yield by 22.58%, while in wheat, weather-station data with an linked sensing device for irrigation (ISDI) indicated irrigation initiation at CWSI ≥0.35 for grain production and water conservation (Sharma H. et al., 2025; Ke et al., 2026). Weather forecasts integrated with AquaCrop soil-water balance models can dynamically adjust Management Allowable Depletion (MAD) and suspend irrigation when significant rainfall is predicted (Nasar et al., 2026). Key operational limitations include calibration drift, mounting-height effects, structural degradation, and obstruction by leaves, spider webs, and bird nests (Singh and Singh, 2026; Saha and Rastogi, 2026). Environmental and meteorological sensors provide air temperature, relative humidity, wind speed, solar radiation, rainfall, and atmospheric pressure as key inputs for ET estimation and climate-responsive irrigation scheduling. These measurements help account for VPD, evapotranspiration, rainfall, and changing weather conditions in irrigation decisions. Integration with weather forecasts can further adjust MAD and irrigation timing, including suspension of irrigation during significant rainfall.

6.4 Water quality and flow sensors

Water-quality and flow sensors monitor irrigation-water characteristics, hydraulic performance, and distribution-network integrity. pH sensors measure acidity and alkalinity, EC sensors salinity, turbidity sensors optical scattering, turbine flow meters discharge, ultrasonic sensors water level, and pressure transducers hydraulic pressure (Singh and Singh, 2026; Oudah et al., 2024; Halder et al., 2025; Gopal et al., 2025). The Flow Pilot system for chili pepper integrates LoRaWAN, soil and weather sensors, in-line flow meters, and automated tank controllers for water-delivery regulation and real-time soil pH adjustment (Moustafa et al., 2026). ESP32-based groundwater systems monitor aquifer level, dissolved oxygen, temperature, and pH, while prepaid radio-frequency smart-card flow meters support-controlled water allocation at pump stations (Wang et al., 2026; Oudah et al., 2024). These systems monitor pH, ECw, turbidity, dissolved oxygen, nitrate, pressure, discharge, and total water volume; turbine, ultrasonic, and pressure sensors provide approximately ±2%, ±0.5–1%, and ±0.5% accuracy, respectively, with pressure transducers operating over 0–10 bar (Nxumalo et al., 2026; Gopal et al., 2025). Real-time pressure monitoring maintained approximately 2.8 m³ h⁻¹ while distributing 2,209 m³ irrigation water during the growing season (Meriç, 2025), and combined flow and pressure sensing supports hydraulic auditing and anomaly detection for pipeline failures, line blowouts, and emitter clogging (Meron et al., 2025). Key limitations include electrode corrosion, salt-crystal accumulation, pressure surges, and leakage-related measurement anomalies (Singh and Singh, 2026; Meriç, 2025). Water quality and flow sensors monitor pH, ECw, turbidity, dissolved oxygen, nitrate, pressure, discharge, and total water volume for irrigation-system assessment. Turbine flow, ultrasonic level, and pressure sensors quantify water movement and hydraulic conditions, supporting continuous flow and pressure monitoring. Their integration enables hydraulic auditing, delivery regulation, leak/clogging detection, and real-time irrigation control.

6.5 Acoustic and ultrasonic sensors

Acoustic and ultrasonic sensing technologies support leak detection, hydraulic diagnostics, and early crop-water-stress identification. Acoustic Emission (AE) sensors use piezoelectric transducers to detect high-frequency elastic-wave bursts from xylem cavitation during drought, while ultrasonic sensors assess water depth and flow through sound-wave propagation and reflection (Saha and Rastogi, 2026; Nxumalo et al., 2026). Acoustic leak-detection sensors identify pipeline leakage with 95–98% accuracy and localize failures within <1 m, while hydroacoustic tomography detects sediment clogging in groundwater irrigation systems (Saha and Rastogi, 2026; Nxumalo et al., 2026). Ultrasonic AE sensors attached to plant stems detect xylem cavitation under osmotic stress, with cavitation measurements correlating with stem water potential (r = 0.89) and detecting plant-water stress 3–7 days before visible wilting. AE sensors are attached to stems and woody branches, leak sensors along pressurized pipelines, and hydroacoustic arrays in groundwater irrigation wells (Saha and Rastogi, 2026; Nxumalo et al., 2026). These systems monitor cavitation frequency, acoustic-wave patterns, water depth, flow characteristics, and pressure anomalies; hydroacoustic tomography provides approximately 85% sediment-clogging detection accuracy in groundwater irrigation wells (Nxumalo et al., 2026). A multi-layer digital-twin model combining point-based acoustic-emission cavitation signals with regional thermal remote sensing has been proposed for deficit-irrigation scheduling. Key limitations include point-specific measurements, limited spatial scalability, sensor-coupling instability, environmental noise, and signal attenuation in dense crop canopies (Saha and Rastogi, 2026) (Table 1).

Table 1. Artificial intelligence models for different crops, their performance in smart water management.

AI, Artificial Intelligence; ANN, Artificial Neural Network; AWD, Alternate Wetting and Drying; CNN, Convolutional Neural Network; DDPG, Deep Deterministic Policy Gradient; DL, Deep Learning; EC, Electrical Conductivity; ET, Evapotranspiration; ET₀, Reference Evapotranspiration; F₁-score, F1 Score; GBDT, Gradient Boosting Decision Tree; GLI, Green Leaf Index; GNSS, Global Navigation Satellite System; IoT, Internet of Things; Kₛ, Soil Water Stress Coefficient; KNN, K-Nearest Neighbors; LAI, Leaf Area Index; LSTM, Long Short-Term Memory; LST, Land Surface Temperature; MAE, Mean Absolute Error; ML, Machine Learning; MLP-ANN, Multilayer Perceptron Artificial Neural Network; NDMI, Normalized Difference Moisture Index; NDVI, Normalized Difference Vegetation Index; NDWI, Normalized Difference Water Index; NIR, Near-Infrared; RF, Random Forest; RH, Relative Humidity; RMSE, Root Mean Square Error; RS, Remote Sensing; Rₙ, Net Radiation; R², Coefficient of Determination; SAC, Soft Actor-Critic; SVM, Support Vector Machine; Tₐ, Air Temperature; Tmax, Maximum Temperature; Tmin, Minimum Temperature; TIR, Thermal Infrared; Umean, Mean Wind Speed; VWC, Volumetric Water Content; ViT, Vision Transformer; VPD, Vapor Pressure Deficit; WUE, Water Use Efficiency; XGBoost, Extreme Gradient Boosting.

