Success in Water Management and Plant Health Prediction with Drone Technology – Our AI is Revolutionizing Agriculture

Author of an article: Weronika Dorocka – VP of Business Development

Introduction

In a world where climate change and resource scarcity threaten our way of life, innovative solutions are essential for sustainable development. This success story tells a story of a promise for the future that we all need, the future where resources are not wasted and the problems are smartly detected to act effectively and fast. We harnessed advanced drone technology to optimize water consumption and predict plant health in agricultural fields. 

Understanding the Initial Problem

All initiatives that center around plants, including but not limited to the agricultural sector, often struggle with two significant issues: inefficient water usage and the difficulty of accurately monitoring plant health.

Agriculture often faces two critical challenges: inefficient water use and difficulty in accurately monitoring plant health. Over-irrigation wastes water and increases costs, while under-irrigation harms crops. Traditional plant health monitoring is labor-intensive and prone to errors, missing early signs of disease and stress. These inefficiencies lead to higher costs and lower yields, threatening sustainability. Optimizing water use and enhancing plant health monitoring can save significant costs, improve crop yields, and contribute to sustainable farming practices.

The Benefits of Our Solution

Optimized Water Usage: By precisely targeting irrigation, we achieved significant water savings, contributing to sustainable farming practices and cost reductions.

Enhanced Plant Health: Early detection and mitigation of water stress and other health issues resulted in more resilient and productive crops.

Increased Efficiency: Automated monitoring and analysis reduced the labor and time required for field inspections, allowing farmers to focus on other critical tasks.

Our Innovative Solution

Recognizing these critical pain points, our team developed three pioneering models, utilizing advanced multispectral drone data to provide precise, actionable insights:

1. Creating a New Precision Standard in Irrigation Systems

Soil Moisture Index (SMI) Estimation using the Triangle Method: This model accurately measures soil moisture levels, allowing farmers to make informed irrigation decisions. By applying water precisely when and where it’s needed, we minimized water waste and ensured that plants received optimal hydration, leading to healthier crops and increased yields.

2. Detecting and Preventing Water Stress and Waterlogging (too much, too little water in specific parts)

Threshold-based Model for Identifying Water Stress and Waterlogging: This model sets specific thresholds to detect areas where plants are either receiving too little water (water stress) or too much water (waterlogging). Early identification of these conditions enables farmers to take corrective actions, such as adjusting irrigation schedules or improving drainage systems. This proactive approach prevents crop damage and maintains plant health.

3. Advanced Health Monitoring: Grouping Plants by Health Status with K-Means Clustering

K-Means Clustering Model for Health Monitoring: This model groups plants based on their health status, providing a clear, detailed picture of the field’s overall condition. By monitoring these clusters, farmers can quickly identify and address issues such as diseases or nutrient deficiencies in specific areas, ensuring targeted and effective interventions.

Discover how we did it!

The Background 

Plants make up 80% of the food we eat and produce 98% of the oxygen we breathe.

From feeding a growing world population to protecting biodiversity and ecosystems, the significance of plants is tremendous. Consequently, the protection of plants is paramount. In an era of changing climate, rising temperatures, and threats like hungry locusts, the preservation and health of plants have never been more important.

How Plant Health and Hydration Are Commonly Monitored

Traditionally, monitoring plant health and hydration has relied on several methods:

1. Visual Inspections: Farmers and agronomists walk through fields to visually assess plant health. This method is labor-intensive, time-consuming, and subjective, often missing early signs of stress or disease.
2. Soil Moisture Sensors: Ground-based sensors measure soil moisture levels at specific points. While accurate, these sensors cover limited areas and may not capture the full variability of field conditions.
3. Satellite Imagery: Satellites provide large-scale images of crop fields, but their resolution is often too low to detect small-scale issues. Additionally, cloud cover can obstruct satellite imagery, limiting its effectiveness.
4. Manual Sampling: Collecting soil and plant samples for laboratory analysis provides detailed information but is labor-intensive and time-consuming. This method also only represents specific sampled locations and may not reflect broader field conditions.

The Goal 

We started with an aim to build an automated plant health prediction and monitoring solution using drone-derived multispectral and thermal data.

Our Approach

Preparation Steps: 

Step 1. 

Identifying the Need for a Test Environment and Its Requirements

Before choosing the right testing environment, we identified the essential components required for our project:

1. Variation in Environmental Conditions: A test environment with diverse conditions, including varying levels of moisture, salinity, and temperature, was crucial. This variation would help us develop and validate our models under different scenarios.
2. Existing Sensor Infrastructure: Having a location with pre-installed sensors to monitor soil moisture and other relevant parameters would facilitate cross-referencing and validating our drone data.
3. Accessibility for Field Data Collection: The ability to easily collect ground-truth data, such as soil moisture levels and plant health indicators, was necessary for accurate model training and validation.

Step 2. 

The Choice of a Case Study Environment: A Golf Course

While a farm field might seem like an obvious choice, we selected a golf course for several compelling reasons:

1. Controlled Environment: A golf course provides a controlled environment with consistent maintenance practices. This consistency made it easier to isolate and study specific variables without the interference that might come from varying agricultural activities on a farm.
2. High Variation in Conditions: Golf courses have significant variation in moisture and salinity levels due to different watering practices for fairways, greens, and roughs. This variation is ideal for testing our models, as it simulates different stress conditions that crops might experience.
3. Existing Infrastructure: Many golf courses already have advanced sensor infrastructure in place to monitor soil moisture and other conditions. This existing infrastructure allowed us to cross-reference and validate our drone data effectively, providing a reliable and efficient testing environment.
4. Dynamic Conditions: Golf courses experience diverse conditions due to seasonal changes, weather patterns, and resource provision (water, nutrients). This diversity made them perfect for testing our approach under varying scenarios and ensuring the robustness of our models.
5. Ease of Data Collection: The structured layout of a golf course and the existing maintenance routines facilitated easier and more consistent collection of ground-truth data. This was critical for validating our model predictions accurately.
   

