Marktstammdatenplotter

Animated choropleth maps of installed wind & solar capacity in Germany, driven by data scraped from the public Marktstammdatenregister (MaStR).

What it does

The MaStR registry contains every grid-connected electricity-generating unit in Germany — millions of solar panels, tens of thousands of wind turbines, plus hydro, biomass, gas and more. Each record carries an install date, capacity in kW, geographic coordinates, and a pile of enum-encoded metadata.

This repo:

1. scrapes the MaStR public JSON API,
2. decodes the enum-heavy rows into a clean PowerPlant dataclass (parser.py),
3. joins turbines to German county polygons extracted from OSM,
4. renders one choropleth PNG per month from 2000 to today, and
5. assembles the frames into an animated GIF with ffmpeg.

Repository layout

Pipeline overview

Three external sources flow into a single rendering loop. Cached artifacts (data-*.json, germany_kreise.gpkg) sit between scrape and analyze so the slow steps are run-once.

Module architecture

Scraping MaStR

The registry offers a full XML export, but the API filters server-side and returns JSON instead of XML, so this repo scrapes that instead:

seq 7 | xargs -P 4 -I{} curl --get \
  'https://www.marktstammdatenregister.de/MaStR/Einheit/EinheitJson/GetErweiterteOeffentlicheEinheitStromerzeugung' \
  --data-urlencode 'sort=' \
  --data-urlencode 'page={}' \
  --data-urlencode 'pageSize=25000' \
  --data-urlencode 'group=' \
  --data-urlencode 'filter=Energieträger~neq~\'2495\'~and~Energieträger~neq~\'2496\'' \
  --data-urlencode 'forExport=true' -o data-{}.json

OSM boundaries

County boundaries come from a Germany OSM extract (e.g. geofabrik.de). Filter to admin levels 4 (state) and 6 (Kreis), then export to GeoPackage:

osmfilter germany-latest.o5m \
  --keep-nodes="boundary=administrative and ( admin_level=6 or admin_level=4 )" \
  --keep-ways="boundary=administrative and ( admin_level=6 or admin_level=4 )" \
  --keep-relations="boundary=administrative and ( admin_level=6 or admin_level=4 )" \
  --drop-version --drop-author \
  -o=germany_admin_levels_4_6.osm

ogr2ogr -f GPKG germany_kreise.gpkg germany_admin_levels_4_6.osm \
  -sql "SELECT name, admin_level, boundary FROM multipolygons \
        WHERE boundary = 'administrative' \
        AND (admin_level = '6' \
             OR (admin_level = '4' AND name IN ('Berlin','Hamburg','Bremen')))" \
  -nlt MULTIPOLYGON -overwrite -nln multipolygons

Kreise are admin_level=6. The three city-states (Berlin, Hamburg, Bremen) sit at admin_level=4 and get pulled in by name.

Parser & enums

MaStR fields are numeric enum codes. parser.py decodes the six enums that matter for plotting, plus the .NET-style /Date(ms)/ timestamps.

Unknown codes resolve to None rather than raising — the registry adds codes over time, so the decoder must tolerate them.

Notebook flow

wind.ipynb loads cached JSON, builds a GeoDataFrame of turbines, joins to germany_kreise.gpkg, aggregates by county, and writes one PNG per month from 2000-01-01 to today.

Bins are pre-computed once on a representative date so the legend stays stable across all frames — otherwise the same color would mean different megawatts in January 2005 versus May 2026.

Animation

PNGs land in fig/ with names like wind-2007-04.png. ffmpeg insists on frame%03d.png, so the assembly script renames frames in a tmpdir, repeats the final frame ~120 times for a 4-second hold, and runs palettegen + paletteuse for a clean palette:

set -l file wind
set -l frames_to_repeat 120
mktemp -d | read -l temp_dir
and cp "$file"-*.png $temp_dir
and begin
    set -l i 1
    set -l last_frame_path ""
    for f in (ls "$temp_dir/$file"*.png | sort)
        mv $f (printf "%s/frame%03d.png" $temp_dir $i)
        set last_frame_path (printf "%s/frame%03d.png" $temp_dir $i)
        set i (math $i + 1)
    end
    set -l current_duplicate_index $i
    for j in (seq 1 $frames_to_repeat)
        cp "$last_frame_path" (printf "%s/frame%03d.png" $temp_dir $current_duplicate_index)
        set current_duplicate_index (math $current_duplicate_index + 1)
    end
end
and ffmpeg -framerate 30 -i "$temp_dir/frame%03d.png" \
    -vf "scale=-1:1200:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse=dither=none" \
    -loop 0 -y "$file".gif
and rm -rf "$temp_dir"

Interactive notebooks

Two marimo reactive notebooks ship with this repo. They render directly to standalone HTML, so the published docs site below contains a fully self-contained interactive map for each technology. Edit the source .py files locally to add your own controls.

