Operational and historical forecast

Overview

Solargis forecast services produce two distinct types of forecast data: operational forecasts and historical forecasts. Understanding the difference between them is essential for correctly interpreting forecast output, evaluating forecast accuracy, and integrating time-series data into solar energy workflows.

Historical forecasts are predictions of solar irradiation, meteorological parameters, and PV or wind power output (PVOUT / WOUT) that cover the period from a defined point in the past up to the current moment ("Now"). Every historical forecast was once an operational forecast. As time progresses, operational forecasts transition into historical forecasts — the underlying data does not change, but the availability of reference data makes accuracy evaluation possible.

A key principle distinguishing the two types is evaluability:

• Operational forecast accuracy cannot be evaluated — the reference data does not yet exist.

• Historical forecast accuracy can be evaluated — the reference data is already available, either as Solargis reference or monitor data, or as measured data provided by the customer.

Operational vs historical forecasts

Operational forecasts are predictions issued from the current moment ("Now") into the future. They cover irradiation, meteorological parameters, and power output. Because the events they describe have not yet occurred, no reference data exists against which to compare it.

Historical forecasts cover the period from the past up to the "Now" moment. They were once operational forecasts. Because the forecasted period has already elapsed, reference data is available and forecast accuracy can be evaluated.

The figures below illustrate this distinction:

Figure 1: Operational forecast service, Solargis reference data, and D1 historical forecast divided by the "Now" line. "Now" moment is 10th December 2024 00:00 (local time in Tokyo).

Figure 1 shows historical forecasts (red), reference data (blue), and operational forecasts (black) separated by the "Now" line. The red arrow indicates the "day-ahead" (D1) operational forecast — the PVOUT forecast for 11th December 2024, issued at the "Now" moment (10th December 2024 00:00). At the time of issuance, forecast accuracy cannot be evaluated because the reference data does not yet exist.

After 2 days (48 hours), the situation changes:

Figure 2: Operational forecast service, Solargis reference data, and D1 historical forecast divided by the "Now" line. "Now" moment is 12th December 2024 00:00 (local time in Tokyo).

The operational "day-ahead" forecast indicated by the red arrow in Figure 1 has become a historical "day-ahead" forecast, indicated by the red arrow in Figure 2. The data itself is identical — the only difference is that the reference data is now available and forecast accuracy can be evaluated. The red arrow in Figure 2 answers the question: "What was the day-ahead PVOUT forecast for 11th December 2024 issued at the Figure 1 'Now' moment (10th December 2024 00:00)?"

Data delivery and Time series structure

Operational and historical forecasts differ not only in evaluability but also in how their time-series data is structured and delivered.

When operational forecasts are delivered, each successive day in the time series represents a different forecast horizon (leadtime). The forecast horizon changes as each new day is issued.

When historical forecasts are considered, each single time-series maintains the same forecast horizon throughout. This consistent structure makes historical forecast data suitable for systematic accuracy evaluation and benchmarking.

Figure 3 illustrates this structural difference:

Figure 3: Operational and historical forecasts divided by the "Now" line, showing the D1 forecast horizon applied consistently across the historical period.

Note: Figure 3 shows short examples for illustrative purposes. In practice, operational forecasts can extend up to 14 days ahead, and historical forecasts can extend from the beginning of 2020 up to the current "Now" moment.

The visualization below further illustrates the concept of successive daily forecast requests, showing how D+0, D+1, and D+2 horizons shift each time a new request is issued:

Figure 4: Successive daily GHI forecast requests showing how forecast horizons D+0, D+1, and D+2 shift with each new issuance date (July 4–6, requested at 6:00 a.m.).

Note: Operational forecasts with longer forecast horizons — from 4 days-ahead up to 14 days-ahead — are not stored after delivery due to the large data volumes involved. Operational forecasts with shorter forecast horizons are stored and, once they become historical, can be accessed upon request by both internal and external stakeholders.

Further reading

• Managing forecast accuracy expectations

• Origin of the data for Monitor and Forecast

• Validation of Forecast data

• Evaluating forecast accuracy - best practices