Acoustic and ultrasonic sensors support detection of xylem cavitation, water depth, flow, and pipeline leakage or clogging in irrigation systems. Acoustic emission (AE) sensors detect drought-induced cavitation and can indicate plant water stress before visible wilting. Pipeline acoustic sensing enables leak detection and localization, while hydroacoustic methods support sediment-clogging assessment and hydraulic monitoring.

7 Imaging technologies for solving agricultural water challenges

7.1 Satellite based remote sensing

Satellite-based remote sensing obtains crop and soil information from reflected or emitted electromagnetic radiation without direct physical contact (Abdulrasaq et al., 2026). Passive optical radiometers operate in visible, near-infrared (NIR), and shortwave infrared (SWIR) regions, while active Synthetic Aperture Radar (SAR) transmits microwave pulses and measures backscatter related to soil-canopy dielectric properties and water content (Krishnan et al., 2023; Yang X. et al., 2025). Landsat-8/9, Sentinel-1/2, MODIS, Aqua, Aura, and ECOSTRESS provide wide-area, multi-band, and multi-temporal observations (Ahmad and Sohel, 2025), with Sentinel-1 C-band SAR enabling day–night, all-weather monitoring and penetration of cloud cover, vegetation canopies, and shallow soil layers (<5 cm) (Demissie et al., 2026; Manivasagam, 2024). MODIS provides daily 250–1,000 m observations, Landsat-8/9 16-day observations with 30 m optical and 100 m thermal bands, Sentinel-2 10–20 m imagery at 5-day intervals, and Sentinel-1 5–40 m imagery at 6-day intervals; commercial CubeSats provide daily 3–5 m observations, while WorldView-3 provides 0.3–3.7 m imagery at approximately 1–4-day intervals (Agyemang et al., 2026; Haq et al., 2025).

Satellite processing generates Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil-Changeed Vegetation Index (SAVI), land surface temperature (LST), actual evapotranspiration (ET_a), crop coefficients (K_c), equivalent water thickness (EWT), and microwave-derived topsoil volumetric moisture estimates (Ahsen et al., 2025; Cudjoe et al., 2026). These products support drought monitoring, water-budget measurement, irrigation-oasis mapping, center-pivot identification, and water-productivity evaluation across agricultural watersheds and river basins (Abdulrasaq et al., 2026; Dutta et al., 2026; Nxumalo et al., 2026). Applications include MODIS NIR–SWIR NDWI retrieval of maize vegetation water content at watershed scale (Cudjoe et al., 2026), Landsat–MODIS STARFM–SEBAL estimation of sugarcane actual evapotranspiration and soil-water balance with R² = 0.87–0.91 under semi-arid conditions (Allu and Mesapam, 2025), and Sentinel-2/Landsat-8 NDVI, SAVI, and NDWI integrated with in-situ soil moisture for avocado irrigation scheduling and crop-water requirement assessment (Torres-Quezada et al., 2025). Multi-temporal observations also support agricultural drought monitoring, crop-water consumption, evapotranspiration changes, and irrigation-performance assessment across heterogeneous regions (Abdulrasaq et al., 2026; Dutta et al., 2026). Despite these applications, satellite observations are constrained by revisit intervals, cloud and dust interference, and mixed-pixel effects in heterogeneous agricultural systems (Cudjoe et al., 2026; Demissie et al., 2026; Sarr et al., 2026), while persistent cloud cover and coarse-resolution imagery reduce field-level measurement accuracy in fragmented landscapes (Demissie et al., 2026; Sarr et al., 2026). Integration with ground sensing, UAV platforms, and data-fusion methods is therefore used to improve spatial and temporal monitoring (Haq et al., 2025; Allu and Mesapam, 2025). Satellite-based remote sensing includes Sentinel-2, Sentinel-3, Landsat-8, and Harmonized Landsat and Sentinel-2 (HLS) data with Leaf Area Index (LAI), Soil Moisture (SM), Land Surface Temperature (LST), and Fraction of Absorbed Photosynthetically Active Radiation (fAPAR) for crop-water assessment as shown in Table 2, respectively. These observations support evapotranspiration estimation, Crop Water Stress Index (CWSI) assessment, Variable Rate Irrigation (VRI), and Regulated Deficit Irrigation (RDI) through integrated models and systems. Satellite-based remote sensing uses passive optical VIS/NIR/SWIR, thermal infrared, and active SAR observations to assess crop and soil-water conditions over large areas. It provides NDVI, NDWI, EVI, SAVI, LST, SIF, ET_a, K_c, EWT, and soil-moisture information for drought, water-stress, irrigation, and water-budget assessment. Sentinel-1/2, Landsat, MODIS, and ECOSTRESS enable repeated spatial monitoring, while integration with ground and UAV observations improves field-level water management.

Table 2. Satellite-based remote sensing technologies and indices for smart water management.

NDVI, Normalized Difference Vegetation Index; CWSI, Crop Water Stress Index; SAR, Synthetic Aperture Radar; LST, Land Surface Temperature; METRIC, Mapping Evapotranspiration at High Resolution with Internalized Calibration; SEBAL, Surface Energy Balance Algorithm for Land; SMAP, Soil Moisture Active Passive; SMOS, Soil Moisture and Ocean Salinity; Kₛ, Water Stress Coefficient; NDMI, Normalized Difference Moisture Index.

