Skip to content

PV Estimation Pipeline

Estimates the return on investment (ROI) of photovoltaic installations on rooftops in Trentino. Uses the celine-roi library to compute financial metrics for each eligible building.

Includes a Streamlit dashboard for policy makers and REC managers to evaluate investment at province and primary substation level.

Quick start

cd apps/pv_estimation

task estimate                   # compute ROI for all buildings (incremental)
task dbt:seed                   # load adoption rate parameters
task dbt:run                    # build dbt models (staging → gold + REC)
task dashboard                  # launch the Streamlit dashboard

Full refresh (recompute everything):

task run:full-refresh

Task reference

task estimate                   Parallel ROI estimation (incremental)
task estimate:full-refresh      Truncate and recompute all buildings
task estimate:dry-run           Test on 10 buildings without writing to DB

task dbt:deps                   Install dbt packages
task dbt:seed                   Load seed data (adoption rate curve)
task dbt:run                    Run all dbt models
task dbt:gold                   Run gold models only
task dbt:rec                    Run REC substation models only
task dbt:full-refresh           Full refresh all dbt models
task dbt:test                   Run dbt tests

task run                        Full pipeline: estimate + seed + dbt
task run:full-refresh           Full refresh of everything

task dashboard                  Launch Streamlit dashboard
task serve                      Start Prefect flow server (dev mode)

Pass extra arguments with --:

task estimate -- --limit 5000 --workers 4
task estimate:full-refresh -- --batch-size 2000

Data flow

This pipeline sits downstream of three other pipelines:

overture → trentino_rooftops → pv_detection → pv_estimation

  1. overture ingests Overture Maps building footprints and produces a silver table with geometry, area and building metadata.

  2. trentino_rooftops filters buildings against regulatory constraints (aree non idonee, vincoli diretti/indiretti) to produce a gold table of suitable rooftops.

  3. pv_detection runs a vision model on aerial imagery to flag which buildings already have PV panels installed.

  4. pv_estimation (this pipeline) takes suitable buildings and estimates the financial return of installing panels.

The pipeline has no meltano extraction step. The Prefect flow reads directly from the upstream gold/silver tables, computes ROI via celine-roi, and writes results to a raw table. dbt then transforms through staging, silver and gold.

Eligible buildings

Controlled by require_detection in config.yaml:

  • require_detection: true — only buildings scanned by pv_detection and confirmed as has_pv = false. Conservative, waits for detection coverage.
  • require_detection: false — all buildings from pv_building_suitability. Useful for full-coverage projections before detection is complete.

In both cases, buildings must pass the regulatory suitability filter from trentino_rooftops and meet the minimum system size (min_kwp, default 3 kWp).

Building classification

Overture Maps building_class is mostly NULL, so the pipeline uses a heuristic to classify buildings based on footprint area and floor count:

if Overture building_class is present:
    use the explicit class mapping from config

else if num_floors > 3:
    → commercial (likely apartment block or office)

else if footprint_area_m2 > 200:
    if area >= 500 and floors <= 1:
        → industrial (large single-story: warehouse, factory, barn)
    else:
        → commercial

else:
    → residential

The thresholds are configurable under estimation.classification in config.yaml. The classification drives both the consumption model and the celine-roi user_type (which affects tax treatment and incentive eligibility).

Future improvements: cross-reference with OSM building tags (already in the osm pipeline) or cadastral data from Provincia di Trento for more accurate classification.

Estimation logic

For each building the pipeline estimates input parameters from the available data (geometry, footprint area, floor count) and calls celine-roi to compute the financial outcome.

PV capacity (kWp)

System size is the minimum of three constraints:

rooftop_kwp = footprint_area_m2 * panel_kwp_per_m2
kwp = min(rooftop_kwp, consumption_kwp, max_kwp)

Where panel_kwp_per_m2 is 0.130 kWp/m2, derived from:

  • 65% usable roof fraction (excluding edges, obstructions, orientation losses)
  • 200 Wp/m2 panel density (standard monocrystalline modules)

Residential sizing: the system is sized to match the household's actual consumption plus a headroom factor (default 20%), not to fill the entire roof:

consumption_kwp = consumption_kwh * (1 + headroom) / specific_yield

A household consuming 3,500 kWh/year gets a system of ~3.8 kWp (producing ~4,200 kWh), regardless of whether the roof could fit 20 kWp. This reflects real homeowner behavior — over-building a system that mostly exports to the grid at low RID tariffs is a poor investment.

Non-residential: the system fills the available rooftop up to max_kwp, since commercial/industrial buildings typically have high enough consumption to absorb the production.

The result is capped at max_kwp (default 20 kWp). This matches the Italian IRPEF tax deduction threshold — systems above 20 kWp do not qualify for the 50% deduction.

Buildings with estimated kwp below min_kwp (default 3 kWp, roughly 6 panels) are skipped. Below this size the fixed costs (inverter, permitting, scaffolding, electrician) make the installation uneconomical.

Annual production (kWh)

annual_production_kwh = kwp * specific_yield

Where specific_yield is 1100 kWh/kWp/year, a conservative estimate for Trentino based on average solar irradiance at typical tilt angles.

This value is passed directly to celine-roi, bypassing the PVGIS API call. This avoids rate limiting when processing hundreds of thousands of buildings and is acceptable for a screening-level estimate.

Annual consumption (kWh)

Two different models depending on building type:

Residential (fixed per household):

annual_consumption_kwh = 3500  (configurable)

A typical Italian household consumes 2700-4500 kWh/year (ARERA data). The default of 3500 kWh represents a mid-range family with modern appliances. This is a fixed value per household, NOT proportional to footprint area — a 60 m2 apartment and a 200 m2 villa have similar electricity consumption.

