📄 other

Random Forest-Based Prediction of Geothermal Heat Flow in the Arabian-Nubian Shield: A Multi-Proxy Satellite Geophysical Approach

This study presents the first spatially continuous geothermal model for the Arabian-Nubian Shield by employing a Random Forest algorithm integrated with sixteen multi-proxy satellite geophysical predictors to map terrestrial heat flow, identify high-potential geothermal zones, and guide future carbon-reduction energy development.

Original authors: Menna Haggag, Mohamed Sobh, Hosni H. Ghazala

Published 2026-07-27

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Menna Haggag, Mohamed Sobh, Hosni H. Ghazala

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Earth's Hidden Thermostat

Imagine the Earth not just as a solid rock we walk on, but as a giant, slow-cooking stew pot. Deep inside, the planet is still hot from its birth billions of years ago, and it's constantly leaking that heat up toward the surface. This invisible warmth is called "terrestrial heat flow." For most of us, this heat is just a background hum, but for scientists and energy companies, it's a potential goldmine. If you can find a spot where the Earth is leaking heat fast enough, you can turn that energy into electricity to power cities or heat water for desalination, all without burning fossil fuels.

The tricky part is that the Earth is a bit like a mystery box. We can't drill deep enough to see the whole picture, so we only have a few "thermometers" (measurements) stuck in the ground here and there. In many places, especially across the vast Arabian-Nubian Shield (a massive stretch of land covering parts of Egypt, Saudi Arabia, and beyond), these thermometers are incredibly scarce. It's like trying to guess the temperature of an entire ocean by sticking a thermometer in just a few random puddles. To solve this, scientists are turning to a clever trick: using the clues the Earth leaves on its surface—like the shape of the ground, the strength of its magnetic field, and how deep the crust is—to guess what's happening deep underground. This is where machine learning comes in, acting as a super-smart detective that can connect the dots between surface clues and deep-seated heat.

The Paper's Story: Teaching a Robot to Read the Earth's Heat

In this study, a team of researchers led by Menna Haggag, Mohamed Sobh, and Hosni H. Ghazala decided to build a "crystal ball" for geothermal energy across the entire Arabian-Nubian Shield. Since they didn't have enough direct heat measurements to draw a map by hand, they taught a computer algorithm called Random Forest Regression to do the guessing for them. Think of this algorithm not as a single smart person, but as a massive crowd of 490 different "decision trees" (a type of computer model). Each tree in the crowd looks at the data slightly differently, and then they all vote on the final answer. This "crowd wisdom" approach helps avoid mistakes that a single model might make.

The team fed this digital crowd a diet of sixteen different types of "clues" gathered from satellites and global databases. These clues included things like how deep the boundary between the Earth's crust and mantle is (the Moho), how thick the rocky shell of the Earth is (the lithosphere), the speed of seismic waves traveling through the ground, and even the shape of the land itself. Before letting the computer learn, they used a special filter called an Isolation Forest to spot and remove any weird, broken data points that looked like errors, ensuring the computer only learned from the truth.

What they found:
After training the model, the computer produced a smooth, continuous heat map for the entire region, filling in the gaps where no measurements existed. The results were surprisingly accurate. When they tested the model against the few real measurements they had, it got a score of 0.92 (on a scale where 1.0 is perfect), meaning it explained 92% of the variations in the real data. The average error was just 20.4 mW m⁻², which is a very small margin of error for such a huge area.

The map revealed three distinct "thermal neighborhoods":

The Hot Zones: Along the Red Sea Rift and the Afar Triangle, the heat flow is scorching, ranging from over 80 to 200 mW m⁻². This is where the Earth's crust is thinning and the hot mantle is rising close to the surface, making it a prime spot for high-energy geothermal power plants.
The Warm Zones: Under the ancient Precambrian shields (the rocky heart of the region), the heat is moderate, sitting between 60 and 80 mW m⁻². This is still hot enough to be useful, especially if there are cracks in the rock to let the heat circulate.
The Cool Zones: On the stable Arabian Platform to the east, the heat flow drops below 60 mW m⁻². Here, the Earth's crust is thick and deep, acting like a heavy blanket that keeps the heat trapped far below.