7.2 Low altitude imaging

Unmanned Aerial Vehicles (UAVs) provide low-altitude remote-sensing imaging below cloud cover, with high-resolution georeferenced imagery acquired on demand (Ahmad and Sohel, 2025). UAV systems commonly use Structure-from-Motion (SfM) photogrammetry for orthophotos and three-dimensional surface reconstruction, with radiometric calibration panels and onboard radiometers for absolute surface-reflectance measurements (Bwambale et al., 2022; Yacoob et al., 2026b). Fixed-wing, multi-rotor, and vertical takeoff and landing (VTOL) platforms are used for large-area, hovering/plot-level, and fragmented-landscape surveys, respectively, while RTK/PPK GNSS and inertial navigation improve geometric accuracy and reduce positioning errors (Mowla et al., 2026). UAVs typically provide 1–10 cm pixel⁻¹ spatial resolution, with millimeter-level resolution at very low altitudes and customizable return frequency (Demissie et al., 2026; Zhang, 2023). UAV data generate canopy-cover maps, plant height, Digital Surface Models (DSMs), three-dimensional point clouds, stand density, and Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge Index (NDRE), and Photochemical Reflectance Index (PRI) (Mowla et al., 2026; Ndlovu et al., 2024). These outputs support crop phenotyping, micro-topographic analysis, stress-hotspot identification, variable-rate irrigation zoning, and early pest and disease detection (Huang et al., 2026; Mowla et al., 2026; Ndlovu et al., 2024). UAV-based sensing supports crop-water monitoring across crops; in winter wheat, early-afternoon acquisition during grain filling was identified for discriminating drought responses, while in cotton, the Water Deficit Index (WDI) outperformed the Crop Water Stress Index (CWSI) during early vegetative stages, with 0.5 m spatial resolution providing a balance between computational efficiency and diagnostic performance. In vineyards, nadir and oblique imagery with three-dimensional canopy reconstruction improved canopy delineation and high-resolution crop-water-stress mapping (Cudjoe et al., 2026). High-resolution imagery supports canopy characterization, vegetation strength, field heterogeneity, and spatially targeted irrigation, while multispectral, thermal, and RGB integration extends monitoring across growth stages and environmental conditions (Mowla et al., 2026; Ndlovu et al., 2024; Huang et al., 2026; Mowla et al., 2026). UAV operations remain constrained by limited battery life, flight endurance, payload capacity, weather dependency, operational costs, and aviation regulations (Ahmad and Sohel, 2025; Ndlovu et al., 2024). Mission planning, image processing, data storage, and the expertise required for high-resolution data interpretation further limit operational use (Fuentes-Peñailillo et al., 2025; Mowla et al., 2026). Low-altitude imaging using UAVs provides high-resolution, georeferenced observations of crop and field variability at cm-scale spatial resolution. RGB, multispectral, hyperspectral, thermal, and LiDAR systems support canopy mapping, water-stress detection, microtopography, and VRT zoning. Flexible flight planning and repeated surveys enable timely monitoring of spatially variable crop-water conditions.

7.3 RGB imaging

RGB imaging systems use silicon-based optical sensors sensitive to the visible spectrum (approximately 400–750 nm), recording reflected radiation in red (600 nm), green (550 nm), and blue (450 nm) regions (Cho et al., 2024). Data are acquired using handheld digital cameras, smartphones, surveillance systems, or UAV-mounted RGB cameras under stable illumination to minimize shadowing and lighting variability (Batra et al., 2026; Ding et al., 2025; Yumnam et al., 2025). RGB imaging provides centimetre- to millimetre-level spatial resolution, low cost, and high temporal flexibility (Cho et al., 2024), while image processing derives canopy-cover fraction, Green Leaf Index (GLI), Excess Greenness Index (ExG), SfM-based plant height, and crop emergence or germination counts (Yacoob et al., 2026b; Yumnam et al., 2025). These data support crop-growth and phenological monitoring, Leaf Area Index (LAI) estimation, weed–crop differentiation, lodging detection, and visible late-stage water-stress assessment such as canopy thinning and leaf yellowing (Cho et al., 2024; Mowla et al., 2026). Machine-learning integration has enabled water-stress assessment: GoogLeNet classified winter-wheat stress at 92.9% accuracy within 200 ms, maize smartphone RGB imagery using ExG, CIVE, and a* achieved R² ≥ 0.97 with Random Forest out-of-bag error of 0.18, and UAV RGB with LightGBM predicted sugarcane leaf water potential with R² = 0.78 (Cho et al., 2024; Salah Ud Din et al., 2025; Mowla et al., 2026). However, RGB imaging cannot detect early physiological water stress because visible wavelengths do not directly capture stomatal closing or transpirational changes, and performance is affected by shadows, solar-angle variation, and soil-background reflectance (Alavi et al., 2024; Ndlovu et al., 2024; Arsenoaia et al., 2026; Fevgas et al., 2025). RGB imaging uses visible wavelengths (400–750 nm) to capture crop and canopy parameters such as canopy cover, GLI, ExG, plant height, emergence, and phenology. Its low cost and high spatial resolution support UAV, smartphone, and handheld monitoring of crop growth and visible stress. RGB-based analysis has supported chlorophyll-deficit assessment, plant detection, and crop-stress classification, although early physiological water stress may remain difficult to detect.

