Forecast
How much solar will a site produce?
Choose a point and a plant size, then press Get forecast.
The roof-mounted case scales our model by 1,441.6 / 1,501 (expert-review values): derived, not a model run. Expert review gives about 1,440 to 1,500 for rooftops.
| Value | kWh per kWp | Range | Period | Source | Label |
|---|---|---|---|---|---|
| PVGIS-ERA5 long-term mean, Hyderabad, 17 deg tilt, 14% losses, free-standing | 1,501.0 | 1,418.0 to 1,548.0 | 2005-2023 | SME reproduced (not re-read here) | [checked-correct] SME |
| PVGIS-ERA5, roof-mounted (building) setting | 1,441.6 | 2005-2023 | SME reproduced (not re-read here) | [checked-correct] SME | |
| Global Solar Atlas (satellite), optimal tilt 21 deg | 1,589.0 | long-term | SME read (not re-read here) | [checked-correct] SME | |
| PVGIS-ERA5 as held on disk, Hyderabad, 17 deg, 14% losses | 1,456.4 | 1,418.0 to 1,521.1 | 2019-2023 | pitch/data/pvgis_1kwp/hyderabad.csv | checked here (local file) |
| Our model, Hyderabad, annual yield | 1,579.7 | Nov 2025 to Oct 2026 | pitch/data/pvmodel_1kwp/hyderabad.csv column P_W_per_kWp_calibrated (hourly sum, 8,760 h; equals target_w in results/test_predictions_monthly.parquet) | checked here (local file) | |
| Our model, mean of the 13 TGSPDCL points, annual yield | 1,604.0 | 1,558.9 to 1,638.7 | Nov 2025 to Oct 2026 | results/test_predictions_monthly.parquet column target_w, sum per city then mean of 13 (the audit's 1,604) | checked here (local file) |
| Our model, Hyderabad, calendar 2025 | 1,545.4 | 2025 | pitch/data/pvmodel_1kwp/hyderabad.csv | checked here (local file) | |
| Our model, Hyderabad, calendar 2024 | 1,451.3 | 2024 | pitch/data/pvmodel_1kwp/hyderabad.csv | checked here (local file) | |
| Our model, Hyderabad, calendar 2023 | 1,568.7 | 2023 | pitch/data/pvmodel_1kwp/hyderabad.csv | checked here (local file) | |
| Our model, Hyderabad, Nov 2025 to Oct 2026, scaled to the roof-mounted case (x 1441.6 / 1501) | 1,517.1 | Nov 2025 to Oct 2026 | arithmetic on model_hyd_nov25_oct26; not a model run | derived here |
- Hyderabad, 17 degree tilt, 14% losses, PVGIS-ERA5: long-term 1,501 kWh per kWp, 1,418 to 1,548 by year (2005-2023). Checked-correct by expert review; the series is not held here.
- Roof-mounted setting 1,441.6, about 4% lower: the free-standing default is optimistic for rooftops. Global Solar Atlas (satellite, 21 degree tilt) 1,589. Both checked-correct by expert review.
- Our model, Nov 2025 to Oct 2026: Hyderabad 1,579.7 kWh per kWp; 13-point mean 1,604.0 (range 1,558.9 to 1,638.7); checked on local files. Above the PVGIS band, close to the satellite product.
- The model uses a 0.954 factor fitted on 2019-2021, the lowest-yield years. Against PVGIS it reads +2.0% high in 2022 and +2.6% high in 2023 (six-point average); Hyderabad alone +2.1% and +3.1%. Fit file not held; bias checked on local files.
- Expert ratios differ: our earlier test found model/PVGIS 0.981 (2019) and 1.031 (2023) for Hyderabad; the expert review has 0.986 and 1.043, basis unknown. Both rise with year.
