Demeter's Oracle

Project Summary

Context

Climate change is increasing the speed and frequency of extreme events. Drought is no longer always a slow process. In many cases, it develops rapidly, leaving little time to react, which usually traduces into negative impacts on plant physiology and heavy wildfire risk.

The Problem

Historically, droughts' development has been slow and spreading in a cascading way, usually starting with a lack of precipitation and, with enough time, this entails plant water stress and agricultural droughts. It is a complex and diverse process.

In last 20 years it has been observed that this process sometime occurs in a very fast fashion. Sometimes, low precipitation coincides with abnormally high temperatures, often during heatwaves, along with high winds and shifts in solar radiation. This phenomena has been denominated by academy as "flash drought".

Flash droughts are rapid-onset events that develop over 1–3 weeks, unlike traditional droughts which evolve over months.

Thus, the issue is not only lack of rainfall, but how quickly the land dries out under heat and dry conditions.

Why It Matters

The ambiguity of the term and difficulty of its detection has made flash droughts not well-know among the general public. Nevertheless, the effects of flash droughts are real, and as soil moisture degrades, the risk of this kind of phenomena will increment.

Graphic from OECD's Global Drought Outlook (link).

For example, last 2025 the Ourense–León–Zamora triangle suffered from a flash drought that concluded in one of the worst wildfires in Spain history (link).

Flash droughts will add to economic damage produced by “regular” droughts:

• Drought has caused over $278 billion in global damage
• In Spain, major events can reduce crop yields by up to 10%

Importantly, flash drought does not only affect dry regions. Humid areas like Galicia are increasingly at risk, because:

• ecosystems are not adapted to such rapid drying
• sudden heat events can trigger stress before any visible warning

Thus flash drought basically leads to:

• crop losses
• forest degradation
• increased wildfire risk

Seeing all this, and within the framework Cassini challenge #3: disaster risk monitoring, we asked ourselves if it was possible to predict flash drought leveraging Sentinel data and we think that it is feasible. 

We present: Demeter's Oracle.

Demeter's Oracle

Our Approach

Ivan Noguera et al. (link) closely defines the criteria to identify a flash drought in Spain using SPEI data. This way, we can use SPEI data provided by CSIC (link) to identify when a flash drought happens. Knowing "when" it happens we can look for Sentinel data that corresponds to flash droughts.

We combine ground SPEI data with Sentinel data corresponding to flash droughts to train a ML model (random forest). The model finds hidden correlations among the data and help predict flash droughts some weeks before they happen.

This usage of ML model to predict flash droughts is being explored by the research community (example paper) and it is expected that the field will continue to advance. This gives us the opportunity to leverage the knowledge and offer it to society, specially in the agriculture and forestall related sectors.

The principal source of data that we use for flash drought prediction is Sentinel-1 SAR, concretely VH/VV and Surface Soil Moisture (SSM).

We plan to use, although with the limitation of clouds, Sentinel-2 data like:

• Moisture index (NDMI)
• Short infrared composite (SWRI)

We also plan to integrate surface temperature, wind and solar radiance data from public meteorological datasets.

This way, the ML model can detect large and rapid soil moisture changes and combine it with weather indicators to predict a flash drought in a specific geographical area. Sentinel-1 provides good-enough observations every 5-7 days, allowing detection of drying trends that traditional systems usually miss.

What We Deliver

We offer an app that can be used to both monitor and predict flash droughts. A prototype video can be seen at attachments. Also we could find some initial insights about the ML model with VH/VV data from Sentinel 1 in a small geographical between Lauroco and La Rúa in Galicia,  which suffered from a devastating wildfire in 2025 as a consequence of a flash drought. To see some result trainin data, see Apendix1.

 We plan to develop a functional app that can:

• Display localized alerts at parcel level
• Detect rapid drying events
• Throw early warning signals of increased flash drought risk

This way, the application can be used to identify when conditions are evolving toward a flash drought, enabling earlier intervention and evading its surprise effect.

What Makes Us Special

Systems such as Copernicus and the European Drought Observatory provide reliable drought large-scale monitoring.  AEMET also has an open free service offering SPEI data (very related with drought events). CSIC has recently deployed a public-accessible tool to monitor flash droughts, but need at least 4 weeks of data to detect the event.

With all this, most found meteorological tools and alternatives are useful to monitor classical droughts. We found few alternatives to our idea being our main competition the Rapid Onset Droughts (ROD) service provided by NOAA in the EEUU. This is showed in the next table:

Our solution:

• Focuses on flash drought prediction, not just flash drought state
• Works under cloud cover
• Has a decent forecast window
• Is designed for operational decisions at local level

The Business Plan

Target Users

Primary:

• Short term:
  • Farmers and land owners
• Mid term:
  • Forest and wildfire management

Secondary:

• Mid term:
  • Water infraestructure administrators
• Long term:
  • Insurance companies

Market Opportunity

• Global precision farming market: $14.2B (2025) → $48.4B (2035)
• Europe precision agriculture market: $4.87B (2025) → $15.47B (2034)
• Spain precision farming market: $383.7M (2024) → $1.19B (2033)

Business Model

Pricing structure

Basic (Awareness)

• flash drought prediction in one small area
• for farmer (200€/yearly)
• light prediction model (less costs for us)

Pro (Monitoring)

• alerts + probability maps + indicators
• mid-tier pricing for coorporatives (3K€/yearly)
• advanced model

Premium (Decision Support)

• prediction + full analytics + API
• full-access model for authorities (20K€/yearly)
• high-value contracts for organizations

Different users = different value ⟶ different pricing

Revenue Logic

Year 1 (Pilot)

• 25K€ revenue
• 150K€ cost
• -125K€ lost
• Validation phase

Year 3 (Growth)

• 100K€ revenue
• Near break-even

Year 5 (Scale)

• 300K€+ revenue
• 100–€200K€ profit

Roadmap

• Phase 1 (0–6 months)
  • Galicia MVP
  • Pilot users
  • Validate system
• Phase 2 (6–18 months)
  • Expand to Spain / Southern Europe
  • Improve prediction models
  • Onboard more clients
• Phase 3 (18+ months)
  • Europe-wide expansion
  • API services
  • Partnerships
• Long-term possibility: Advanced soil moisture modeling via L-band satellites launches.

The system is scalable, with low marginal cost for new regions.

The team

• Isabel Schweim - CEO, Project Manager
• Marco A. Villa - CTO
• Miguel Carralero - Data Analyst, Head of Business & Marketing
• Brian Fortich - Back-End Expert
• Elizabeth Oppong - AI / Machine Learning Division Leader
• Luciana Diaz - Front-end Leader 

Repository

Public source repository:

https://github.com/Eliiss/flash_drought_app