7.4 Multispectral and hyper spectral imaging

Multispectral imaging measures reflected electromagnetic energy through 4–12 discrete bands across visible, red-edge, and near-infrared regions using optical filters, prisms, or multi-lens assemblies (Ahmad et al., 2021; Haq et al., 2025). MicaSense RedEdge and Parrot Sequoia are commonly mounted on fixed-wing and multi-rotor unmanned aerial vehicles or satellite platforms (Batra et al., 2026; Ndlovu et al., 2024), with radiometric calibration using reflectance panels before and after flights (Yacoob et al., 2026b). Low-altitude unmanned aerial vehicles can provide approximately 5.2 cm spatial resolution at 120 m, whereas satellite imagery generally provides 10–30 m resolution (Yacoob et al., 2026a; Yacoob et al., 2026b). Multispectral data provide red, green, blue, red-edge, and near-infrared reflectance for calculating Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge Index (NDRE), Soil-Changed Vegetation Index (SAVI), and Enhanced Vegetation Index (EVI) (Yacoob et al., 2026b). These indices support variable-rate irrigation, cumulative water-deficit assessment, crop-coefficient mapping, canopy-nitrogen evaluation, and crop-density characterization (Ahmad and Sohel, 2025; Mowla et al., 2026; Ndlovu et al., 2024). In winter wheat, UAV multispectral imagery integrated with the PROSAIL radiative-transfer model and local meteorological data retrieved leaf area index (LAI), canopy chlorophyll content (CCC), and stomatal conductance (G_s), with rRMSE values of 0.109, 0.136, and 0.191, respectively (Yang et al., 2025b). UAV multispectral and thermal imagery with Gradient-Boosting Decision Tree (GBDT) estimated normalized greenness index and equivalent water content with R² = 0.88 and 0.90, respectively (Cho et al., 2024). In maize, UAV NDVI, SAVI, and EVI with basal crop coefficient (K_cb) and thermal-stress coefficients (Ks) estimated actual evapotranspiration (ET_a) with R² = 0.84 and RMSE = 0.50 mm day⁻¹ (Sarr et al., 2026). Multispectral and thermal imagery with Random Forest predicted maize Crop Water Stress Index (CWSI) with R² = 0.85 and RMSE = 0.05, while cotton UAV vegetation indices and texture features with Random Forest predicted leaf-water content and canopy-water status (Cho et al., 2024). Multi-source RGB, 19 multispectral indices, and thermal observations with XGBoost achieved cotton water-content prediction of R² = 0.860 across five irrigation regimes, outperforming Support Vector Regression (SVR) and Partial Least Squares Regression (PLSR). In sugarcane, Random Forest provided the highest salinity-stress prediction performance among Random Forest, Support Vector Machine, Cubist, and XGBoost, with R² = 0.90 and RMSE = 0.59 dS m⁻¹ (Mowla et al., 2026). In almond orchards, UAV multispectral reflectance and soil physicochemical properties combined with Random Forest and Artificial Neural Network (ANN) predicted stem water potential (SWP), providing an alternative to pressure-chamber measurements (Yang X. et al., 2025; Mortazavi et al., 2025). Multispectral vegetation indices, canopy reflectance, and machine-learning methods also support spatial assessment of crop intensity, irrigation variability, field heterogeneity, and crop performance (Ahmad and Sohel, 2025; Ndlovu et al., 2024; Mowla et al., 2026). However, broad spectral bands may miss subtle physiological and biochemical responses, vegetation indices may saturate in dense canopies, and limited Short-Wave Infrared (SWIR) capability restricts direct water-index calculation such as Normalized Difference Water Index (NDWI) (Kumari et al., 2025; Yacoob et al., 2026a, 2026b). Multispectral imaging uses 4–12 discrete VIS, red-edge, and NIR bands to assess crop and water-related conditions. It supports NDVI, NDRE, SAVI, EVI, Kcb, canopy density, water deficit, and stress assessment, with applications in VRT and irrigation management.

Hyperspectral imaging provides hundreds of narrow contiguous bands (<10 nm) across approximately 400–2,500 nm, covering visible, near-infrared, and shortwave infrared regions (Kumari et al., 2025; Zhang, 2023). Push-broom, whisker-broom, and plane-scan systems are commonly deployed on high-capacity unmanned aerial vehicles, helicopters, and robotic platforms (Mowla et al., 2026), with ground spectroradiometers supporting calibration and validation (Krishnan et al., 2023). These systems provide centimetre- to sub-meter-level spatial resolution but lower temporal resolution because of data-acquisition and processing requirements (Kumari et al., 2025; Mowla et al., 2026). Hyperspectral data support estimation of equivalent water thickness (EWT), leaf-water content (LWC), nitrogen, chlorophyll, carotenoids, and soil salinity, as well as drought detection up to 10 days before visible symptoms, differentiation of nutrient and water stress, and drought phenotyping (Yang et al., 2025b). In winter wheat, Continuous Wavelet Transform level-3 (CWT3) spectral and texture features (450–950 nm) combined with thermal infrared canopy temperature and CNN–LSTM predicted leaf-water content with R² = 0.827–0.836 across the phenological cycle (Kumari et al., 2025). In rice, UAV hyperspectral data with XGBoost showed improved water-status prediction using near-infrared and shortwave infrared bands, with R² > 0.85 (Mowla et al., 2026). In potato, 400–1,000 nm hyperspectral data identified Vogelmann Red Edge Index 1 (VOG REI 1) as the most effective soil-moisture indicator (R² = 0.90), followed by VOG REI 2 (R² = 0.886), with approximately 17% soil moisture maximizing dry-tuber yield (Sharma V. et al., 2025). UAV hyperspectral imaging identified 670–780 nm as the most sensitive pearl-millet range for water-stress detection; Support Vector Machine (SVM) achieved 95.38% classification accuracy and 80.76% at early stress stages (Cho et al., 2024). In sorghum, UAV-derived three-dimensional drought indices (TDDIe) with Random Forest Regression predicted vegetation-moisture content with r = −0.93 under fully irrigated conditions (Chen et al., 2024). UAV hyperspectral and multispectral indices, including NDVI and NDRE, differentiated physiological drought responses among peanut genotypes, while hyperspectral methods also support biochemical and physiological trait retrieval, spatial plant-water assessment, and drought-response characterization (Kumari et al., 2025; Yang et al., 2025b; Cudjoe et al., 2026). Major limitations include spectral redundancy and collinearity, advanced preprocessing requirements such as fractional derivatives, Savitzky–Golay filtering, and continuous wavelet transforms, and high sensor cost, payload weight, and software-compatibility constraints (Mowla et al., 2026; Yang et al., 2025a, 2025b; Yang X. et al., 2025; Arsenoaia et al., 2026; Kumari et al., 2025). Hyperspectral imaging records hundreds of contiguous bands across 400–2,500 nm, providing detailed spectral information on crop condition. It supports assessment of EWT, LWC, chlorophyll, carotenoids, salinity, and early drought stress, with applications in water-status detection and crop phenotyping.

8 Precision irrigation systems for site-specific water management

Precision irrigation technologies integrate variable-rate application with real-time field information to adjust irrigation and other inputs according to spatial and temporal variation in soil, crop, and water status. Map-based Variable Rate Technology uses apparent electrical conductivity-based management zones, Geographic Information System prescription maps, Global Navigation Satellite System positioning, and center-pivot speed or nozzle control, whereas sensor-based systems use real-time soil, crop, and plant observations for on-the-go adjustment; integration with Internet of Things networks, unmanned aerial vehicle and satellite imaging, artificial intelligence and machine-learning models, and digital twins supports dynamic prescriptions and site-specific application. A 0–0.40 m root-zone apparent electrical conductivity map based on 2,482 samples achieved a root-mean-square error of 0.097702, mean error of −0.000030, and mean deviation of −0.000013 (Millán et al., 2025). Map-based Variable Rate Technology with GreenSeeker reduced wheat nitrogen application by 26% and increased nitrogen recovery efficiency by 21.9% (Singh and Singh, 2026), while Geographic Information System zoning with Variable Rate Technology increased crop yield by 13.4% to 7,850 kg/ha, reduced fertilizer by 16.0% and water intake by 12.5%, and improved soil organic matter and nitrogen recovery efficiency by 16.0 and 25.9%, respectively (Batra et al., 2026). A retrofittable Variable Rate Applicator Controller achieved 96.17% accuracy, 95.15% precision, and 2.82% error (Abdalla and Mirzakhani Nafchi, 2024), while fuzzy center-pivot speed control achieved 37–47% accuracy compared with 35% under traditional control (Morchid et al., 2025a, 2025b). Variable Rate Technology also reduced nitrogen use by 20%, while commercial systems increased potato yield by 12% compared with uniform management (Kumar et al., 2026b; Ahmad and Sharma, 2023). IoT-based precision irrigation connects soil, plant, weather, flow, and pressure sensors with ESP32, NodeMCU, Raspberry Pi, and wireless networks such as LoRa/LoRaWAN, ZigBee, Wi-Fi, NB-IoT, 4G, and 5G. Real-time data are transferred to edge–cloud platforms for monitoring, irrigation scheduling, automated valve control, and closed-loop operation. Integration with AI/ML, VRT, UAV imaging, and digital twins supports dynamic and site-specific irrigation decisions.

