AI-Powered Drone Technology for Water Management and Plant Health Prediction

This project uses AI powered drone imagery to optimise irrigation and predict plant health at field scale. By detecting water stress and waterlogging early, the solution reduces water waste, improves crop resilience and replaces labour intensive monitoring with automated, data driven insights, supporting climate resilient land management.

Introduction

Climate change, erratic weather patterns and increasing pressure on freshwater resources are reshaping modern agriculture. Farmers must produce more with fewer inputs, making efficient water use and early detection of crop stress critical for sustainability. However, traditional monitoring methods are often reactive, labour-intensive and unable to capture the spatial variability that exists across large fields.

This success story demonstrates how drone-based multispectral imaging and artificial intelligence enable precise water management and reliable plant health prediction. By converting high-resolution aerial data into actionable insights, the solution allows irrigation to be applied exactly where it is needed, identifies water stress and waterlogging at an early stage, and supports timely interventions. The result is reduced water waste, healthier crops and a practical pathway toward climate-resilient, data-driven farming.

Understanding the Challenge

All plant-centred initiatives across agriculture, horticulture and forestry face two closely linked challenges: inefficient water use and limited visibility into plant health. Over-irrigation leads to water waste and higher operating costs, while under-irrigation weakens crops and reduces yields. Traditional monitoring methods such as visual field inspections or manual sampling are labour-intensive, time-consuming and often fail to detect early signs of stress or disease.

These limitations result in delayed interventions, rising costs and increased vulnerability to climate variability. Improving water efficiency and enabling timely, accurate plant health monitoring are therefore essential for sustainable and climate-resilient farming, and they form a critical foundation for scaling AI-driven weather prediction and decision-support systems.

Benefits of Our Solution

Our approach delivered three clear benefits for growers:

1. 1. Optimised water usage. By targeting irrigation precisely, we achieved significant water savings that reduce operating costs and support sustainable farming practices.
   2. Enhanced plant health. Early detection of water stress and other health issues allowed timely interventions, resulting in more resilient crops and higher yields.
   3. Increased efficiency. Automated monitoring and analysis freed farmers from labour‑intensive field inspections, enabling them to focus on other critical tasks.

These kinds of AI-driven, water-efficient solutions are increasingly being adopted across the global agri-food ecosystem, as highlighted by leading companies and organizations advancing sustainable agriculture that are combining data, automation and environmental intelligence to improve productivity while reducing resource use.

Our Innovative Solution

To tackle the twin challenges of water management and plant health, our team developed three complementary models based on multispectral drone data. These models created a high‑precision irrigation standard, detected and prevented water stress or waterlogging, and grouped plants by health status so that interventions could be targeted effectively.

1. Creating a New Precision Standard in Irrigation

Our Soil Moisture Index (SMI) estimation model uses the triangle method to measure soil moisture levels accurately. By applying water precisely when and where it is needed, farmers minimise water waste and ensure that plants receive optimal hydration. The result is healthier crops and increased yields without additional input costs.

2. Detecting and Preventing Water Stress and Waterlogging

The second model uses threshold‑based analysis to detect areas where plants are receiving either too little water (water stress) or too much water (waterlogging). Early identification of these conditions allows growers to adjust irrigation schedules or improve drainage systems before crops are damaged. This proactive approach preserves plant health and prevents yield loss.

3. Advanced Health Monitoring with K‑Means Clustering

Our third model groups plants based on their health status using a K‑means clustering algorithm. The resulting clusters provide a clear, detailed picture of the field’s overall condition. By monitoring these clusters, farmers can rapidly identify issues such as disease or nutrient deficiency and apply corrective measures exactly where they are needed.

The Significance of Plant Protection

Fig.1 Disease in chili plant.

Plants produce roughly 80 % of the food we eat and 98 % of the oxygen we breathe. Their well‑being underpins biodiversity, ecosystem services and our ability to feed a growing global population. Changing climate patterns, rising temperatures and threats such as locust infestations highlight the importance of preserving plant health. Protecting vegetation is not optional; it is essential for environmental sustainability and food security.

