From Hourly Irradiance Discrepancy to Day-Ahead Photovoltaic Energy: Evaluating Free Weather APIs for Urban Energy Management

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Freely accessible weather APIs are an attractive input for photovoltaic (PV) power prediction, yet how their reference-based error carries through—under temporal aggregation—into day-ahead PV energy is rarely quantified before model development. This study audits eight freely accessible weather-data services—seven forecast services and the Meteostat observational archive—and finds that, for the endpoints, subscription tiers, and collection period evaluated here, only two of the forecast services expose hourly global horizontal irradiance (GHI), while a third advertised irradiance field returns empty—itself a material result for municipal integrators. The two irradiance-capable services are propagated through a common physical PV model of the 548 kWp east–west rooftop plant being built on a university campus in Bratislava, Slovakia, against independent references (CAMS irradiance; NASA POWER temperature and wind). Throughout, “error” denotes discrepancy against the reference, not measurement truth. Over a common 62-day spring–summer window, the day-ahead daily-energy mean absolute error (MAE, relative to mean reference daily energy) is 10–11% for both sources; Open-Meteo’s winter-inclusive own window raises its value to about 13%. On the matched window, Open-Meteo has the lower hourly GHI RMSE (121 vs. 128 W·m−2), yet this advantage does not carry through to day-ahead energy—the point-estimate ordering even changes—and neither matched-window difference is statistically resolved. Because daily aggregation rewards low bias over low scatter, this ordering change is already present in daily GHI energy—before the PV conversion—so hourly irradiance accuracy alone does not determine day-ahead energy. Propagated input-data auditing is therefore an advisable step before ML-based PV forecasting and day-ahead scheduling of urban distributed PV.

Links and sources

• PDF from original source

Comments