Sensor-based variable rate technology uses real-time soil, crop, and plant physiological measurements. UAV microbolometer imaging estimated tomato stem water potential with R² = 0.81 and mean absolute error = 0.67 bars, compared with R² = 0.74 and mean absolute error = 0.63 bars for proximal sensing (Sharma V. et al., 2025). Tensiometer-based irrigation saved 2,024 L of water, reducing use by 27.29% compared with manually scheduled drip irrigation (Ashraf et al., 2025), while capacitive sensing in sweet corn increased productivity by 12.05% and reduced water use by 35.2% compared with standard crop evapotranspiration scheduling (Sharma H. et al., 2025). Soil-moisture-based drip irrigation reduced paddy-rice water use by 41.5% compared with flood irrigation and 13% compared with conventional drip irrigation (Oudah et al., 2024). Active sensing reduced rice fertilizer use by up to 45%, while plant-wearable electrophysiology estimated cotton water stress with R² = 0.98 (Kumar et al., 2026a; Gupta et al., 2026). A fuzzy controller combining Crop Water Stress Index and root-zone capacitive soil moisture reduced water use by 59.61% and energy use by 67.35%, while increasing yield by 22.58% (Sharma V. et al., 2025). Wireless Sensor Network-based sugarcane irrigation using crop evapotranspiration, crop coefficient, and daily weather reduced water use by 25.08%, while Levenberg–Marquardt localization improved spatial node accuracy by 19–58% and reduced water use by 28% (Morchid et al., 2026b; Morchid et al., 2025a). Real-time Variable Rate Technology integrates continuous sensing, machine learning, digital twins, remote sensing, and automated control. AgriSens improved crop yield by up to 10.21% (Moustafa et al., 2026). Random Forest prediction of crop water demand achieved R² = 0.93 for maize and 0.91 for wheat, with water savings of 30 and 27%, respectively (Nasar et al., 2026). An evapotranspiration-ARIMA controller reduced root-mean-square error by up to 33% and normalized root-mean-square error by 40% across 12-, 24-, and 48-h forecasts (Ali et al., 2026). An Artificial Neural Network classified potato and wheat irrigation requirements with 96.81% accuracy versus 91% for K-Nearest Neighbors and 87% for Support Vector Machine (Jaiswal et al., 2025). Real-time machine learning reduced lawn irrigation runoff by 74% with 87% classification accuracy, while machine-to-machine communication with reinforcement learning achieved 97.88% crop-water-demand accuracy (Morchid et al., 2026b; Morchid et al., 2026c). A soil-moisture and relative-humidity classifier achieved 99.3% accuracy, 99% precision, 98.5% recall, and 99% F1-score (Rani and Banu, 2026). Solar-powered ESP32 and LoRa-based control limited valve operation to 9.34% of engine running time compared with 20.34% for drip and 31.34% for manual irrigation (Gao and Li, 2025). AquaCrop-IoT achieved 32% wheat irrigation-water savings, ThingsBoard-based smart irrigation reduced water use by 70%, and the DT-SIS digital twin achieved 38% water savings with 16% higher crop yield than conventional manual scheduling (Morchid et al., 2026a; Morchid et al., 2025a; Salwadkar, 2024). Sensor-based VRT uses real-time soil and crop sensing to adjust irrigation according to spatial and temporal variation in field conditions. Capacitive/TDR sensors, tensiometers, canopy thermal sensing, and plant-based measurements provide inputs for on-the-go irrigation adjustment. Integration with IoT, AI/ML, and automated valve control supports site-specific water application and closed-loop irrigation.

9 Robotics for smart water management

Aerial robotics combines UAV sensing, imaging, navigation, communication, and autonomous flight for crop and soil-water monitoring, water-stress detection, evapotranspiration assessment, field mapping, and targeted irrigation through IoT, AI/ML, and ground robotic systems. UAV-based Variable Rate Technology integrates positioning, payload sensing, wireless communication, flight control, commercial platforms, and crop-specific models. Real-Time Kinematic (RTK) and Post-Processed Kinematic (PPK) receivers provide centimeter-level flight and hovering precision, while Inertial Measurement Units (IMUs) stabilize and correct flight trajectories in real time (Guebsi et al., 2024). LiDAR-based Simultaneous Localization and Mapping (SLAM) supports navigation in GNSS-denied environments and dense crop canopies (Ren et al., 2024), while TimeSync provides microsecond synchronization of optical and GNSS observations and Pure Pursuit with pose correction and active disturbance rejection improves path tracking under wind (Mowla et al., 2026; Ren et al., 2024). RGB cameras assess plant morphology and row structure; multispectral sensors provide NDVI, NDRE, and GNDVI; hyperspectral sensors cover SWIR (800–2,500 nm) for crop composition, nitrogen, and early water stress; thermal infrared sensors map canopy temperature (T_c) and CWSI; and LiDAR provides 3D crop structure (Chouhan et al., 2025; Mowla et al., 2026; Chouhan et al., 2025; Sow et al., 2024; Chouhan et al., 2025; Mowla et al., 2026; Agelli et al., 2024; Chouhan et al., 2025). Communication uses Wi-Fi, LoRa, ZigBee, Bluetooth, 4G/5G, and Radio Frequency (RF) protocols (Faqeih et al., 2025); 5G Private Mobile Networks support low-latency UAV-to-Multi-access Edge Computing (MEC) video transmission for YOLOv8 inference (Nguyen-Tan and Le-Trung, 2024); MIMO LoRa improves large-area communication (Almalki and Angelides, 2024); and UAV-assisted LoRaWAN gateways collect root-zone telemetry from Wireless Underground Sensor Networks (WUSNs) (Abdelmoneim et al., 2025; Ogbodo et al., 2026).