Common Approaches to Monitoring Plant Health

Traditional monitoring methods form the baseline against which we measured our drone‑based solution. They include:

1. 1. • Visual inspections. Farmers and agronomists walk through fields to assess plant health. This method is labour‑intensive, subjective and can miss early signs of stress or disease.
      • Soil moisture sensors. Ground‑based sensors provide accurate moisture readings at specific points but do not capture the full variability of conditions across a field.
      • Satellite imagery. Satellites offer large‑scale images but have limited resolution and can be obstructed by cloud cover.
      • Manual sampling. Laboratory analysis of soil and plant samples yields detailed information but is time‑consuming and only represents small areas.

Understanding the limitations of these methods emphasises why a more comprehensive, data‑driven approach is necessary.

Our Goal and Approach

We set out to build an automated plant health prediction and monitoring solution using drone‑derived multispectral and thermal data. To achieve this, we followed a systematic approach consisting of five major steps. A brief summary of each step is presented below.

Step 1 – Identifying the Right Test Environment

Before choosing a testing site, we defined the conditions needed for robust model development:

1. 1. • Variability. The environment had to exhibit diverse conditions, including different levels of moisture, salinity and temperature, so that the model would perform under multiple scenarios.
      • Existing sensors. A location with installed sensors for soil moisture and other parameters would allow us to cross‑reference and validate drone data.
      • Accessibility. The site needed to be accessible for ground‑truth data collection, including manual soil moisture measurements and plant health indicators.

Step 2 – Choosing a Case Study: The Golf Course

While a farm field might seem the obvious choice, we selected a golf course as our case study for several compelling reasons. Golf courses are controlled environments with consistent maintenance practices, making it easier to isolate and study specific variables without interference from unpredictable agricultural activities. They also exhibit wide variation in moisture and salinity because watering practices differ across fairways, greens and roughs, allowing us to simulate different stress conditions. Many courses already have sensor infrastructure for monitoring soil moisture, facilitating effective cross‑validation with drone data. Seasonal changes and maintenance schedules provide dynamic conditions for testing the robustness of our models, and the structured layout simplifies ground‑truth data collection.

Step 3 – Validating the SMI Model with Field Data

Validation was essential to ensure that the Soil Moisture Index predictions were accurate and trustworthy. After reviewing the normalised difference vegetation index (NDVI) and thermal data from the drones, we created scatterplots plotting temperature against NDVI values. These scatterplots helped identify “dry edges” (hot, dry areas) and “wet edges” (cool, wet areas). We downsampled the data to remove noise and speed up processing, then generated an SMI map that visually shows areas of water stress and waterlogging.

To confirm accuracy, we compared the SMI predictions with actual soil moisture data collected manually. Scatterplots of predicted SMI values against measured soil moisture readings allowed us to identify discrepancies and adjust the model. This iterative process improved predictive accuracy and built confidence in the system.

Figure 2: Temperature and soil moisture field data points and NDVI, Thermal and SMI prediction of the greens on Hole

Step 4 – Identifying Problem Areas with a Threshold‑Based Model

The fourth step focused on turning predictive insights into actionable interventions. We defined thresholds – specific values used to classify areas as healthy, water‑stressed or waterlogged – based on the data. The image data was processed through several stages:

Downsampling to reduce resolution and accelerate processing; rescaling to standardise values; masking to concentrate on areas of interest; polygonisation to convert image data into vector format for detailed analysis; and finally thresholding to mark problem areas clearly.

This process produced a map that accurately identified zones of water stress and waterlogging. By highlighting these areas, the model allowed farmers to focus their interventions where they would have the greatest impact, improving plant health and optimising water use.

Fig.3 Waterlogged regions on Fairway 2 from 16-06-2021. Left: RGB image, Right: Identified waterlogged regions (Source: Omdena)

Step 5 – Grouping Plant Health Status with K‑Means Clustering

Clustering techniques were used to group areas of the golf course based on plant health. Data processing involved masking to isolate relevant zones, filtering to remove outliers, scaling to normalise values and applying the K‑means algorithm to categorise the landscape into distinct clusters. Post‑processing removed small, insignificant clusters and converted the results into vector format for easy visualisation.

The resulting map offered a clear representation of plant health across the golf course, with different clusters corresponding to varying levels of vigour and stress. Course managers could quickly identify problem areas and direct resources where they were needed most, thereby improving efficiency and resource management. 

