Intraday spatiotemporal PV power prediction at national scale using satellite-based solar forecast models

TL;DR

This paper introduces a novel, country-scale framework for intraday spatiotemporal PV nowcasting and benchmarks seven models across Switzerland using a large, satellite-derived SSI dataset and thousands of PV installations. It combines satellite-based SSI forecasts (SolarSTEPS, IrradianceNet, SHADECast) and a physics-based IFS-ENS model with a station-specific ML irradiance-to-power converter to produce PV forecasts, allowing direct comparison to ground truth. The findings show satellite-based approaches outperform IFS-ENS at short lead times, with SHADECast providing the most reliable probabilistic forecasts and IrradianceNet delivering the strongest point-forecast accuracy, while high-elevation and high-variability conditions remain challenging. The work demonstrates the practicality of large-scale, probabilistic, spatiotemporal PV forecasting and highlights avenues for operational deployment and further improvement through bias correction, cloud-evolution modeling, and higher-resolution regional analyses.

Abstract

We present a novel framework for spatiotemporal photovoltaic (PV) power forecasting and use it to evaluate the reliability, sharpness, and overall performance of seven intraday PV power nowcasting models. The model suite includes satellite-based deep learning and optical-flow approaches and physics-based numerical weather prediction models, covering both deterministic and probabilistic formulations. Forecasts are first validated against satellite-derived surface solar irradiance (SSI). Irradiance fields are then converted into PV power using station-specific machine learning models, enabling comparison with production data from 6434 PV stations across Switzerland. To our knowledge, this is the first study to investigate spatiotemporal PV forecasting at a national scale. We additionally provide the first visualizations of how mesoscale cloud systems shape national PV production on hourly and sub-hourly timescales. Our results show that satellite-based approaches outperform the Integrated Forecast System (IFS-ENS), particularly at short lead times. Among them, SolarSTEPS and SHADECast deliver the most accurate SSI and PV power predictions, with SHADECast providing the most reliable ensemble spread. The deterministic model IrradianceNet achieves the lowest root mean square error, while probabilistic forecasts of SolarSTEPS and SHADECast provide better-calibrated uncertainty. Forecast skill generally decreases with elevation. At a national scale, satellite-based models forecast the daily total PV generation with relative errors below 10% for 82% of the days in 2019-2020, demonstrating robustness and their potential for operational use.

Figures (23)

• Figure 1: (a-c) Satellite-based SSI fields and (d-f) PV power output of the 6434 PV stations, observed on 6 August 2019 at three time stamps: 12:00 UTC, 13:00 UTC and 14:00 UTC. The black lines denote national borders. Note that the PV power output is normalized using the station-specific 95th percentile of the power time series.
• Figure 2: Empirical probability density function and cumulative density function (CDF) of the PV stations (a) elevation and (b) $P_{95}$ value.
• Figure 3: Two-year average of PV power output normalized with the station-specific $P_{95}$ value during (a) MAM, (b) JJA, (c) SON, and (d) DJF. The averaging is performed over timestamps between local sunrise and sunset times.
• Figure 4: Solar zenith and elevation angles computed over a full year at 30-minute resolution for two PV systems located in the vicinity of (a) Geneva and (b) Biasca. The correction factor represents the fraction of time since the previous timestamp during which the sun remains above the local horizon, ranging from 0 (entirely below the horizon) to 1 (entirely above the horizon). This figure was generated using the HORAYZON library Steger2022.
• Figure 5: PV power measurements, XGBoost predictions of PV production, and SSI observations at 15-minute resolution for stations S-GE (a–c) and S-TI (d–f) over three days that belong to the test set. SSI observations from the HANNA dataset are interpolated to the corresponding station locations. The PV power values are normalized with the station-specific $P_{95}$ value. The light-blue shaded area highlights the period when the sun is above the horizon, determined from local SZA using HORAYZON. The vertical black dashed lines indicate local sunrise and sunset times. The nMAE and nRMSE values are computed for each individual day and normalized using the station-specific $P_{95}$ value.