Autonomous flight uses Deep Reinforcement Learning algorithms including DQN, DDQN, and Soft Actor-Critic (SAC) for path planning, collision avoidance, and trajectory tracking (Chouhan et al., 2025), Ant Colony Optimization for fleet synchronization and scheduling (Almalki and Angelides, 2024), and MAVLink/FANET for waypoint navigation and communication (Krishnan et al., 2023). Commercial platforms include DJI Phantom, Mavic, Matrice, senseFly eBee/eBee X, DeltaQuad Pro, DJI Agras, Parrot Bluegrass, Skydio 2+, and Percepto Sparrow (Mowla et al., 2026; Mowla et al., 2026; Ren et al., 2024; Chouhan et al., 2025). Multispectral and thermal imaging support soil dryness, soil moisture, ETa, Kc, water-stress hotspots, and georeferenced targets for ground systems (Wanyama et al., 2024; Mowla et al., 2026), while UAVs also collect water samples and support soil-moisture sensor placement (Chouhan et al., 2025; Almalki and Angelides, 2024) (Table 3).

Table 3. Field robotic platforms for agricultural water management.

AI, Artificial Intelligence; APH, Advanced Plant Habitat; CWT, Continuous Wavelet Transform; CNN, Convolutional Neural Network; CV, Computer Vision; DL, Deep Learning; DQN, Deep Q-Network; EC, Electrical Conductivity; eVTOL, electric Vertical Take-Off and Landing; EP, Electrophysiological; FFLM, Fuzzy Logic and Learning Machine; GIS, Geographic Information System; GNSS, Global Navigation Satellite System; GPS, Global Positioning System; GPR, Ground Penetrating Radar; HSI, Hyperspectral Imaging; IoT, Internet of Things; IP, Internet Protocol; IWUE, Irrigation Water Use Efficiency; K-means, K-means Clustering; LiDAR, Light Detection and Ranging; ML, Machine Learning; NDVI, Normalized Difference Vegetation Index; NIR, Near-Infrared; PRI, Photochemical Reflectance Index; RF, Random Forest; RL, Reinforcement Learning; ROS, Robot Operating System; RTK, Real-Time Kinematic; RS, Remote Sensing; SAC, Soft Actor-Critic; SPAW, Soil–Plant–Atmosphere–Water; TIR, Thermal Infrared; UGV, Unmanned Ground Vehicle; UAV, Unmanned Aerial Vehicle; VRI, Variable Rate Irrigation; WSN, Wireless Sensor Network; XAI, Explainable Artificial Intelligence; XGBoost, Extreme Gradient Boosting; YL-69, Soil Moisture Sensor Model; 2D, Two-Dimensional; 3D, Three-Dimensional; 1D, One-Dimensional; R², Coefficient of Determination.

Crop-specific applications include cotton leaf-water prediction using RGB, multispectral, 19 vegetation indices, and thermal imagery with XGBoost (R² = 0.860) (Mowla et al., 2026), cotton seedling detection using CenterNet–DEEP SORT (R² = 0.96, RMSE = 0.39) (Gupta et al., 2026), sugarcane soil-salinity mapping using Random Forest (R² = 0.90, RMSE = 0.59 dS m⁻¹) (Mowla et al., 2026), wheat irrigation using AquaCrop-IoT with approximately 32% water savings (Morchid et al., 2026a), wheat soil-salinity prediction using Transformer models (R² = 0.84, RMSE = 2.41 g kg⁻¹, MAE = 1.84 g kg⁻¹), tomato water-stress classification using Inception-ResNet-v2 with 83%, 90%, and 98% accuracy for AWC classes of 25%, 50%, and 100% (Mowla et al., 2026), vineyard water-status prediction using Random Forest (R² = 0.92) (Gupta et al., 2026), banana water savings of 31.4% with 5,010.7 mL per tree versus 7,541 mL (Morchid et al., 2026a), peach orchard transpiration estimation using TSEB (R² = 0.89) (Mowla et al., 2026), and soybean grain-yield and leaf-moisture estimation using DNNs (Gupta et al., 2026). Aerial robotics uses Unmanned Aerial Vehicles equipped with Red–Green–Blue, multispectral, hyperspectral, thermal infrared, and Light Detection and Ranging sensors, supported by Real-Time Kinematic, Post-Processed Kinematic, and Inertial Measurement Unit systems. These platforms support crop-water stress detection, evapotranspiration assessment, field mapping, and identification of irrigation intervention zones. Integration with Artificial Intelligence, Machine Learning, Internet of Things, and Unmanned Ground Vehicles enables targeted water management.

On-ground robotics combines UGV platforms, mobile soil and crop sensing, embedded controllers, wireless communication, and automated irrigation for localized monitoring and site-specific water management. The AgriOne UGV uses a VWC sensor on a linear-actuator manipulator with surface-aware filtering to remove invalid measurements when soil penetration fails (Tsimpidi et al., 2025). The Vineyard Non-Invasive AGV integrates dual-frequency GPS geodetic, Local Tangent Plane (LTP), and canopy coordinates with canopy temperature, NDVI, and PRI, validated against manual Ψleaf measurements for vineyard water-stress mapping (Saiz-Rubio et al., 2021). A ZigBee irrigation robot transmits temperature and soil-moisture data to an LPC2148 controller, while an Arduino–Raspberry Pi KNN rover processes soil-moisture and temperature data for automatic localized pump control (Talaviya et al., 2020). The Photovoltaic Multi-Purpose Smart Farming Robot integrates photovoltaic panels, batteries, soil and plant sensors, and computer vision for plant-level precision irrigation and selective pesticide application (Nurmalitasari et al., 2025). On-ground robotics uses Unmanned Ground Vehicles and Automated Guided Vehicles with mobile soil and crop sensors, embedded controllers, wireless communication, and automated irrigation systems. These platforms support soil-water assessment, plant-water monitoring, leak detection, multi-depth sensing, and localized irrigation intervention. Integration of computer vision, hyperspectral sensing, and machine learning enables field-level and site-specific water management. A moving-bridge manipulator performs closed-loop variable irrigation, a drip-pipeline leak-detection rover uses visible and thermal infrared cameras, and a weather-adaptive LSTM watering robot adjusts irrigation intervals using soil temperature, humidity, and weather forecasts (Parra-López et al., 2025). An on-the-go hyperspectral UGV uses a robotic arm for direct-insertion, multi-depth soil-moisture classification, while an optical soil-image irrigation robot uses smartphone RGB imaging, grayscale wetness estimation, GSM transmission, and a central microcontroller, achieving 58.57% water savings versus flood irrigation and 86.97% versus conventional drip irrigation (Zhang et al., 2025; Roy and Chakraborty, 2025). Different field robotic platforms, i.e., VineScout, BoniRob (Flourish), TerraSentia, SwarmFarm, BerryBot, Niryo Robot etc., for field water management were shown in Table 4, respectively. It summarizes commercially available sensing and robotic technologies used for soil-moisture, irrigation, water-quality, and nutrient monitoring in smart agriculture. These technologies provide field-level sensing, automated monitoring, and targeted management, supporting precision irrigation and improved water and input-use efficiency.