Knowing All That,
Why using Drones in our Solution?

The drones we decided to use are equipped with advanced multispectral and thermal sensors that capture a range of data across different spectral bands. In this context, “bands” refer to specific ranges of wavelengths in the electromagnetic spectrum that the sensors can detect.  These specialized drones, can detect subtle differences in plant health and soil conditions from a distance. The multispectral sensors capture light in the blue, green, red, red edge, and near-infrared bands, providing detailed information about plant vigor and stress.

Here we also decided to give a bit of comparison between other common techniques and our choice:

Visual Inspections vs. Drones:

• Visual Inspections: Labor-intensive, subjective, and can miss early signs of issues.
• Drones: Provide high-resolution, objective data that covers large areas quickly and can detect early signs of stress or disease that might be missed during visual inspections.

Satellite Imagery vs. Drones:

• Satellite Imagery: Covers large areas but with lower resolution and potential issues with cloud cover.
• Drones: Provide high-resolution, real-time imagery unaffected by cloud cover, allowing for detailed analysis of specific areas.

Manual Sampling vs. Drones:

• Manual Sampling: Provides detailed information but is labor-intensive and time-consuming, covering only specific points.
• Drones: Can quickly cover large areas and provide detailed data on soil and plant health, reducing the need for extensive manual sampling.

Why Use Drones Instead of Ground Sensors you may think? 

1. More Comprehensive Coverage: Drones can cover large areas quickly and efficiently, providing comprehensive data across the entire golf course. Ground sensors, while useful, are limited to specific locations and may miss variations in other parts of the field.
2. High-Resolution Data: Drones equipped with multispectral and thermal sensors capture high-resolution images and data that ground sensors cannot match. This detailed data allows for more precise analysis and better identification of problem areas.
3. Flexibility and Accessibility: Drones can easily access areas that might be difficult or impractical to reach with ground sensors. This ensures that no part of the golf course is left unmonitored.
4. Real-Time Monitoring: Drones can provide real-time data and imagery, allowing for immediate analysis and decision-making. Ground sensors typically require data retrieval and processing, which can delay response times.
5. More Cost-Effective for Scaling: Deploying drones is often more cost-effective than installing and maintaining a large number of ground sensors across a field. Drones can be easily deployed as needed without the extensive setup and maintenance required for ground-based systems.

Our AI Approach 

Step 1: Understanding the Dataset and Problem Analysis

What We Did:  The dataset provided to us consisted of multispectral drone imagery of a 7 month time frame, spanning different seasons and weather conditions. The seven bands were captured using Altum sensors from MicaSense, which include Blue, Green, Red, Red Edge, Near-infrared, Thermal, Transparency. Additionally, field scouts of soil moisture survey data were provided. We began by thoroughly understanding the dataset and the problem at hand. We also had a detailed discussion with a domain expert,, to understand specific needs and expectations of the industry.

Step 2: Setting Up a Framework for Moisture Measurement – Soil Moisture Index (SMI) Estimation

Important Aspects:

What We Did: We integrated thermal data with the Normalized Difference Vegetation Index (NDVI) to identify areas of water stress (where plants are too dry) and waterlogging (where plants have too much water). The SMI was calculated using the Red, Near-Infrared, and Thermal bands from the drone data.

Process:

1. Data Understanding: We started by reviewing the NDVI and thermal data from the drones to understand the current conditions.
2. Scatterplots: We created scatterplots (graphs) plotting temperature against NDVI values to identify patterns. This helped us find “dry edges” (hot, dry areas) and “wet edges” (cool, wet areas).
3. Downsampling: We reduced the size of the data to remove noise and make processing faster and more efficient.
4. Output: We generated an SMI map that visually shows areas of water stress and waterlogging.

Step 3: Ensuring Accuracy through Validation – Field Data Comparison and Accuracy Scatterplots

What We Did: Validation is a crucial step to ensure the accuracy and reliability of our Soil Moisture Index (SMI) predictions. It involves comparing the SMI predictions with actual field data and fine-tuning the model based on these comparisons.

Why It Was Needed: Validation is needed to confirm that our model’s predictions are accurate and trustworthy. By comparing the SMI values generated by the model with actual soil moisture readings, we can identify any discrepancies and make necessary adjustments to improve the model’s performance.

Validation Process:

Step 4: Identifying Problem Areas to Target Interventions – Threshold-Based Model

Important Aspects:

What We Did: We developed a model that sets specific thresholds to detect areas with water stress and waterlogging.

Results:

Step 5: Grouping Plant Health Status for Effective Monitoring – Clustering Model

Important aspects:

What We Did: We used clustering algorithms to group plants based on their health status.

Results:

*To explore even more technical aspects of this project check out also this article: Data-Centric AI for a Sustainable Water Irrigation System.

Benefits and other Applications of these Methodologies

Applied Methodology 1:

AI model utilizing drones equipped with advanced multispectral and thermal sensors

Can also be utilised in:

Mining and Resource Extraction

Construction and Infrastructure

Utilities and Energy

Disaster Management and Relief

Fisheries and Coastal Management

Real Estate and Property Management

Tourism and Recreation

Manufacturing and Industrial Facilities

Smart Cities and Urban Planning

Further Possibilities

The success of our current model opens up numerous avenues for further enhancements and applications. Potential next steps include:

These steps would significantly boost the effectiveness and reach of our solution, driving sustainable agricultural practices on a global scale.