You’re a developer or engineering firm who is working on a large solar project and you’re deciding which solar resource dataset to use. You know you need to use a high-quality resource dataset to get the project across the line. How do you make a good choice on the resource data, which will be the backbone of your project?
Three key questions will help you make a wise choice:
- Are hourly data available, or only monthly means?
- How old is the dataset?
- What is the dataset’suncertainty?
If you can answer these questions about the resource data used in your projects, then you’re already on the path to making better choices then many. Based on these answers and how much risk you are willing to live with, you can make an informed tradeoff between schedule, accuracy, and cost. This article focuses on the first question.
Long-term monthly mean meteorological data is sometimes used as a basis for predicting the yield of photovoltaic systems, since such data is widely available at little or no cost. Also, software tools like PVsyst and Plant Predict allow users to input monthly mean data, and then generate from this a year of synthetic hourly data that can be used as an input to photovoltaic yield modeling.
At very early stages of project design, this can be a quick way to get indicative numbers. The problem with this approach isthat – depending on the location and type of project - it can add too much resource risk, even in the early prospecting stages of project development. This is especially true for tracking PV plants and locations - such as India - with high resource variability.
In a recent Vaisala study, PVsyst was used to compare the P50 energy results derived from using multiple years of Vaisala’s 3TIER Services hourly meteorological data (Vaisala, 2017) to those derived using two types of synthetic years based on the same data: a PVsyst synthetic year (hereafter: PVsyst-SYN) and a Typical Meteorological Year (TMY). The same underlying resource dataset was used, and the only difference in the projects was the temporal resolution.
Modeling was performed for ten Megawatt-scale photovoltaic projects in six different countries including India. For each project, simulations were run for two configurations: an equator-facing fixed tilt orientation and one with horizontal single-axis East-West trackers with backtracking.
PVsyst was used to run all simulations, with only the meteorological data varying between the three approaches.
- Full timeseries approach: The hourly timeseries for the ten project locations covered roughly 20 years of individual simulations run separately in PVsyst to derive the corresponding annual yield. The full timeseries yield estimate (P50) was then taken to be the median of the individual annual yields.
- PVsyst-SYN approach: In this approach, the long-term monthly means of global horizontal irradiance (GHI) are first used to stochastically derive synthetic hourly GHI data as described by Meteotest in 2017.
- TMY approach: Vaisala creates TMY datasets using an empirical approach that selects four-day samples from the full timeseries to create a “typical year” of data with 8760 hours, while conserving the monthly and annual mean of GHI. The process is iterated until the annual means of all solar variables in the TMY dataset match the means of the full timeseries to within 1% or less.
Finally, the simulated yields from the PVsyst-SYN and TMY approaches were compared to the P50 yields from the full timeseries approach for each of the ten PV systems, for fixed and tracking configurations.
For all project types and locations, we saw significantly more deviation from the PVsyst-SYN approach. Notably, it was difficult to predict whether the bias was high or low – this means it cannot be treated as a consistent known bias.
The deviation between the energy yields from the long-term hourly timeseries and the TMY files relates to how well the TMY averages match the long-term dataset. If the TMY was not well matched for GHI we would expect these energy yield biases to be larger. Vaisala’s TMY creation process is designed to minimize these deviations for that very reason.
For tracking plants, the average difference in the hourly vs monthly average approach was about 2%, but the maximum difference was over 7%!For tracking plants, the monthly averagePVsyst-SYN approach has too much uncertainty to use even in the early prospecting stages of project development. A 7% difference in energy from predicted to actual production is more than enough to make or a break a project; thus, we recommend hourly data at all project phases if you’re developing tracking plants.
The news is better for fixed PV. The energy differences between the models was less than 1%, and the maximum difference was around 2%. That is a deviation that’s generally acceptable at early project stages. In later development stages, you should incorporating hourly timeseries data to reduce the uncertainty, especially if you are in a competitive financing situation.
A Proactive Approach
So as a proactive project developer or engineer how can you make sure you are making the best resource choices for your projects? Vaisala and other providers cover most geographies with satellite-based datasets at and hourly or even sub-hourly resolution at a reasonable cost.
If you have received a report from an engineering firm and are wondering what resource methodology they have used in order to understand the accuracy of the estimates, you should look past the name of the data provider. Many prominent engineering firms use the low-cost monthly means but not the hourly time series. Any time the words “synthetic” or “generated” are used in regards to the resource data, that is an indication that hourly values are derived and not native to the resource timeseries.
Given the increase in availability and the decrease in costs for hourly timeseries we would encourage others in the industry to drop the synthetic data creation practice so we can all build projects we will be proud of.
Authors: Gwendalyn Bender, Sophie Pelland Ph.D,Louise Leahy Ph.D. and Rajni Umakanthan
Vaisala, Seattle (U.S.A.) and Bangalore (India)
- Font Size
- Reading Mode