Table 4. Commercially available sensing and robotic technologies for soil moisture, irrigation, water quality, and nutrient monitoring.

10 Recommendation models for smart water management in agriculture

Recommendation models for smart water management combine soil, plant, weather, remote-sensing, and irrigation-system observations with AI/ML, IoT, VRT, digital twins, DSS, and robotics to support real-time, site-specific, and adaptive irrigation across heterogeneous fields. AI models including LSTM, CNN-LSTM-Attention, Random Forest, MLP-ANN, 1D-CNN, and CNN-ViT have been applied across wheat, rice, maize, sugarcane, olive, camelina, soybean, tomato, and chili for irrigation scheduling, soil-moisture forecasting, evapotranspiration estimation, water-stress detection, and water-demand optimization, with high prediction accuracy and water savings as summarized in Table 1. A linked AI–IoT–robotics system can integrate root-zone capacitive, FDR, and TDR sensors, tensiometers, plant-wearable EP sensors, and multi-depth probes with ESP32 edge controllers to assess VWC, soil-water potential, and plant water status; EP signals within 1–5 min can trigger CNN-based valve control, while soil-moisture feedback can reduce water use by 20–55%, with sweet corn showing 12.05% higher grain yield and 12.7% lower water use (Abdulrasaq et al., 2026; Chen et al., 2025; Sharma V. et al., 2025). Multi-depth sensing at 15, 35, and 50 cm with flow and pressure monitoring can prevent deep percolation and maintain approximately 2.8 m³ h⁻¹ discharge (Meriç, 2025). UAV RGB, multispectral, hyperspectral, and thermal imagery, spectral indices for plant, i.e., CWSI, NDMI, NDWI, red-edge indices, and plant physiological indicators, can detect water stress before visible symptoms; TIR–RGB integration can detect stress up to 84 h earlier than conventional soil sensors (Fevgas et al., 2025; Sharma V. et al., 2025). Predictive irrigation scheduling can integrate Random Forest, Artificial Neural Networks, CNN–LSTM, Gradient-Boosting Decision Trees, and Soft Actor–Critic reinforcement learning with soil, crop, weather, and remote-sensing data to estimate crop water demand and optimize irrigation; Random Forest achieved R² = 0.93 for maize and R² = 0.91 for wheat, while Soft Actor–Critic adapts soil-water depletion under constrained water allocation (Nasar et al., 2026; Xu et al., 2026). Dung Beetle Optimization can optimize Random Forest parameters for water-demand prediction, while Artificial Neural Network edge inference enables real-time valve control (Hu et al., 2025; Bautista et al., 2026). A Cloud–Edge–End framework can use LoRaWAN, 4G, and 5G for telemetry, ESP32 nodes for local decisions during connectivity interruptions, and cloud platforms for model training, historical analysis, and dashboards; ThingsBoard-based systems and LoRaWAN have supported up to 70% water saving and communication ranges of up to 3 km, respectively (Wang et al., 2025; Ali et al., 2026; Morchid et al., 2024; Aghaabbasi and Ganjouei, 2026).

Spatially variable irrigation can combine ECₐ-based management zones, GIS, UAV water-stress maps, and VRT controllers for site-specific prescriptions; ECₐ kriging achieved RMSE = 0.097702, while a retrofittable Variable Rate Applicator Controller achieved approximately 97% material-discharge accuracy, with fuzzy-logic and neural-network controllers enabling center-pivot speed adjustment (Millán et al., 2025; Abdalla and Mirzakhani Nafchi, 2024; Morchid et al., 2025a). UAVs can identify water-stress hotspots and generate georeferenced prescription maps for UGV-based localized sensing and intervention. UGV manipulators support direct multi-depth soil sampling with rejection of invalid readings, while UAV multispectral and thermal observations provide complementary canopy-level water-status information (Tsimpidi et al., 2025; Mowla et al., 2026). Digital twins and DSS can integrate real-time soil and plant sensors, weather forecasts, satellite-derived variables, crop-growth models, and irrigation hydraulics to update soil-water balance and automate RDI and VRI; Irri_DesK reduced irrigation water use by 26% and monitoring labour by 80%, while the SAFY–FEST-EWB model achieved >10% water savings and >13% higher water productivity (Millán et al., 2025; Sarr et al., 2026). Crop-specific management should link root-zone and canopy sensing with crop growth stage, soil-water availability, ET₀, VPD, forecast rainfall, and plant water stress, using AWD or RDI for cereals, multi-depth and plant-water sensing with UAV and digital twins for orchards and vineyards, and real-time soil, plant, and climate sensing with AI-based scheduling for vegetables. Crop specific precision water management and field performance are given in Figure 3. Across cropping systems, these linked inputs can support closed-loop management of water use, crop yield, energy use, fertilizer efficiency, and crop quality (Ahmad and Sohel, 2025; Ke et al., 2026; Sharma V. et al., 2025). Smart water management integrates multi-depth soil and plant-based sensing, weather observations, imaging, and flow-pressure monitoring rather than relying on individual parameters. Sensor and remote-sensing data with learning, and optimization models for estimation of crop water demand, evapotranspiration, and plant water stress synergistically work with Internet of Things and edge–cloud systems which provide continuous data transfer and rapid local control, with automated valves enabling closed-loop irrigation. Variable rate application uses management zones, prescription maps, and real-time observations for site-specific application. Digital twins and decision-support systems used to integrate soil-water balance, weather forecasts, crop growth, and irrigation states for adaptive scheduling with robotics complement fixed sensors through field-scale mapping and localized sensing or intervention under variable field conditions to improve water-use efficiency without compromising crop productivity.