The "Why" Behind the Map:
The researchers didn't just make a pretty picture; they figured out why the heat is where it is. By asking the computer which clues mattered most, they discovered that the two biggest factors controlling the heat are the depth of the Moho (how deep the crust is) and the depth of the Lithosphere-Asthenosphere Boundary (how deep the rocky shell is). In simple terms, the thinner the Earth's shell, the closer the super-hot mantle is to the surface, and the hotter it gets. This confirms that the tectonic stretching of the Red Sea is the main engine driving the high heat in that area.

While the deep structure sets the stage, the study highlights that volcanic centers and fault networks act as the delivery system. Although the final, most efficient model relied on the top 12 predictors (excluding proximity to volcanoes as a primary driver), the analysis confirmed that areas where high heat flow overlaps with dense fracture networks create "High-Enthalpy Corridors." These are the sweet spots where heat can actually travel up to the surface via fluid circulation. The researchers used a separate topographic analysis (CET) to detect these dense fracture networks, proving that these features are essential targets for finding usable energy, even if they weren't the top statistical drivers in the final heat flow calculation.

How sure are they?
The authors are very confident in their map for the areas where they have data to check against, noting that the model's predictions match real-world measurements almost perfectly. However, they are careful to point out that in areas with very few measurements (like some parts of the deep desert), the "uncertainty" is higher. They even created a special "confidence map" that shows exactly where the computer is guessing with high certainty and where it's a bit more unsure. They suggest that while the map is a powerful tool for finding new drilling targets, explorers should still be cautious in the "high uncertainty" zones and might need to gather more local data before digging deep.

In the end, this paper hands energy planners a new, high-resolution map that turns a data-sparse desert into a clear guide for where to look for clean, renewable geothermal energy, potentially helping countries in the region move toward a greener future.

Technical Summary: Random Forest-Based Prediction of Geothermal Heat Flow in the Arabian-Nubian Shield

Problem Statement
The Arabian-Nubian Shield (ANS) represents a tectonically active region with significant geothermal potential, yet its development is hindered by a severe "data-sparsity" problem. Direct terrestrial heat flow (THF) measurements are scarce, unevenly distributed, and insufficient for conventional interpolation methods. This lack of continuous thermal data increases exploration risk and limits the ability to identify viable targets for geothermal energy, particularly for strategic applications like seawater desalination and green hydrogen production in the Middle East and North Africa (MENA) region. Traditional geostatistical techniques often fail to resolve the complex, non-linear thermal gradients created by the ANS's diverse tectonic settings, ranging from active rifting to stable cratonic lithosphere.

Methodology
To address these limitations, the authors developed a data-driven, machine learning workflow using Random Forest Regression (RFR) to construct the first spatially continuous, high-resolution (0.5°) geothermal model for the entire ANS. The methodology follows a six-stage pipeline:

