# Solar forecast URL: https://pro-weather.com/docs/features/solar-forecast > An hour-by-hour power forecast for your own solar panels, built from weather models scored against your station''s solar sensor, with an API for evcc, EMHASS and Home Assistant. If you have solar panels, a battery or an electric car, the useful question about tomorrow is how many kilowatt-hours your roof will make and when. The solar forecast answers it for your own panels, hour by hour for the next two days, and hands the numbers to the software that schedules your charging. It is part of the Pro plan. Set it up under **Setup > Solar forecast** in your site's dashboard. ## How it works 1. Every hour, Pro Weather fetches the sunshine forecast (solar radiation on a flat surface) from several weather models for your station's location. 2. It checks each model's earlier forecasts against what your station's solar sensor actually measured, and gives the models that got your sky right more say. That is the part a generic solar forecast cannot do: it has no sensor on your roof. 3. It turns the result into power for each of your panel arrays: the sun's position, the angle your panels face, the air temperature (hot panels make less), your system losses and your inverter's limit. The forecast says **corrected** once your solar sensor has scored the models for a couple of days. Until then, and for a station without a solar sensor, every model counts the same. It still works; it just has not learned your sky yet. What your solar sensor cannot know: shade from trees or chimneys, and panels that need a wash. It sees the sky, not your roof. Send what your panels actually make from Home Assistant and the forecast learns that too (see [Teach it your roof](#teach-it-your-roof)). ## Setting up your panels Add one row per roof face, up to four: - **kWp**: the peak power of that array, from your installer's paperwork or the panel count times the panel rating. - **Tilt**: 0 for flat panels, 90 for panels on a wall. A typical pitched roof is 30 to 45. - **Azimuth**: where the panels face. 0 is south, -90 east, 90 west, 180 north. South-east is -45. - **Losses**: wiring, dirt and the inverter. 14 % suits most systems; lower it if the forecast keeps coming in under what your inverter reports. - **Inverter kW**: the most the inverter can put out. Leave it empty if you do not know. An east-west roof is two rows, one facing -90 and one facing 90. ## Connect evcc [evcc](https://evcc.io) uses a solar forecast to decide when to charge your car. Create a token under **Setup > Solar forecast**, then add this to `evcc.yaml` with your token in place of ``: ```yaml tariffs: solar: - type: custom tariff: solar forecast: source: http uri: https://pro-weather.com/api/solar headers: - Authorization: Bearer jq: .slots | tostring interval: 1h ``` ## Connect EMHASS or a script [EMHASS](https://emhass.readthedocs.io) and most battery schedulers accept a list of watts per time step. With a 60-minute optimization time step, pass the `value` of each slot: ```bash curl -s -H "Authorization: Bearer " https://pro-weather.com/api/solar \ | jq '{pv_power_forecast: [.slots[].value]}' ``` and post that to EMHASS's `dayahead-optim` action. Node-RED, Home Assistant automations and your own scripts can read the same URL. ## Home Assistant The Pro Weather Solar Forecast integration puts the forecast in Home Assistant's Energy dashboard and gives you sensors for automations. Install it through [HACS](https://hacs.xyz) as a custom repository: 1. In HACS, open the three-dot menu, choose **Custom repositories**, add `https://github.com/LNNRT-STN/ha-pro-weather` as an **Integration**, then download **Pro Weather Solar Forecast** and restart Home Assistant. 2. Under **Settings > Devices & services > Add integration**, pick **Pro Weather Solar Forecast** and paste your token. 3. Under **Settings > Dashboards > Energy**, edit your solar production and tick the Pro Weather entry as its **Solar production forecast**. The integration adds power now, energy this hour and next, energy today, remaining today and tomorrow, and today's peak. "Today" follows your Home Assistant time zone. A **Corrected by station** sensor shows whether your solar sensor has scored the models yet. If you regenerate the token, Home Assistant asks for the new one. ### Teach it your roof The forecast knows the sky over your station, not the tree that shades your panels after four or the dust on them. Your inverter does. In the integration's options (**Settings > Devices & services > Pro Weather Solar Forecast > Configure**), pick **Solar production sensor**: the energy sensor in kWh that counts what your panels make, the same one the Energy dashboard uses for solar production. Once an hour the integration sends each finished hour of the last day to Pro Weather. From those hours Pro Weather learns one number for your site: how much of the forecast your panels actually deliver. It uses bright hours only, and, where your station measures sunshine, only hours whose sky the forecast got right, so a cloudy surprise does not count against your roof. Hours with a power cut, snow on the panels, a full battery