AquaCrop - Optimizing Irrigation System using Aerospace Data

Inspiration

Our inspiration stems from a critical global challenge: ensuring food security in the face of increasing water scarcity, climate change, and the economic pressures on farmers. In agricultural powerhouses like Indonesia, millions of smallholder farmers are on the front lines, battling unpredictable rainfall, prolonged droughts, and decaying irrigation infrastructure. While precision agriculture offers a solution, existing technologies rely on expensive, high-maintenance, and logistically complex physical IoT sensors. This hardware-centric approach is simply not scalable for the vast majority of farmers who need it most.

We were inspired by a strategic choice: what if we could deliver the same, or even better, field-level intelligence without a single piece of hardware in the ground? The inspiration for AquaCrop is to replace the cost and complexity of physical IoT with the power of aerospace data and sophisticated AI, creating a "virtual sensor" that is infinitely scalable, affordable, and accessible to every farmer with a smartphone.

What it does

AquaCrop is a Software-as-a-Service (SaaS) platform that delivers precise, actionable irrigation schedules directly to farmers' mobile devices. It eliminates guesswork by telling a plantation manager or smallholder farmer exactly when and how much to water each specific field, every single day.

At its core, AquaCrop functions as a "virtual soil moisture probe" and a "virtual crop sensor." It uses a powerful AI model to fuse publicly available satellite data with weather information, generating a high-resolution (10-meter) map of soil moisture and real-time crop water demand. The platform translates this complex data into simple, justifiable recommendations, such as:

"Recommendation for Field 12: Irrigate for 60 minutes. Reason: Our virtual sensor shows soil moisture has dropped to the 45% depletion threshold, and the forecast calls for high water loss tomorrow."

By providing this intelligence without requiring any on-farm hardware, AquaCrop makes precision irrigation accessible, affordable, and immediately valuable, helping farmers save water, reduce energy costs, and secure higher, more resilient yields.

How we built it

We built AquaCrop on a three-layer architecture designed for massive scalability and scientific accuracy, with Google Earth Engine (GEE) as its computational backbone.

The Virtual Sensing Fabric: This is our core innovation. We built a data fusion AI model that learns the complex relationship between satellite observations and on-the-ground soil moisture. This model is trained on data from research farms with physical sensors, allowing it to generate accurate virtual measurements. It continuously ingests and processes multiple data streams via GEE:
GEE Sentinel-1 (Radar): The workhorse for soil moisture detection, its radar signal penetrates clouds, ensuring data availability even during tropical rainy seasons.
GEE Sentinel-2 (Optical): Provides high-resolution data on crop health (NDVI), allowing us to dynamically calculate the crop's real-time water demand (Kc), replacing static textbook values.
GEE & Open-Meteo Data: We integrate low-resolution satellite moisture data (SMAP), topography (Digital Elevation Models), and recent weather data (rainfall, evapotranspiration) to provide critical context for the AI model.
The Contextual Intelligence Engine: This is the irrigation brain. Instead of a generic LLM, we use a Physics-Informed AI Scheduler. This hybrid model is built on the rigid skeleton of the soil-water balance equation, a fundamental law of hydrology. It uses the "virtual sensor" data from Layer 1 to fill in the hard-to-measure variables, ensuring its recommendations are not just plausible but scientifically grounded and trustworthy. This layer also uses historical GEE data to automatically delineate management zones within a field for advanced Variable Rate Irrigation (VRI).
The Action & Advisory Layer: This layer delivers value to the user. We developed a simple mobile interface that provides an Explainable Irrigation Schedule, always giving the "why" behind a recommendation to build farmer trust. It includes a Water-Use Efficiency Dashboard to track ROI through water, energy, and cost savings. Finally, we built a robust API that allows our intelligence to be pushed directly to compatible smart irrigation controllers, enabling farmers to upgrade their existing hardware with our superior, data-driven insights.