- High yield means more export, so it is the cautious side for hosting capacity, but it overstates energy and savings.
| Lead | Nowcast | Day-ahead | Same as last hour | Change |
|---|---|---|---|---|
| 1 h | 22.5 | 28.9 | 98.1 | 22.2% lower |
| 2 h | 27.1 | 28.9 | 183.9 | 6.3% lower |
| 3 h | 28.5 | 28.9 | 254.1 | 1.6% lower |
Mean error, W per kWp, against a physics model, not measured output.
Scored on days not used to fit. Live use needs a meter feed; 15-minute steps are not validated.
- Days compared
- 14
- Mean actual
- 4.59 kWh/m2
- Mean forecast
- 4.54 kWh/m2
- Forecast minus actual
- -1.1%
- Daily error (MAE)
- 0.34 kWh/m2
- Hourly error, sunny hours
- 40.6 W/m2 (182 h)
- 60 days to 9 Nov 2026: forecast minus actual
- +0.6% (60 days)
| Day | Actual kWh/m2 | Forecast kWh/m2 | Forecast minus actual |
|---|---|---|---|
| 9 Nov 2026 | 5.64 | 5.66 | +0.3% |
| 8 Nov 2026 | 5.39 | 5.58 | +3.5% |
| 7 Nov 2026 | 5.17 | 5.12 | -0.9% |
| 6 Nov 2026 | 5.02 | 5.03 | +0.2% |
| 5 Nov 2026 | 4.47 | 3.72 | -16.8% |
| 4 Nov 2026 | 3.72 | 3.17 | -14.8% |
| 3 Nov 2026 | 3.51 | 3.60 | +2.6% |
| 2 Nov 2026 | 3.59 | 4.51 | +25.6% |
| 1 Nov 2026 | 4.96 | 4.57 | -7.9% |
| 31 Oct 2026 | 5.01 | 4.72 | -5.8% |
| 30 Oct 2026 | 5.01 | 5.01 | +0.2% |
| 29 Oct 2026 | 4.92 | 4.19 | -14.9% |
| 28 Oct 2026 | 2.19 | 2.88 | +31.3% |
| 27 Oct 2026 | 5.65 | 5.78 | +2.3% |
Compares Open-Meteo's day-before sunlight forecast with its archive of what happened (satellite-based, not a ground station). It rates the weather forecast only; this is not metered rooftop output. Sources: irradiance_openmeteo_tgspdcl/hyderabad.csv (actual) and prev_day1_tgspdcl/hyderabad.csv (day-before forecast), Open-Meteo, CC BY 4.0. The newest days of the archive are provisional (E11). Daily total = sum of 24 hourly values; days with a missing hour show not held.
Retrained monthly (as live) or trained once on 2024. Scored versus a physics model, not versus measurement (archive weather, W per kWp, daylight hours). No metered rooftop output is held.
Coverage target is 80%. Scored versus a physics model, not versus measurement (cannot be checked yet, because no metered rooftop output is held).
Live forecast from Open-Meteo (read-only) and the shipped model; if the call fails, the stored example is shown and says so. Accuracy not yet measured against real output; results are estimates versus a physics model, not versus measurement.
Live nowcast (last 15 minutes / last hour): not built; it needs a meter feed.
| Model | MAE | RMSE | Bias | Daily energy error kWh/kWp/day | P10 to P90 coverage | Hours |
|---|---|---|---|---|---|---|
| New model, retrained monthly | 31.22 | 57.54 | -1.1 | 0.267 | 83.8% | 115,719 |
| Earlier run A | 38.24 | 67.46 | -7.5 | 0.332 | no band | 115,719 |
| Raw physics on the forecast | 39.07 | 72.81 | -6.4 | 0.356 | no band | 115,719 |
| Same time yesterday | 43.59 | 84.43 | 0.0 | 0.419 | no band | 115,719 |
Coverage target is 80%. Scored versus a physics model, not versus measurement (cannot be checked yet, because no metered rooftop output is held).