that switched the panels off, or an inverter or export limit holding the output flat are left out. After 24 good hours the forecast starts using it, and it keeps learning over the last 30 days, so it follows the seasons and a panel clean. **Setup > Solar forecast** shows the number once it is learned, for example "your panels make 92 % of what the forecast expected". It never goes below 50 % or above 130 %: a gap that big means the kWp, tilt or azimuth is wrong, and the forecast should show that rather than hide it. Changing your panels starts the calculation over from the hours already sent. A battery that is often full while the house uses only part of the sun can still make the number read low, because your inverter then produces less than the roof could. If that is your setup, keep an eye on it in late spring. For [Predbat](https://springfall2008.github.io/batpred/), the energy today and tomorrow sensors carry a half-hourly `detailedForecast` attribute in the same shape as the Solcast integration's. Point `pv_forecast_today` and `pv_forecast_tomorrow` in Predbat's `apps.yaml` at those two sensors. Its `pv_estimate10` and `pv_estimate90` are the cautious and bright case described under [The API](#the-api), so Predbat's pessimistic weighting has a real range to work with (from integration version 0.2.0). ### Without the integration A REST sensor gives you the rest of today's forecast for automations, with nothing to install. In `secrets.yaml`, store the whole header value: `proweather_solar: "Bearer "`, then: ```yaml rest: - resource: https://pro-weather.com/api/solar headers: Authorization: !secret proweather_solar scan_interval: 1800 sensor: - name: Solar forecast rest of today unit_of_measurement: kWh value_template: "{{ ((value_json.daily.values() | list)[0] / 1000) | round(1) }}" ``` A REST sensor cannot feed the Energy dashboard's forecast line; only an integration can. ## The API `GET https://pro-weather.com/api/solar` with the header `Authorization: Bearer ` returns the next 48 hours: ```json { "generatedAt": "2026-09-24T09:12:03.120Z", "issuedAt": "2026-09-24T09:00:00.000Z", "corrected": true, "station": { "name": "Ardooie", "timezone": "Europe/Brussels" }, "arrays": [{ "name": "Roof south", "kwp": 5.2 }], "scale": { "active": true, "value": 0.92, "samples": 212 }, "slots": [ { "start": "2026-09-24T09:00:00Z", "end": "2026-09-24T10:00:00Z", "value": 2140, "wh": 2140, "ghi": 512, "p10": 1310, "p90": 2780 } ], "wh_hours": { "2026-09-24T09:00:00Z": 2140 }, "daily": { "2026-09-24": 14800, "2026-09-25": 21300 }, "daily_p10": { "2026-09-24": 9600, "2026-09-25": 13900 }, "daily_p90": { "2026-09-24": 18900, "2026-09-25": 26100 } } ``` - `slots`: one per hour, `value` the average power in watts, `wh` the energy in watt-hours, `ghi` the sunshine forecast behind it in W/m². Times are UTC. `p10` and `p90` are the cautious and bright case in watts: one hour in ten is expected to come in below `p10`, and one in ten above `p90`. The band is sized from how far your own station's solar sensor has landed from the forecast; a station without a sensor borrows the width measured at other stations. At night, and on a forecast from before the band existed, both equal `value`. - `wh_hours`: the same energy keyed by the start of each hour. - `daily_p10`, `daily_p90`: each day's hourly `p10` and `p90` added up. A whole day at every hour's cautious case is rarer than one day in ten, so treat `daily_p10` as a floor for a battery reserve, not as a typical bad day. - `daily`: watt-hours per day in your station's timezone, starting with the rest of today. - `scale`: what your measured production has taught the forecast. `value` is already applied to every number above; it is 1 while `active` is false, and `samples` is how many hours it rests on. - A new forecast is ready once an hour; polling every 15 to 60 minutes is plenty. More than 60 requests an hour from one address get a `429`. Errors: `401` for a missing or unknown token, `403` when the site's plan does not include the solar forecast, `409` when no panels are set up or the station has no location. Creating a new token stops the old one immediately. ### Sending production from your own script The Home Assistant integration does this for you. Anything else that knows your hourly production can post it with the same token: ```bash curl -s -X POST https://pro-weather.com/api/solar/production \ -H "Authorization: Bearer " -H "Content-Type: application/json" \ -d '{"hours": [{"start": "2026-09-24T09:00:00Z", "wh": 2140}]}' ``` - `start` is the start of a finished hour in UTC, on the hour. `wh` is what all your panels together made in that hour, in watt-hours. - Up to 168 hours per request, none older than 7 days. Sending an hour again replaces it, so resending the last day every hour is fine and fills any gap. - The whole request is refused with `400` and the bad hour named if any hour is wrong, for example not finished yet or more than your panels could make. `413` above 32 KB, `415` without the JSON content type, and more than 30 requests an hour from one address get a `429`. - The answer says how many hours were accepted and the `scale` as it stands.