One of the main objectives of Project SINAG is to create a forecasting model for solar PV power output in the Philippines using remotely sensed data. And did you know, you can use various irradiance data to forecast solar PV output?
In forecasting time series, you can use the past behavior of a variable you want to predict, commonly referred to as the Box-Jenkins [1] approach, often interchangeably with Autoregressive Integrated Moving Average (ARIMA). ARIMA [2] is a three-step process [3] which considers the past behavior and error terms of a variable you want to forecast, provided that the time series data are made stationary.
An extension of ARIMA, the seasonal ARIMA (SARIMA) [4] accounts for the seasonality and trend of the time series. Meanwhile, SARIMA with exogenous variables (SARIMAX) [5] accounts for other variables you want to consider in forecasting your variable of interest.
In our IIS paper, we used the SARIMAX approach to evaluate the difference between forecasted solar PV outputs using various irradiance data from the following sources:
Himawari 08, shortwave radiation (SWR) [6]
ERA5, surface solar radiation downwards (SSRD) [7]
Fengyun-4A (SSI) [8]
PAG-ASA ground weather stations
Results show that using the irradiance data from PAG-ASA yields better forecasting accuracy at 9.32% (MAE) compared to SSRD at 9.39%, SSI at 9.62%, and SWR at 9.56%.
You can read more about our IIS paper here.
References:
[3] P. Kennedy, A Guide to Econometrics, 3rd ed. Cambridge, Mass: MIT Press, 1992.