Typical Meteorological Year Data: Solargis Approach: Sciencedirect

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Available online at www.sciencedirect.

com

ScienceDirect
Energy Procedia 69 (2015) 1958 – 1969

International Conference on Concentrating Solar Power and Chemical Energy Systems,


SolarPACES 2014

Typical Meteorological Year data: SolarGIS approach


T. Cebecauer* and M. Suri
GeoModel Solar, Pionierska 15, 831 02 Bratislava, Slovakia

Abstract

We analyze factors to be considered when Typical Meteorological Year (TMY) is constructed from a time series data. We
discuss characteristics of the input data (time period, time frequency, quality), calculation methods (weighting of parameters,
coherence of parameters, selection criteria) and requirements of the resulting TMY to characterize average solar climate (P50
case) or a year with less-favorable solar resource (P90 or Pxx cases). Overview of the existing methods is presented, with
summary of their features, limitations and associated risks. SolarGIS method is presented, which overcomes limitations of the
older methods. We explain the underlying concept of uncertainty that is used for calculation of annual P90 or Pxx values for
Direct Normal Irradiation (DNI) and Global Horizontal Irradiation (GHI). SolarGIS method fulfills the criteria required in solar
industry, such as geographical representativeness, consistency of solar radiation and meteorological parameters, and yields a
good representation of the P50, P90 or any other Pxx cases.

© 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
© 2015 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer review by the scientific conference committee of SolarPACES 2014 under responsibility of PSE AG.
Peer review by the scientific conference committee of SolarPACES 2014 under responsibility of PSE AG
Keywords: Typical Meteorological Year; DNI; GHI; uncertainty, P50, P90

1. Introduction

Since its introduction in 1970s, the need for TMY was driven by limitations of computing power and storage,
while the applications focused on energy simulation of buildings and hot water systems. Stoffel et al. [1] summarize
the phases of development in the US, starting with the first method by Sandia [2], later by NREL with TMY2 [3]

* Corresponding author. Tel.: +421 2 49212491; fax: +421 2 49212422.


E-mail address: [email protected]

1876-6102 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer review by the scientific conference committee of SolarPACES 2014 under responsibility of PSE AG
doi:10.1016/j.egypro.2015.03.195
T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969 1959

and TMY3 [4]. Alternative methods were presented by e.g. Kalogirou [5] and Faiman et al. [6]. Originally, TMY
was derived from ground-measured or their combination with modeled data.
New requirements came with energy simulation in photovoltaics and concentrated solar power (CSP) systems.
Simulation of a CSP system is complex, because of the non-linearity of thermal systems, capacity limitations and
storage. Use of satellite-modeled solar resource data became standard. Especially in CSP, prior to generation TMY,
historical satellite-based time series are correlated with short-term ground measurements and they are adapted for
the specific conditions of a site. Choice of approach and input data for TMY should follow the requirements, for
which the data set is to be used.

Nomenclature

CDF Cumulative Distribution Function


CSP Concentrated Solar Power
DNI Direct Normal Irradiance
DIF Diffuse horizontal irradiance
GHI Global Horizontal Irradiance
KSI Kolmogorov Smirnoff Integral
RH Relative Humidity
RMSD Root Mean Square Deviation
TEMP Air Temperature at 2 m above Surface
TMY Typical Meteorological Year

2. Factors determining the most suitable TMY

Multiyear time series of solar radiation and meteorological data are transformed to a simpler data representation
with an aim to reduce data volume and speed-up energy simulation. Typical Meteorological Year (TMY) condenses
hourly or sub-hourly time series, typically representing more than 10 years of historical data, into one year. In this
way a significant reduction in number of data points by factor 10 and more can be achieved. Of course, such
reduction of data volume is not possible without loss of information contained in the original time series.
To meet the specific needs of energy simulation for a solar power plant, by the most suitable and representative
TMY data, the following factors have to be considered:
x Input data: origin, resolution, representativeness, accuracy and weighting;
x TMY generation approach: choice of representative period of time, statistical and construction criteria.

3. Input data

3.1. Data sources

TMY data are required by industry at any site of planned solar energy project development. Time series used for
construction of TMYs may come from various sources; the selection depends on data availability, temporal
coverage, quality and completeness. Two data sources are available; optimally they should be used in a
combination:
x Ground-measured solar and meteorological parameters;
x Solar resource from satellite-based models and meteorological parameters from meteorological models;
Solar measurements from high-accuracy and well-maintained sensors are supposed to be more accurate compared
to the modeled data. However, accuracy of ground measurements is often deteriorated by insufficient maintenance
of sensors and absence of quality control. Solar data should be quality-controlled before further use. Availability of
ground-measured data is very limited, and they do not have long enough and continuous time coverage. Thus the
only option is data from the satellite-based models, which offer site-specific information almost globally, for 12 to
20 or more years, with high spatial resolution (up to 3 km) and temporal resolution up to 15 and 30 minutes [7].
1960 T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969

Meteorological data should cover exactly the same period as solar resource data as the consistency and
dependency of all data within TMY should be preserved. This is very important because the actual performance of
solar energy technologies is directly dependent on combination of solar resource and parameters such as air
temperature, wind speed, wind direction and relative humidity. If long record of local meteorological measurements
is available, this type of data can be used. If such data are not available or it is incomplete, data from meteorological
models have to be used. Due to regional representation, and geographical smoothing, meteorological parameters
from models represent broader situation in the region rather than conditions of the local site.
Optimally, to achieve the best combination of characteristics (Table 1), data from models should be correlated
with local measurements and adapted for site-specific conditions.