| Window | Model | MAE | Bias | Coverage |
|---|---|---|---|---|
| 2025-01..2025-09 (feed = ICON) | New model, retrained monthly | 34.18 | -10.1 | 81.8% |
| 2025-01..2025-09 (feed = ICON) | Same time yesterday | 43.46 | +0.1 | no band |
| 2025-01..2025-09 (feed = ICON) | Earlier run A | 44.56 | -18.7 | no band |
| 2025-01..2025-09 (feed = ICON) | Raw physics on the forecast | 50.43 | -21.1 | no band |
| 2025-10..2026-11 (feed switched) | New model, retrained monthly | 29.17 | +5.1 | 85.3% |
| 2025-10..2026-11 (feed switched) | Raw physics on the forecast | 31.17 | +3.8 | no band |
| 2025-10..2026-11 (feed switched) | Earlier run A | 33.84 | +0.4 | no band |
| 2025-10..2026-11 (feed switched) | Same time yesterday | 43.68 | -0.1 | no band |
Not proven that the feed switch caused the bias change between the windows; weather models, archive source and a wet 2025 monsoon are not separated. Cannot be checked yet.
| Model | Area MAE W/kWp | Bias | Error correlation | P10 to P90 coverage | Band |
|---|---|---|---|---|---|
| New model, retrained monthly | 28.19 | +1.0 | 0.341 | 77.2% | Below the 80% target: the band is a little narrow. |
| New model, trained once on 2024 | 29.98 | -6.5 | 0.341 | 72.0% | Below the 80% target: the band is a little narrow. |
| Earlier run A | 35.80 | -6.8 | 0.374 | 74.6% | Below the 80% target: the band is a little narrow. |
| Raw physics on the forecast | 36.46 | -6.4 | n/a | no band | No band (point forecast only). |
| Point | New model, retrained monthly | Raw physics on the forecast | Earlier run A |
|---|---|---|---|
| gadwal | 29.1 | 36.7 | 35.8 |
| hyderabad | 32.8 | 42.1 | 41.1 |
| mahabubnagar | 31.8 | 39.9 | 38.8 |
| medak | 30.2 | 38.0 | 37.3 |
| medchal | 32.0 | 40.4 | 39.6 |
| nagarkurnool | 31.5 | 39.6 | 38.9 |
| nalgonda | 29.6 | 36.6 | 35.9 |
| sangareddy | 32.1 | 40.9 | 39.7 |
| siddipet | 32.7 | 39.9 | 39.7 |
| suryapet | 30.1 | 36.6 | 36.0 |
| vikarabad | 32.8 | 41.3 | 39.7 |
| wanaparthy | 31.0 | 38.2 | 37.6 |
| yadadri | 30.2 | 37.8 | 37.1 |
Live forecast for one TGSPDCL point and plant size, plus stored day-ahead accuracy. Power values are MODEL estimates versus a physics model; the weather check is REAL (Open-Meteo, not metered output). No live nowcast.
- Open-Meteo weather actuals, 13 TGSPDCL points (hourly) REAL 2019-01-01 to 2026-11-09, last date 9 Nov 2026. Source: data/irradiance_openmeteo_tgspdcl/*.csv (Open-Meteo, CC BY 4.0).
- Open-Meteo day-before forecasts, 13 TGSPDCL points (hourly) REAL 2024-01-19 to 2026-11-09, last date 9 Nov 2026. Source: data/prev_day1_tgspdcl/*.csv (Open-Meteo previous-runs API, CC BY 4.0).
- PVGIS 1 kWp hourly output (published model reference) MODEL not a dated file. Source: data/pvgis_1kwp/*.csv.
- Open-Meteo live forecast (called when you ask, LIVE badge) REAL not a dated file. Source: Open-Meteo API (CC BY 4.0) via backend/app/integrations/open_meteo/.
- Not held: Metered rooftop solar output. Not public and not downloadable (PVGIS stops at 2023; no metered source was found). Forecast accuracy is versus a physics model, not versus measurement.
- Not held: Ground-measured sunlight (GHI). No ground station file is held. The Open-Meteo actuals are satellite/reanalysis-based; NIWE or IMD stations would be the check.