Similar approaches are widely used in precision agriculture workflows, where image segmentation techniques for weed or crop detection help isolate vegetation patterns, stress zones and crop variability at a highly granular level.

Why Use Drones?

Our solution depends on drones equipped with advanced multispectral and thermal sensors that capture data across the blue, green, red, red‑edge and near‑infrared bands. These bands reveal subtle differences in plant health and soil conditions that are invisible to the naked eye. Visible bands (blue, green and red) provide general information on plant health, the red‑edge band highlights changes in chlorophyll content (an indicator of stress), and the near‑infrared band reflects healthy vegetation strongly, allowing us to assess biomass and vigour. Thermal sensors measure the surface temperature of plants and soil, helping to identify water stress and waterlogging. This high‑resolution data enables AI weather prediction tools to monitor conditions across large fields quickly and accurately.

Figure 4: Drone Gathering General Information On Plant Health

Compared with traditional methods, drones offer several advantages. They cover large areas rapidly and objectively, overcoming the subjectivity and limited scope of visual inspections. Unlike satellite imagery, drones are not hampered by cloud cover and provide higher spatial resolution. They also reduce reliance on labour‑intensive manual sampling, delivering detailed insights without the need to collect and analyse numerous physical samples. While ground sensors provide point measurements, drones offer comprehensive coverage, making them ideal for precision agriculture and weather forecasting for agriculture.

When combined with satellite-based monitoring, these drone insights become even more powerful. For example, using satellite imagery to detect and assess the damage of armyworms in farming demonstrates how multispectral data can support early pest detection and large-scale crop risk assessment alongside field-level analytics.

Applications Beyond Agriculture

Although our models were designed for agriculture, the combination of drones and AI has broad applicability across many industries. In mining and resource extraction, high‑resolution thermal and multispectral data enhance environmental monitoring and ensure regulatory compliance by detecting subtle changes invisible to traditional cameras. Construction and infrastructure projects benefit from improved site surveys and progress monitoring, allowing engineers to spot heat leaks, water intrusion and material degradation early. In utilities and energy, drones detect thermal hotspots and vegetation encroachment that could cause power outages or fires, and they monitor solar panels and wind turbines for efficiency losses.

For disaster management and relief, drones equipped with thermal imaging can locate survivors in debris, while multispectral data assess the extent of vegetation and structural damage. Fisheries and coastal management teams use these sensors to detect changes in water quality and coastal vegetation health, protecting marine biodiversity. In real estate and property management, drones identify stressed vegetation and heat loss, helping maintain landscapes and improve energy efficiency. Tourism and recreation facilities maintain pristine grounds by optimising irrigation based on high‑resolution plant health data. Manufacturing and industrial facilities employ drones to detect heat anomalies and equipment malfunctions early, while smart cities and urban planners use them to monitor urban heat islands, vegetation health and infrastructure condition. Together, these applications demonstrate the versatility of drone‑based AI solutions beyond the farm.

Future Possibilities

The success of the current models opens clear opportunities for further enhancement. Integrating real-time and forecasted weather data would enable dynamic predictions and improve accuracy in identifying water stress and waterlogging. Expanding the models to support different crops and vegetation types would extend their use across diverse agricultural systems.

Additional improvements include a mobile application for real-time insights and recommendations, the use of advanced machine-learning techniques to refine predictions, and the integration of satellite data to scale coverage. Connecting the models with automated irrigation systems and adding predictive maintenance and soil-health indicators would further optimise water use, reduce downtime and strengthen long-term farm sustainability.

Conclusion

By combining multispectral drone imagery, thermal data and advanced AI models, this project demonstrates that precise and scalable water management and plant health prediction are achievable in real-world settings. The solution enables targeted irrigation, early detection of stress and actionable, field-level insights that support more efficient and resilient farming practices. More broadly, it shows how data-driven decision-making and AI-enabled weather intelligence can help agriculture adapt to growing climate uncertainty.

Beyond farming, the same techniques can deliver value across sectors such as energy, construction, disaster management and urban planning. As the models evolve and incorporate additional data sources, their potential impact will continue to expand. Organisations seeking to improve resource efficiency, resilience and sustainability can build on these innovations to drive meaningful, long-term change.