Geographic Information System and Machine Learning Approach for Solar Photovoltaic Site Selection: A Case Study in Pakistan

Hamza Ashraf, Jiajun Li, Zeyu Li, Azam Sohail, Ahmed Raza, Maryam Butt, Hameed UllahProcesses 2025 · DOI

TL;DR

This paper presents an integrated GIS-ML framework to pinpoint optimal locations for solar photovoltaic (PV) power plants in Pakistan's Cholistan Desert, a region facing significant energy shortages. By combining field data with advanced machine learning models, the study identifies high-potential zones and offers a scalable solution for renewable energy planning.

What Problem Does This Paper Address?

Pakistan, particularly the Punjab province, is grappling with a severe energy crisis characterized by substantial electricity shortages and a heavy reliance on imported fossil fuels. This deficit impacts both residential and industrial sectors, hindering economic stability. While the Cholistan Desert offers immense solar potential due to high irradiance, effectively identifying suitable sites for large-scale PV installations is complex. Existing methods often rely on fragmented approaches, theoretical data, or fail to account for the unique challenges of hyper-arid environments like extreme temperature gradients and seasonal variability. This research aims to bridge this gap by developing a robust, data-driven framework for precise and adaptable solar PV site selection.

Methods

The study employs a hybrid Geographic Information System (GIS) and Machine Learning (ML) approach. The core methodology involves:

  • Data Collection and Preprocessing: Gathering 14 conditioning factors (geophysical, socio-economic, and resource-related) from various sources, including field surveys, satellite imagery, and meteorological datasets. All data was resampled to a uniform 1 km resolution.
  • Machine Learning Models: Three widely used ML algorithms were implemented and compared:
  • Multilayer Perceptron (MLP): A feed-forward neural network with three hidden layers, optimized using the Adam optimizer.
  • Random Forest (RF): An ensemble learning method that builds multiple decision trees and aggregates their outputs, known for its robustness against overfitting.
  • Extreme Gradient Boosting (XGBoost): Another powerful ensemble technique that builds trees sequentially, minimizing residual errors.
  • Model Evaluation: Models were evaluated using Receiver Operating Characteristic (ROC) curves, Area Under the Curve (AUC), and Classification Accuracy (CA).
  • Feature Importance Analysis: The SHAP (SHapley Additive exPlanations) method was used to understand the contribution of each feature to the model's predictions, providing insights into what drives site suitability.

Datasets

The study utilized a combination of datasets, all processed to a 1 km spatial resolution:

  • Field Survey Data: Collected from 355 location points (both PV and non-PV sites) to serve as ground truth.
  • Solar Irradiation Data: Obtained from solargis.com.
  • Digital Elevation Model (DEM): Acquired from USGS.
  • Vegetation Data (NDVI): Derived from Sentinel-2 imagery.
  • Road Network and Population Density Data: Retrieved from data.humdata.org.
  • Wind Speed Data: Sourced from the ERA5 Monthly Aggregates dataset.
  • Temperature Data: Obtained from solargis.com.
  • Other factors: Including aspect, slope, carbon emissions, distance to residential areas, and sunshine duration.

No new benchmark datasets were introduced; rather, existing publicly available and field-collected data were integrated.

Key Insights and Ideas

  • Synergy of GIS and ML is Crucial: The study effectively demonstrates how combining GIS for spatial data management and ML for complex pattern recognition leads to more accurate and nuanced site selection than either method alone.
  • Field Data Matters: The inclusion of field survey data significantly enhances the reliability and practical relevance of the models, moving beyond purely theoretical or satellite-derived inputs.
  • Random Forest Reigns Supreme: Among the tested models, Random Forest consistently outperformed MLP and XGBoost in both predictive accuracy (AUC) and classification performance, making it the preferred choice for this application.
  • Key Drivers of Suitability: Global Horizontal Irradiance (GHI) and population density emerged as the most influential factors across models, highlighting the dual importance of solar resource availability and accessibility.
  • "Sunshine Duration" is Less Critical Than Expected: Surprisingly, sunshine duration showed negligible importance in all models. The authors hypothesize this is due to its redundancy with GHI, potential data limitations, or correlation with other features like temperature. This suggests that models can effectively capture solar potential through other, more direct indicators.

Key Results

  • Optimal Model: The Random Forest (RF) model achieved the highest performance with an AUC of 0.921 and a Classification Accuracy (CA) of 0.723.
  • High-Potential Zones Identified: Bahawalnagar (eastern region) and Bahawalpur (central region) were identified as prime locations, with 10.50% and 11.06% of their areas classified as "high" or "very high" probability for solar PV installation, respectively.
  • Area Suitability Distribution: For the RF model, 76.99% of the study area was classified as low suitability, 14.37% as moderate, 2.89% as high, and 5.75% as very high.
  • Estimated Solar Potential: The identified high- and very-high-suitability zones contribute disproportionately to the region's total solar energy potential, estimated at 120,475 TWh/year.
  • Wind Energy Potential: The identified high-suitability solar zones also show promise for wind farm feasibility due to favorable wind conditions and flat terrain.

Limitations and Future Work

  • Limitations:
    • The study acknowledges that "sunshine duration" had negligible importance, suggesting potential data limitations or redundancy with other features.
  • While the RF model showed strong performance, the study notes that population density had a moderate impact on accuracy, warranting further investigation.
  • Future Work:
    • Integrating real-time solar and wind data is recommended to further refine site selection and support multi-source renewable energy planning.
  • The methodology is presented as transferable to other arid or high-irradiance regions, offering a scalable blueprint for global climate resilience.
  • Further research could explore the integration of other renewable energy sources and a more detailed analysis of socio-economic and environmental constraints.

Summary updated

Abstract

Punjab, the most populous province in Pakistan, is currently facing substantial electricity shortages that are adversely affecting both residential and industrial sectors. To address this issue, the Cholistan Desert presents a promising solution due to its high solar irradiance, making it an ideal location for solar energy production. This study aims to identify the most suitable area for solar photovoltaic (PV) power plants in the Cholistan Desert using Geographic Information System (GIS) and machine learning techniques. The analysis included field survey data encompassing 14 conditioning factors such as geophysical, socio-economic, and resource conditions. Three machine learning models were utilized: Random Forest, XGBoost, and Multilayer Perceptron (MLP). The Random Forest model demonstrated superior performance with an AUC of 0.92, and feature importance was measured through SHAP. The resulting suitability map indicates that Bahawalnagar in the eastern region and Bahawalpur in the central region have 10.50% and 11.06% of their areas classified as having a “high” and “very high” probability for solar PV installation, respectively. For stakeholders in the wind industry, these regions also present potential for wind farm feasibility due to favorable wind conditions and flat terrain. The methodology can be adapted to prioritize wind energy sites by incorporating factors such as land availability, wind direction, and other related factors. Co-locating solar and wind farms in these regions could optimize land use, enhance grid stability, and support Pakistan’s renewable energy targets. Future research integrating real-time solar and wind data could further refine site selection and support multi-source renewable energy planning, providing actionable insights for policymakers and investors.
BibTeX
@inproceedings{ashraf2025geographic,
  author = {Hamza Ashraf and Jiajun Li and Zeyu Li and Azam Sohail and Ahmed Raza and Maryam Butt and Hameed Ullah},
  title = {{Geographic Information System and Machine Learning Approach for Solar Photovoltaic Site Selection: A Case Study in Pakistan}},
  booktitle = {Processes},
  year = {2025},
  doi = {10.3390/pr13040981},
}
Figure