Run locally

python -m marimo edit pv.py     # reactive editor
python -m marimo edit wind.py
python -m marimo run pv.py      # read-only app
python -m marimo export html pv.py -o docs/pv.html

Sample renders (real MaStR data)

Generated from the open-MaStR Zenodo full registry dump (cutoff 2025-02-09) merged with the MaStR API delta through May 2026. Wind: full registry (~42 k turbines, every Kreis covered). PV: full registry (~4.86 M plants, including all rooftop and balcony solar — no power threshold). BESS: top 200 k storage units by installed power.

Per-unit size-bin distribution

Same size-bin axis as the BESS chart above. Wind sits almost entirely in the 1-10 MW bin (75 GW of the 79.7 GW total). PV is spread across 10 kW – 100 MW — utility ground-mount (1-10 MW + 10-100 MW) carries roughly equal weight, with ≥ 100 MW reserved for a handful of mega-parks.

Installed capacity by Bundesland

21 years of wind growth + YTD 2026

21 years of PV growth + YTD 2026

Energy-type mix

National installed capacity per Energietraeger, snapshot 2026-05-01. Numbers reflect the open-MaStR Zenodo full registry merged with the API delta through May 2026 — Wind, PV, and storage all use the complete registry, no power threshold.

Build-out per Bundesland

Monthly cumulative wind + PV per state, 2000 → May 2026. Bayern (22 GW), Niedersachsen (19 GW) and Brandenburg (18 GW) lead. Sharp inflection around 2010 = PV boom; the post-2020 wedge is mostly utility-scale solar in the former-lignite belt.

PV by installation type

Which sub-type of plant carries the GW? Top 200 k PV plants split by ArtDerSolaranlageId. Free-standing utility parks (16 k plants) hold ~ 43 GW; building-mounted commercial rooftops (184 k plants) hold ~ 30 GW. The residential rooftop and balcony long tail under 49 kW is excluded — that's where the count is, but not the GW.

PV orientation

Capacity-weighted distribution of panel facing + tilt across the top 200 k PV plants. South-facing dominates at ~ 62 % of installed capacity; east-west flat-mount (a utility-park staple) is the strong second at 14 %. Trackers stay rare — only 0.5 GW of the captured fleet. Most plants tilt 0–19° (flat, utility-park footprints).

Wind fleet age + repowering signal

Top panel: installed MW per commission year (blue) vs. decommissioned MW per removal year (red, plotted below the axis). The decommission wave kicks in around 2019 as the first big EEG cohort reaches its 20-year payback boundary — ~ 0.6 → 0.75 GW removed each year since 2023.

Bottom panel: mean per-turbine rotor MW by commission year. The fleet has upsized from ~ 0.5 MW per turbine in the late 90s to ~ 7 MW in 2026 — a 15× growth at the unit level, driven by the new offshore 15 MW class.

Battery storage (BESS)

The German electricity registry tracks storage units under Energieträger=2496 (Speicher) — a separate slice from generation. The numbers below cover batteries only (Stromspeichertechnologie = "Batterie"). Pumped-hydro (PSH) is reported in its own section below. Scraped the top 200 000 units by Bruttoleistung; that captures everything from grid-scale Li-ion projects down to ~10 kW home batteries.

Snapshot 2026-05-01 (batteries only, active): 193 599 units · 6.49 GW · 9.4 GWh.

HSS / CSS / LSS — three-sector view

Heimspeicher (HSS, < 30 kWh) — residential PV+battery, 5-15 kWh LFP units typical. Gewerblicher Speicher (CSS, 30 kWh – 1 MWh) — C&I PV+battery, UPS, peak-shaving. Großspeicher (LSS, ≥ 1 MWh) — grid-services + standalone utility BESS + Pumpspeicher.

Battery sub-sector breakdown (active, 2026-05-01):

• HSS < 30 kWh: 176 411 units · 2.47 GW · 2.54 GWh
• CSS 30 kWh – 1 MWh: 16 655 units · 0.68 GW · 1.20 GWh
• LSS ≥ 1 MWh: 526 units · 3.34 GW · 5.65 GWh

Pipeline: another ~ 8 GW of LSS Li-ion under construction or permitted (LEAG GigaBattery Jänschwalde 1 GW, Boxberg 400 MW, ECO POWER 1–6 = 6 × 300 MW, Westfalen 538 MW…). Combined with PSH the German storage stack reaches ~ 25 GW by ~ 2028.

Size-bin breakdown — GW + GWh per per-unit size class

Battery-only view. Residential (10-100 kW) carries ~ 192 k units but only 2.9 GW / 3.5 GWh. The MW-scale grid-services bins (10-100 MW + 1-10 MW) carry most of the GW + GWh.