Non-residential (area-based):

annual_consumption_kwh = footprint_area_m2 * consumption_per_m2

Where consumption_per_m2 depends on user_type:

office:       80 kWh/m2/year
commercial:  100 kWh/m2/year
industrial:  150 kWh/m2/year
agricultural: 30 kWh/m2/year

For non-residential buildings, energy consumption does scale with building size (lighting, HVAC, machinery, refrigeration), so the area-based model is appropriate.

CAPEX (EUR)

capex = kwp * capex_per_kwp

Where capex_per_kwp is 1500 EUR/kWp. This is a mid-range estimate for residential systems under 20 kWp including panels, inverter, mounting and installation. It does not include VAT (celine-roi expects net-of-IVA values).

Incentive regime

Configured as regime in config.yaml (default: RID). Options:

  • RID: feed-in tariff only (Ritiro Dedicato)
  • CER: community energy sharing incentives only
  • RID_CER: both combined

Output metrics

For each building, celine-roi returns:

  • npv: net present value over 25 years at 5.5% discount rate (EUR)
  • irr: internal rate of return
  • payback_simple: years to recover investment from gross cash flows
  • payback_discounted: years to recover investment from discounted cash flows
  • tasso_autoconsumo: fraction of production consumed on-site (0-1)

Installation plan (dbt gold models)

On top of per-building estimates, dbt produces a 5-year installation plan:

pv_installation_ranking ranks all positive-NPV buildings by payback period (best first) and computes each building's cumulative percentile position.

pv_installation_plan assigns each building to an installation year using an adoption rate curve defined in the seed file data/installation_plan_params.csv:

year  cumulative_adoption_pct  degradation_rate
1     2.0                      0.0045
2     6.0                      0.0045
3     12.0                     0.0045
4     20.0                     0.0045
5     30.0                     0.0045

The best-return buildings are installed first. To change the rollout pace, edit the CSV and re-run task dbt:seed && task dbt:gold.

pv_installation_plan_summary aggregates per year with cumulative totals: new and cumulative buildings, kWp/MWp, production MWh (with degradation), investment EUR/MEUR, coverage %, weighted IRR, average payback.

REC substation analysis (dbt gold/rec_it)

Italian RECs operate under GSE primary substation (cabina primaria) coverage areas. The rec_it dbt models break down the estimation by substation:

rec_building_cabina spatially joins each building to its primary substation using gse_cabine_primarie geometry from the rec_it pipeline. Requires GIST indexes on both geometry columns for performance.

rec_cabina_opportunities joins the building-cabina mapping with the per-building ROI estimates.

rec_cabina_summary aggregates per substation: total buildings, capacity, production, consumption, self-consumption, grid export, investment, NPV, weighted IRR, payback, and building type breakdown.

rec_cabina_plan breaks down the 5-year installation plan by substation with per-year and cumulative totals.

Build with task dbt:rec after the main gold models are up.

Dashboard

The Streamlit dashboard (tools/dashboard.py) provides two views:

Province Overview — aggregate PV potential across Trentino: - Total KPIs: buildings, MWp, GWh/year, investment, NPV, IRR, payback - 5-year installation plan with capacity and investment charts - Building type breakdown (residential/commercial/industrial) - Financial distributions: IRR, payback, NPV, self-consumption - Energy balance: self-consumed vs grid export per yearly cohort

REC Substation Analysis — per-cabina primaria investment view: - Multi-select substations to scope a REC investment area - Community investment summary: total potential and aggregate NPV - ROI breakdown with explicit return on investment percentage - Revenue streams: self-consumption savings vs RID export revenue - Per-substation detail table with ROI % column - 5-year plan with cumulative NPV and ROI tracking - Side-by-side substation comparison

Launch with task dashboard.

Known limitations

  • Building classification relies on a simple area + floors heuristic. Overture Maps building_class is mostly NULL and num_floors is often missing. Misclassified buildings get wrong consumption estimates and tax treatment. Cross-referencing with OSM tags or cadastral data would improve accuracy.

  • The 20 kWp cap applies to all buildings. Large commercial/industrial rooftops could host bigger systems with different economics (no IRPEF deduction, IRES/IRAP tax treatment, lower EUR/kWp). A future version could model these separately.

  • Residential consumption uses a single fixed value. Real consumption varies widely based on occupancy, appliances, heating, and building efficiency. Heat pump adoption (5000-8000 kWh/year) is not modeled.

  • Production uses a flat regional specific yield. Per-building estimates via PVGIS or the Trentino Solar API would account for local shading, orientation and tilt but are too slow for batch processing.

  • CAPEX is flat per kWp. Real costs decrease with system size (economies of scale) and vary by installer, roof type, and mounting system.

Configuration

All estimation parameters are in flows/config.yaml. Key values:

estimation:
  specific_yield: 1100       # kWh/kWp/year
  panel_kwp_per_m2: 0.130    # kWp per m2 of footprint
  max_kwp: 20                # residential cap (IRPEF threshold)
  min_kwp: 3                 # minimum viable system
  capex_per_kwp: 1500        # EUR/kWp net of IVA
  regime: RID                # RID, CER, or RID_CER

  residential_consumption_kwh: 3500  # fixed per household
  residential_sizing_headroom: 0.20  # size system to 120% of consumption

  consumption_per_m2:        # area-based, non-residential only
    commercial: 100
    industrial: 150
    office: 80
    agricultural: 30

  classification:            # building type heuristic
    max_residential_area_m2: 200
    max_residential_floors: 3
    industrial_min_area_m2: 500