Psi Factor

💎 Idea of Psi Factor

The Problem: Disasters caused by unpredictable surface runoff demand answers, not just current reactions. The rapid surface runoff and soil instability cause profound damage to small catchments, yet the root causes and specific geographic origins remain hidden. This lack of precise data forces authorities to underestimate risks and waste resources on temporary fixes that fail to produce long-term results.

The Solution: We are building an investigative diagnostic tool designed to trace hazardous water flows back to their exact source. By integrating localized geotechnical measurements with spatial data through a robust, reproducible data pipeline, we map out areas with high Ψ (Psi) factor and thus of high significance for hazardous flows. This turns guesswork into certainty, empowering decision-makers to fund and deploy interventions only in the exact locations where they will successfully protect the catchment, improve the water retention, and help the community.

💧 EU Space for Water

Our project directly addresses the Challenge Case 3 (Land Cover Type) & Case 1 (Risk). 

By providing a high-fidelity diagnostic tool for small catchments, we contribute to water resource management by predicting and mapping rapid runoff events before they escalate. This ensures that environmental modifications (such as retention basins or soil stabilization) are placed effectively, reducing disaster risk, preventing erosion, and securing the broader watershed ecosystem.

Main objective

Quantitative prediction of peak flows (flood water with a probability of 1%) for small catchments (up to 50 km²) and ungauged catchments. 

The system's advantage will be precise knowledge of the impact of local land cover changes for each catchment point (towns, bridges, culverts, and temporal changes during rainfall). At the same time, it is possible to simulate projected spatial development changes and assess their impact. 

The project consists of several stages; this is the first phase, which allows us to estimate which rainfall distribution is the most unfavorable for difficult, ungauged catchments with minimal input from the designer. Additional benefits include identifying regions (sub-catchments) that are crucial for the runoff system. At this stage, we can detect hydrosystem damages that lead to significant negative changes in runoff. We can also assess the quality of works aimed at increasing the system's resilience to sudden weather events. 

We have introduced an interesting categorical dependence of the Ψ (Psi) index (based on "Field measurement of soil erosion and runoff", by N. W. Hudson, Silsoe Associates Ampthill, Bedford, United Kingdom; Food and Agriculture Organization of the United Nations Rome, 1993). The Ψ index on Topographic Wetness Index (TWI), substrate type, and strongly emphasized satellite measurements —specifically, catchment temperature measurements from Copernicus and soil moisture data from NASA.

Vision and Spatial Module (AI / Vision LLM + GIS)

Project objective

Automatic detection, land cover classification, and determination of runoff and roughness coefficients based on spatial data fusion. This module is responsible for the acquisition and semantic understanding of space to provide parameters for the hydrological model.

Input Data and 🛰️ EU space technologies used: 

• Images and remote sensing: Orthophotomaps.
• Copernicus satellite data – including the imperviousness layer, temperature etc. The data from Sentinel-2, Sentinel-3, and other satellite sources are integrated across different spatial resolutions to analyze land surface temperatures and various vegetation and soil moisture indices (such as NDMI, MSI, and SSM) on specific dates, ultimately to assess surface wetness and plant water stress.
• LiDAR (DSM / DTM): To determine vegetation height.
• Vector databases:
  • BDOT10k: General information on spatial development.
  • BDOT500 / GESUT: Precise information on surfaces (squares, roads, sidewalks) and underground infrastructure of the latter (we currently do not have these - it's work for the future).
  • Cadastral map (Parcels): Essential for categorization and ownership regarding investments in blue-green infrastructure or land designation - forest/field/road.- 
• Thematic maps: Soil (county offices / rasters) and geological (for general hydrogeological characteristics) – we drew core of our knowledge about soils and substrate structure from here.

Tasks for the Vision LLM / AI model

Intelligent terrain classification: 

• Utilizing LiDAR data and imagery to recognize forest types (deciduous/coniferous) and forestation density.
• Invasive species detection: Identification of zones occupied by invasive vegetation (e.g., Japanese knotweed, Sosnowsky's hogweed, wild cucumber) - not yet implemented(!).
• Detection of fast runoff zones: Verification of BDOT data against the actual state from orthophotomaps to identify new housing developments, warehouse halls, parking lots, and roads.
• Runoff coefficient (Ψ) map generation: Assigning Ψ values to specific polygons/pixels on the map (0,…,1) based on data fusion.
• Surface categorization for roughness (n)

The operational scope for Copernicus maps is to check how the surface behaves differently upon wetting - for example, how rapidly moisture changes in a situation of decreasing humidity (drying) or increasing humidity (wetting). Additionally, finding correlations with cover intensity.

The formula calculated by us for Ψ is coherent with the above almost everywhere.

We assume that the runoff coefficient is equal to the sum of the material, morphological, and temperature-related components. For paved roads, squares, roofs, and surface water, we assign its value directly.

In other cases, the following values have been introduced:
To describe the non-linear relationship between temperature and its impact on the runoff coefficient, we proposed the CTCT function, which in a simplified linear form looks as follows for cohesive soils:

Liquid and soft-plastic cohesive materials:

Plastic cohesive materials:

Hard-plastic cohesive materials:

For granular (non-cohesive) soils, the formula takes the following form:

Wet granular soils:

CT(t)=0.50

Moist granular soils:

Slightly moist granular soils:

The last proposed component of the formula is the CSCS component, which is related to the scaled value of the Topographic Wetness Index (TWI - Urbański J. (2012). GIS in natural science research. University of Gdańsk Publishing House; Sørensen, R., Zinko, U., & Seibert, J. (2006). On the calculation of the topographic wetness index: evaluation of different methods based on field observations. Hydrology and Earth System Sciences, 10(1), 101-112.), where:

Where aa – specific catchment area. This is the catchment area situated above a given point, from which water flows to this point, divided by the grid cell width; and tan⁡βtanβ – local terrain slope (the tangent of the slope angle). It tells us how quickly water will drain from this location. The scaling was performed without prior calibration, in the form of a hypothesis, as:

Project repository

https://github.com/mdziezyc/psi-patrol-cassini-hackhaton/

🤼 Team: Psi (Ψ) Patrol

• Maciej Dzieżyc - Senior Data Scientist with over 10 years of experience in industry and academia. Expert in bringing business value with AI.
• Janusz Vitalis Kozubal - PhD Eng., researcher of Civil Engineering in the field of Geotechnics and Hydrotechnics (WUST). Scientifically involved in water treatment and retention projects.
• Łukasz Janiec – Assistant and Robotics Researcher at the Department of Cybernetics and Robotics (WUST), AI/ML & Software Development Specialist in systems automation and robotization.
• Magdalena Kozubal - Civil engineer, WUST Master student, specialization in geotechnics. Interested in underground infrastructure, particularly subterranean tunnels.
• Kamil Górniak - Civil engineer, WUST Master student, specialization in geotechnics. Interested in numerical methods and analysis of the stability of the lunar lava caves.
• Adam Cybulski - Senior Quality Assurance Engineer with focus on data and analytics for the last 4 years. Passionate about new technologies and AI.