Challenges we ran into

Model Accuracy and Ground-Truthing: Our biggest challenge was ensuring the "virtual sensor" was accurate. Satellite data provides a measurement for a 10x10 meter pixel, which is different from a physical sensor measuring a single point. This required extensive work to gather high-quality training data from partner farms and research institutions to teach our AI model how to correlate the two, a process known as validation.
Data Complexity and Processing: Fusing heterogeneous data—radar, optical, thermal, and meteorological—in near real-time is computationally intensive. Building a stable and efficient pipeline on Google Earth Engine to process this data across millions of hectares required significant expertise in both geospatial analysis and machine learning.
Farmer Adoption and Trust: Technology is only useful if it's used. We ran into the classic agritech challenge of bridging the digital divide, especially with older, smallholder farmers who may have low digital literacy and are rightfully skeptical of new tools. We had to design an extremely simple interface and focus on building trust by providing transparent, explainable recommendations rather than just black-box AI outputs.

Accomplishments that we're proud of

A Truly Hardware-Free Solution: Our proudest accomplishment is creating a scalable precision irrigation platform that requires zero capital expenditure from the farmer. By completely virtualizing the sensing layer, we have removed the primary barriers—cost, maintenance, and logistics—that have prevented widespread adoption of precision agriculture.
All-Weather Reliability: By building our core model on Sentinel-1 radar data, we overcame a major hurdle for remote sensing in tropical agriculture: clouds. Our system delivers reliable data year-round, even during the monsoon season, which is a significant advantage over optical-only solutions.
Scientifically Valid AI: We are proud of developing a physics-informed AI scheduler. This isn't just pattern recognition; it's an engine whose recommendations are grounded in the established science of hydrology. This makes the output reliable, defensible, and, most importantly, trustworthy for farmers.

Demonstrable Impact: Through our initial deployments, we've proven the model works. Our partners have seen measurable reductions in water and energy consumption while protecting or increasing their yields, validating the ROI and demonstrating a clear path to profitability and sustainability.

What we learned

Solve the Business Problem, Not the Tech Problem: We learned that farmers don't buy "AI" or "remote sensing"; they buy solutions to their problems. The key was to focus on delivering a clear ROI—saving water, cutting fuel costs, and increasing yield stability. The technology is the "how," but the value proposition is the "what."

Trust is the Most Important Input: The most sophisticated algorithm is useless if the farmer doesn't trust its recommendation. We learned that transparency is critical. Explaining why a decision is recommended, in simple terms, is more powerful than explaining the complexity of the model behind it.

Scalability Requires Simplicity: To reach millions of farmers, the solution has to be incredibly simple to deploy and use. By centralizing all complexity in the cloud and delivering a simple answer to a smartphone, we created a model that can scale frictionlessly, unlike hardware-based solutions that require physical installation and maintenance for every new user.

Data is the Ultimate Asset: We learned that while our primary product is the irrigation schedule, the long-term strategic asset is the aggregated data on water use, crop health, and yields. This data has immense secondary value for governments, insurers, and the entire food supply chain, creating a powerful and defensible business model.

What's next for AquaCrop - Optimizing Irrigation System using Aerospace Data

Our vision is to become the foundational intelligence layer for water management in global agriculture. Our next steps are focused on expansion and deepening our value.

Geographic and Crop Expansion: Having validated our model in Indonesia for key crops like rice and palm oil, our immediate next step is to expand to other major agricultural nations facing similar water challenges, such as Malaysia, Thailand, and the small-farm sector in the United States. We will simultaneously expand our AI models to support other high-value crops like maize, coffee, and sugarcane.

Enhanced Product Features: We will move beyond just irrigation scheduling to offer integrated fertigation recommendations (applying fertilizer through irrigation) and predictive alerts for pest and disease risk based on our granular moisture and climate data.

Building the Data-as-a-Service (DaaS) Platform: We will formalize our DaaS offering, providing anonymized, aggregated insights to government agencies for food security planning, financial institutions for de-risking agricultural loans, and commodity traders for supply chain optimization.

Deepening Automation and Integration: We will expand our API partnerships with smart irrigation hardware manufacturers to enable a seamless transition from recommendation to automated action, creating a fully integrated "sense-and-actuate" system for farms that are equipped for it.