Data Acquisition and Integration: Sixteen geophysical and geological predictor variables were compiled from global and regional repositories. These predictors span four categories:
- Lithospheric Structure: Moho depth, Lithosphere-Asthenosphere Boundary (LAB) depth, and lithospheric/crustal densities.
- Thermal and Seismic Indicators: Curie Point Depth (CPD), and P-wave/S-wave velocity perturbations at 150 km depth.
- Potential Field Anomalies: Bouguer and free-air gravity anomalies, geoid height, gravity shape index, and magnetic anomalies.
- Surface and Tectonic Controls: Digital Elevation Model (DEM), tectonic regionalization classes, lithological units (GLiM), and proximity to Neogene-Quaternary volcanic centers.
Preprocessing: Data were harmonized onto a unified 0.5° grid. Multivariate outliers were rigorously removed using an Isolation Forest algorithm to distinguish erroneous measurements from genuine thermal anomalies. Continuous variables were normalized, and missing values were interpolated.
Model Development: An ensemble of decision trees was trained using RFR. The model leverages "bagging" (bootstrap aggregating) and random feature selection to reduce variance and prevent overfitting, effectively capturing non-linear relationships between surface heat flow and subsurface parameters without requiring linear transformations.
Hyperparameter Optimization: A grid search coupled with 5-fold cross-validation was employed to optimize hyperparameters (e.g., number of trees, tree depth, minimum samples per split), minimizing Mean Squared Error (MSE).
Feature Importance and Selection: The model ranked predictors by their contribution to variance reduction. A recursive feature selection process determined that a model utilizing the top 12 observables achieved optimal performance, balancing accuracy with complexity.
Uncertainty Quantification: An ensemble-based approach was used to generate uncertainty maps, defining prediction confidence ranges and coefficients of variation (CV) for each grid cell. Additionally, Computer-Enhanced Topology (CET) analysis was applied to Digital Elevation Models to detect structural lineaments (faults and fractures) and correlate them with heat flow anomalies.

Key Results
The optimized RFR model demonstrated strong predictive performance, achieving a coefficient of determination ( $R^2$ ) of 0.92 and a Root Mean Square Error (RMSE) of 20.4 mW m⁻² on the validation set. The model successfully reproduced the statistical distribution of observed heat flow, capturing both low-heat-flow cratonic regions and high-heat-flow rift zones.

The study delineated three distinct thermal provinces across the ANS:

High-Potential Zones (>80–90 mW m⁻²): Concentrated along the Red Sea Rift axis, the Afar Triple Junction, and the western margins of the Arabian Shield. These areas are characterized by crustal thinning, shallow LAB, and active magmatism, with predicted values reaching up to 200 mW m⁻².
Moderate-Potential Zones (60–80 mW m⁻²): Found beneath the Precambrian shields (Arabian and Nubian), particularly near rift-proximal areas and volcanic centers (harrats).
Low-Potential Zones (<60 mW m⁻²): Dominating the stable Arabian Platform, where thick, cold lithospheric roots suppress heat flow to values between 35 and 50 mW m⁻².

Variable importance analysis identified Moho depth and LAB depth as the dominant controls on regional thermal variability, followed by geoid height and S-wave velocity anomalies. This confirms that vertical thermal coupling from the mantle to the surface is the primary driver of the ANS's thermal state. Furthermore, the integration of CET analysis revealed that high-heat-flow zones often coincide with dense networks of NNW-SSE trending lineaments, suggesting these structural features facilitate hydrothermal circulation.

Significance and Claims
The authors claim this work provides the first high-resolution, spatially continuous geothermal heat flow map for the entire Arabian-Nubian Shield, transforming sparse borehole data into an actionable, physically interpretable resource. The study asserts that its primary contribution is the reduction of geothermal exploration risk by identifying "drill-ready targets" where high thermal flux overlaps with structural permeability (faults/fractures).

By linking statistical predictions to fundamental geophysical parameters (specifically Moho and LAB depths), the model offers a diagnostic tool for understanding lithospheric controls on the geothermal regime. The authors position this framework as a transferable solution for other data-sparse, tectonically active regions globally, requiring only publicly available datasets and open-source software. Ultimately, the paper argues that this approach supports regional energy strategies (such as Saudi Vision 2030 and Egypt's Sustainable Energy Strategy 2035) by enabling the identification of high-enthalpy targets for flash-steam power plants and moderate-enthalpy zones for binary cycle or Enhanced Geothermal Systems (EGS), thereby facilitating carbon-reduction goals.

The Earth's Hidden Thermostat

The Paper's Story: Teaching a Robot to Read the Earth's Heat

More like this