Sharing a Project: GeoAI & WebGIS for Industrial Solar Rooftop Potential (EJIP) ☀️🏭
Hello Esri YPN Community! 👋
My name is Andre. Following up on my previous post about disaster mitigation, I am excited to share another recent spatial analysis project. This time, I focused on the intersection of renewable energy, deep learning, and spatial planning.
The project is titled GeoAI Solar Rooftop Potential for Industrial Decarbonization.
The Challenge Industrial estates hold massive solar potential and offer a highly scalable, no-land-take decarbonization pathway. However, decision-makers often face a critical barrier: the lack of a reliable, scalable rooftop inventory. Traditional manual assessments are time-consuming and result in fragmented data, which hinders effective investment prioritization.
The GeoAI & WebGIS Workflow To bridge this gap, I developed a workflow combining deep learning and spatial calculation, focusing on the East Jakarta Industrial Park (EJIP) as a case study.
Here is how I approached it using the ArcGIS ecosystem:
1. GeoAI Rooftop Extraction (ArcGIS Pro) Using high-resolution imagery, I trained a deep learning model to automate building footprint extraction.
Model: Mask R-CNN (ResNet-50) for instance segmentation.
Training: Based on 250 labeled training chips (256 px, batch size 3, 20 epochs).
Performance: The model achieved an Average Precision (AP) of 0.589 on the validation set. While this is screening-level performance, it efficiently generates the polygon footprint layer required for massive-scale area estimation.
2. Solar Energy & Economic Calculator Once the building footprints were refined, I integrated them with Global Horizontal Irradiation (GHI) data. I applied transparent calculation parameters to estimate the potential:
Effective rooftop factor: 0.75 (assuming 75% of the detected roof is usable).
PV module efficiency: 0.19.
Performance ratio (PR): 0.80.
Grid emission factor: 0.82 kg CO2/kWh (representing the local predominantly fossil-fuel-based grid).
Industrial Tariff: 1,444 IDR/kWh.
3. Decision-Ready WebGIS (ArcGIS Dashboards) Static data is rarely enough for stakeholders. I published the quantified data to ArcGIS Online and built a fully interactive Dashboard.
Explore the live demo here: 🔗 https://www.arcgis.com/apps/dashboards/eef44857c8b548f3a83b7732e6b1b96a
How to interact with the dashboard:
Click directly on any building footprint on the map to view its specific energy and economic metrics pop-up.
Use the "Select Industrial Building" dropdown filter on the side panel to easily search and isolate specific facilities.
Click on any building within the "Top 10 Buildings" list to automatically highlight its solar potential and guide prioritization.
The Impact
For the EJIP industrial area, this automated screening identified:
Total Solar Potential: 272.1 GWh/year.
Avoided Emissions: 223.1k tons CO2/year.
Estimated Cost Savings: IDR 392.9 billion/year.
This approach transforms raw imagery into decision-ready indicators, directly supporting SDG 7 (Affordable and Clean Energy), SDG 9, SDG 12, and SDG 13.
Explore the Full StoryMap I have documented the complete methodology, calculations, limitations, and the interactive Web Maps in an ArcGIS StoryMap. You can explore it here: 🔗 https://storymaps.arcgis.com/templates/28e9feaf42404325871246081c191001
Let's Discuss!
I would love to learn from this community's expertise. Has anyone else here utilized Mask R-CNN or other deep learning models for feature extraction in ArcGIS Pro? How do you typically handle quality assurance and manual corrections for automated building footprints before presenting them to stakeholders?
Looking forward to your insights!
Best regards,
Andre