11 Challenges and future perspectives

Despite significant advances in water management in precision irrigation systems, it continues constrained by multiple technical, operational, and adoption challenges. The discussed smart water management technologies face limitations related to sensor reliability, connectivity, sensing coverage, system cost, interoperability, model generalization, and automation as discussed in Table 5, respectively. In addition, limited, noisy, incomplete, and heterogeneous datasets constrain the development and validation of AI-based water-management models, highlighting the need for improved sensing, data integration, standardized datasets, and adaptive modelling approaches. Point-level soil sensors often fail to represent field-level soil-water heterogeneity, while long-term deployment is affected by calibration drift and environmental noise, reducing measurement reliability (Anjum et al., 2023; Oğuztürk, 2025). Remote sensing is limited by cloud cover, atmospheric interference, and UAV constraints such as short battery endurance and low flight speed, restricting continuous large-area monitoring (Singh and Singh, 2026; Ahmad and Sohel, 2025). AI-based irrigation models frequently remain crop- and site-specific, requiring retraining or transfer learning through different environments, while static crop-coefficient curves inadequately represent crop development under non-stationary climatic conditions (Carlo Tancredi et al., 2025; Sarr et al., 2026). Deficit irrigation without appropriate drainage modelling may accelerate soil salinization and biodiversity loss, but alternate wetting and drying can reduce methane emissions but simultaneously increase nitrous oxide emissions (Srivastav et al., 2026; Ramzan et al., 2026). Precision irrigation in remote regions is further constrained by intermittent wireless connectivity, communication dropout resulting in more than 8% missing sensor observations, gateway-storage limitations, weak cryptographic capability of low-power IoT devices, vulnerability to wireless attacks and false-data injection, and the absence of standardized preprocessing pipelines for real-time anomaly detection (Abdelmoneim et al., 2025; Wang et al., 2026; Jaiswal et al., 2025; Cariou et al., 2023; Lyu et al., 2026). Future implementation should therefore link financial incentives, technical capacity building, harmonized data governance, cooperative digital infrastructure, and adaptive water policies. Governments can reduce investment barriers through capital subsidies, tax incentives, and low-interest loans, while agricultural organizations can strengthen digital extension services and participatory training programmes (Dutta et al., 2026). Future research should prioritize biodegradable IoT sensors, TinyML-enabled edge intelligence, standardized open-access data, federated learning, Explainable Artificial Intelligence (XAI), and quantum computing to enhance sustainability, reproducibility, computational efficiency, privacy preservation, model transparency, and large-level environmental sensing (Joseph et al., 2026; Bautista et al., 2026; Jaiswal et al., 2025; Ogbodo et al., 2026; Mowla et al., 2026; Morchid et al., 2026b).

Table 5. Current challenges, data limitations, and future directions for smart water management technologies.

12 Conclusion

The review shows a connected pathway for smart agricultural water management, linking crop, soil, and weather conditions with crop water requirement estimation, water-status monitoring, prediction, decision-making, and irrigation application. Crop characteristics, growth stage, rooting depth, soil-water availability, evapotranspiration, and weather conditions provide the basis for determining crop water requirement. Soil and plant sensors provide real-time information on soil moisture, plant-water status, and physiological response, while unmanned aerial vehicles, satellite, and proximal imaging extend this assessment to crop water stress, evapotranspiration, and spatial field variability. The Internet of Things, wireless communication, and edge–cloud systems connect these observations for continuous monitoring and data integration. Artificial intelligence, machine learning, and deep learning models support soil-moisture forecasting, evapotranspiration estimation, crop water-demand prediction, water-stress detection, and irrigation scheduling. Digital twins and decision-support systems link field observations and model outputs with irrigation decisions, while map-based, sensor-based, and real-time variable-rate technologies support site-specific water application according to spatial and temporal field variation. The review also shows a progression from water-status assessment to automated field management. Unmanned aerial vehicles and unmanned ground or autonomous ground vehicles support localized sensing, stress-zone identification, leak detection, and targeted intervention, while automated pumps, valves, nozzles, and irrigation systems provide controlled water application. Integration of these components enables a feedback pathway in which post-irrigation soil, plant, evapotranspiration, and crop responses can be used to update irrigation prescriptions and support adaptive management. A key scientific insight is that the effectiveness of smart water management depends on the integration of complementary sensing, modelling, decision-support, and control technologies rather than on a single technology alone.

The review also identifies important challenges and research gaps related to sensor drift and maintenance, limited spatial coverage, connectivity and power constraints, interoperability, high system cost, limitations in unmanned aerial vehicle operation, model generalization, explainability, scalability, and the complexity of integrated systems. Limited data availability, missing and noisy observations, sensor calibration issues, limited ground-truth information, differences in spatial and temporal resolution, and insufficient crop-, site-, and season-specific datasets remain important constraints for the development, training, validation, and application of artificial intelligence, machine learning, and deep learning models. These limitations can affect soil-moisture estimation, evapotranspiration prediction, crop water-demand assessment, water-stress detection, irrigation scheduling, and site-specific water application. Future research should therefore focus on improved sensor calibration and data fusion, integration of ground, unmanned aerial vehicle, and satellite observations, standardized and open agricultural water datasets, stronger multi-crop and multi-season validation, and models with improved transferability and explainability. Further development of edge computing, low-resource machine learning, federated learning, physics-informed approaches, digital twins, interoperable systems, autonomous robotics, and closed-loop irrigation can strengthen real-time and adaptive water management. Overall, these developments provide a pathway toward more reliable crop-specific and site-specific irrigation decisions and support the transition to climate-resilient agricultural water management.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Generative AI tools were used solely for grammar correction and language refinement.

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: Ajay Singh, Texas A&M AgriLife Temple Research and Extension Center, United States

Reviewed by: Muhammad Umer Arshad, University of Illinois at Urbana-Champaign, United States

Hajar Hamdaoui, Mohamed Premier University, Morocco