BESS power per Kreis

BESS energy per Kreis

Duration distribution

Cumulative growth

GWh trajectory by sector — historical + planned pipeline

9 years of battery storage growth (2017 → May 2026)

Pumped-hydro storage (PSH)

PSH is reported separately from batteries everywhere serious (battery-charts.de, BVES, EASE, EU SET-Plan) because it's a fundamentally different technology — multi-decade lifecycle, multi-hour duration, mechanical-hydro chemistry instead of electrochemical. Lumping it with Li-ion BESS hides both stories.

Snapshot 2026-05-01 (active): 41 sites · 6.48 GW · 927.5 GWh. Median site duration ~ 8 h — orders of magnitude longer than the 1-h Li-ion median. Goldisthal alone (4 PSS units × 265 MW each / 9.64 GWh each) accounts for 1.06 GW + 38.5 GWh.

PSH energy per Kreis

Distribution + duration

Top sites

The "other electricity storage" tail (Wasserstoffspeicher 26 units, Druckluft 10, Schwungrad 1) sums to 0.0 GW / 0.006 GWh combined — pilot installations, not yet material.

Capacity density (MW/km²)

Absolute capacity makes big Kreise look impressive even when their per-km² intensity is low. Normalising by Kreis area reshuffles the ranking: Dithmarschen and Nordfriesland top wind at 1.1–1.6 MW/km²; the city-Kreise of Straubing, Amberg and Memmingen top PV because they host utility parks on small footprints.

Largest individual plants

Top 10 by single-unit capacity (Bruttoleistung) in this scrape, snapshot 2026-05-01. Wind list shows the new Nordseecluster + Iberdrola Windanker 15 MW turbines — all offshore, so Kreis is empty. PV list captures every solar park ≥ 100 MW: Schafhöfen (200 MW, Bayern), SILUX 1/2/4 + EnBW Gottesgabe + SunInvest all clustered in the former-lignite belt around Leipzig + the Oder-Spree border.

Top 10 PV plants

Top 10 wind turbines (single-unit capacity)

Raw data: largest-plants.json.

Capacity added during 2024

New plants registered between 2024-01-01 and 2024-12-31, aggregated per Kreis (Wind + PV ≥ 49 kW from this scrape). Total in this slice: ~13.4 GW — close to the public BNetzA tally of 16 GW once you add back the long tail of sub-49 kW rooftops.

Top operators

Top 30 owners by combined wind + utility-scale PV capacity, snapshot 2026-05-01. Names are normalised (legal-form suffixes stripped) but not consolidated across parent companies — most offshore wind farms appear as standalone project vehicles, which is why the top of the chart is dominated by single-farm LLCs rather than utilities.

Offshore wind

Germany's offshore wind fleet at 2026-05-01: 1 732 turbines, 10.4 GW total — 8.6 GW in the Nordsee, 1.8 GW in the Ostsee. The StandortAnonymisiert field carries a sea label ("Nordsee…" / "Ostsee…"), but the actual Laengengrad / Breitengrad coordinates are real (verified across all 1 909 offshore rows). Offshore points still fall outside every Kreis polygon, so they don't contribute to per-Kreis aggregations — we group by operator instead.

Top Kreise by installed capacity

Sortable, filterable table of all 434 German Kreise (snapshot 2026-05-01), driven by the same MaStR scrape behind the maps above. Click any column to sort; type in the filter box to search by Kreis name or Bundesland. PV column uses the top-50 000 utility-scale plants (≥ 200 kW); Wind # and PV # are turbine / plant counts active at the snapshot date.

Bulk downloads

Cleaned, joined-to-Kreis snapshots are regenerated by the weekly CI workflow. parquet is recommended for analysis (zstd-compressed, types preserved); csv.gz for tools without parquet support.

Data © Bundesnetzagentur (MaStR) and OpenStreetMap contributors (Kreis polygons, ODbL-1.0). The compressed snapshot omits owner contact details — names only, no PII.

PowerPlant dataclass

Defined in parser.py. Fields after decoding:

Gotchas

• Hamburg's geometry contains a national park as a phantom part. The notebook removes "Nationalpark Hamburgisches Wattenmeer" by part-id 2. Verify the part-id still matches if you regenerate the GeoPackage from a newer OSM extract.
• Capacity units. Bruttoleistung is in kW; the notebook divides by 1 000 000 to express county totals in GW for the legend.
• Offshore coordinates are real, despite the label. Records flagged WindAnLandOderSeeId == 889 carry a StandortAnonymisiert string ("Nordsee…" / "Ostsee…") and a real Laengengrad / Breitengrad. The original notebook draws them in synthetic sea rectangles anyway — that workaround was based on a wrong assumption about anonymisation; you can drop it if you want point-accurate offshore rendering. Spatial join to Kreis polygons still excludes them (no Kreis covers open sea).
• Timezone discipline. parse_dotnet_date returns tz-aware UTC datetimes. The notebook strips tz before comparing to date-only filters — keep that conversion local. Do not propagate tz-naive values back into other code.

Troubleshooting