Rooftop Solar and Battery Sizing Advisor Under Net Billing Using Open Building Footprints

The problem: households buy solar on an installer's guess

Pakistan's rooftop solar growth has been extraordinary. pv magazine reported in June 2025 that net-metered capacity had reached 5.3 GW, and imports of panels far exceeded that because many systems are not net-metered at all. Most households size their system the same way: an installer looks at the last bill and suggests a number of panels and a battery.

The rules have also changed. In February 2026 NEPRA issued Prosumer Regulations that replace unit-for-unit net metering with net billing for new applicants, settling exports and imports at different rates. Under net billing, the size that maximised savings before can now be wrong. Energy you use yourself at the moment of generation is worth more than energy you export, so the timing of your load matters, and so does the battery.

What already exists and the gap

• Global Solar Atlas (World Bank) and NASA POWER give free irradiance and generation estimates for any point, but know nothing about your roof or your bill.
• PVWatts and PVGIS simulate a system you describe, but you must already know the size and orientation, and neither models a South Asian tariff.
• Google Project Sunroof pairs roofs with solar estimates in some countries, not in Pakistan.
• Installer calculators exist to sell a system and rarely show assumptions.

The gap is a neutral tool that starts from the actual rooftop and the actual bill history, models the current settlement rules, and shows trade-offs instead of one answer.

Inspiration

Google's Project Sunroof showed the value of combining roof geometry with solar simulation at scale. The open building footprints you need come from Google's Open Buildings dataset, built with the method in Sirko and colleagues' 2021 paper "Continental-Scale Building Detection from High Resolution Satellite Imagery", which now covers South Asia. The simulation side follows the open pvlib library, widely used in solar research.

What you build

• A homeowner clicks their building on a map, confirms roof area and obstructions, and enters 12 months of bill units.
• The tool simulates hourly generation for a typical year, builds a load profile, and tries several panel and battery sizes.
• The result page shows, for each option, how much energy is used directly, how much is exported, how much the battery stores, and a payback range under editable tariff assumptions.
• An installer receives a shareable report with every assumption listed, which makes quotes comparable.

The technically interesting part

• From footprint to usable area. Footprints include water tanks, stair towers and parapets. Apply setbacks, let users mark obstructions, and measure how much this changes the result.
• Monthly bills to hourly load. You only have monthly units. Build a load profile by combining the monthly totals with typical daily shapes for summer and winter, adjusted by a short appliance questionnaire (air conditioners, water pumps, inverter ACs). State clearly that this is an estimate.
• Settlement logic. Write the rules as data, not code: import rate, export rate, any caps, and how each billing period nets. When regulations change again, the rules file changes, not the simulator.
• Battery dispatch. A simple rule (charge from surplus, discharge in the evening peak) is enough. Compare it with a small optimiser to see how much smarter dispatch is worth.
• Uncertainty. Dust, heat and inverter clipping reduce real output. Calibrate a derating factor against real inverter logs instead of trusting nameplate values.

How to evaluate it

• Generation accuracy: compare simulated monthly generation with inverter logs from at least three real systems, and report the error.
• Area accuracy: compare tool-estimated usable roof area with tape-measured area on ten roofs.
• Decision usefulness: give five households their report and record whether it changed the size they were considering, and why.
• Sensitivity: show how the recommended size moves as the export rate changes.

Suggested plan

• Weeks 1 to 3: pvlib basics, irradiance data download, simulate one known system.
• Weeks 4 to 6: footprint selection, usable area, obstructions.
• Weeks 7 to 9: load profile from bills, tariff rules file, battery dispatch.
• Weeks 10 to 12: sizing comparison, report page, validation with inverter logs.
• Weeks 13 to 15: user tests, write-up and demo.

Path to a product

Banks and microfinance lenders offering solar loans, consumer groups and fair installers all benefit from a neutral sizing tool. A free public calculator with a paid version for lenders, who need consistent assessments across thousands of applications, is a realistic model. The skills transfer to energy analytics roles, a growing field across South Asia and the Gulf.

Scope warning

Do not quote prices for panels or batteries; let users enter their own quotes. Do not claim to model shading from trees or neighbouring buildings unless you have the height data. Keep the settlement rules editable and dated, because they will change during your project.

Learn what you need on Sefism

• Data science roadmap: modelling, uncertainty and presenting estimates honestly.
• Data analytics roadmap: turning bill history into a usable load profile.
• pandas User Guide: hourly time series and resampling.
• FastAPI Tutorial: User Guide: serving the simulator to the web front end.
• Streamlit Documentation: a quick first version of the report page.

Get help with this idea

The risks are a confident number built on a guessed load profile, and rules that change under you. Tauseef can help you decide which assumptions to expose, plan the validation with real inverter logs and keep the scope to one semester. If you are new here, create a free account first.

• Discuss this idea with a mentor
• Send a query to Tauseef
• Book a 1:1 session