Table 1. Input data sources


Data source Advantages Limitations
1 Ground-mounted High frequency measurements (seconds to minutes) Available only at a limited number of sites
instruments High accuracy, if best instruments are used and the Limited time coverage
meteo station is professional managed Quality issues if not properly maintained and
Redundant measurements enable more stringent quality-controlled
quality control Representation is limited by geography and the
level of data aggregation
2 Satellite-based solar Available almost globally Time step 15 and 30 minutes
models History of 12 to 20 or more years Lower instantaneous accuracy
Geographical stability Representation of the area (typically a grid cell
High data availability, gaps are filled by algorithms 3 to 6 km)
3 Meteorological data from Available globally for the same period of time as Time step of 60 minutes
meteorological models satellite data Representation of the area (typically a grid cell
Stability of data (reanalysis models) 30 km or more)
No measurement issues, no gaps

The choice of data for constructing TMY is typically limited. Four criteria are important:
x Geographical representativeness: data should represent the local or regional climate:
o Ground measurements in monotonous landscape and low terrain may reliably represent climate
within a radius of few tens of kilometers. In mountains, close to sea/water and industrial/urban
centers and in rapidly changing landscape the representativeness may be limited to few kilometers;
o Satellite-based solar radiation represents an area given by a grid cell of size 3 to 6 kilometers;
o Meteorological characteristics, such as temperature or humidity, derived from numerical weather
models may be represented by a grid cell with size of 30 or more kilometers, which may be
enhanced by more detailed modelling;
x Temporal representativeness: time representation of input time series data should be at least 10 years, to
reduce uncertainty from interannual weather variability. Optimally 15, 20 or more years of continuous time
series data should be used. All TMY weather parameters have to be available for the same period;
x Time step (granularity): typically hourly data are used, though time step of 1 to 15 minute is preferred,
especially in CSP modelling;
x Importance (weight) of parameters: end use of the data should be considered, allowing to control which
parameters (GHI, DNI, TEMP, etc.) will be represented with minimum loss at the cost of lower
representativeness of other parameters. TMY can be fine-tuned for specific applications.

3.2. Weighting solar and meteo parameters

Use of TMYs in solar energy simulation brought new requirements for higher preference of solar irradiation
compared to factors of lower importance. The aim is to achieve the best possible representation of especially GHI
(for PV) and DNI (for CSP), when compared to multiyear time series. The weights for parameters can be modified
according to their impact on performance of particular technology. Typically, higher weight is given to DNI in case
T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969 1961

that TMY is to be used for simulation of CSP and CPV technologies, and also for flat-plate PV tracking
technologies. If TMY is to be used for fixed-mounted PV technologies, the highest weight is given to GHI.
CSP and CPV technologies require a specially developed TMY, as use of suboptimal data may lead to distortions
in energy simulation. The risk of using a TMY, which is not optimized for CSP simulation, has not been quantified
yet, but the difference of mean annual value of GHI and DNI of suboptimal TMY compared to original time series
may be several percent.
Recently, for the US territory, NREL developed two new TMY products with full (100%) weight put either to
GHI or to DNI. Others, e.g. [8] or SolarGIS approach presented here, propose increased weight for selected solar
radiation parameters, leaving some influence to parameters, which have also control on energy conversion process
(e.g. diffuse irradiance and air temperature, Table 2). Moreover, in SolarGIS approach, we suggest that ensuring
optimum representativeness of TMY in different geographical conditions requires adapted weights.

Table 2. Weight assigned to solar and meteo parameters considered in various TMY approaches
Meteo parameter Sandia (TMY2) NREL (TMY3) Kalogirou Meyer et al. SolarGIS SolarGIS
[1] [3] [5] [8] (PV) (CSP/CPV)
Max. temperature of the dry bulb 1/24 1/20 1/32 1 — —
Min. temperature of the dry bulb 1/24 1/20 1/32 2 — —
Average temperature of the dry bulb 2/24 2/20 2/32 1 0.05* 0.04*
Temp. deviation of the dry bulb — — 1/32 — — —
Max. temperature of Dew point 1/24 1/20 — 2 — —
Min. temperature of Dew point 1/24 1/20 — — — —
Average temperature of Dew point 2/24 2/20 — 1 — —
Max. relative humidity — — 1/32 — — —
Min. relative humidity — — 1/32 — — —
Average relative humidity — — 2/32 — — —
Deviation of relative humidity — — 1/32 — — —
Wind speed max. 2/24 1/20 1/32 4 — —
Average wind speed 2/24 1/20 2/32 2 — —
Deviation of wind speed — — 1/32 — — —
Average wind direction — — 1/32 1 — —
Global irradiance 12/24 5/20 8/32 — 0.75* 0.23*
Direct irradiance — 5/20 8/32 85 — 0.70*
Diffuse irradiance — — — — 0.20* 0.03*

* Indicative weight, which may changes depending on the geographical conditions of a site

3.3. Time step (granularity) of data

Non-linear response of CSP systems, in conjunction with the transition states, system inertia and the presence of
thresholds, may introduce a source of substantial uncertainty in simulations, when DNI is used at different time step.
Conventional hourly time-step of TMY is not suitable for simulating a CSP system, as it overestimates the energy
production. An ideal time-step would be approximately 1 minute, which raises practical problems, in terms of
procurement of such data, its volume and existing simulation capabilities. In absence of local historical data with
this time-step, statistical methods have been proposed to develop DNI datasets with time-step of one minute from
satellite data, for example [9]. It is more practical to derive TMY with a longer time step. Hirsch et al. [10] suggest
that the simulation time step of 10 minutes results with errors of energy production of about 2% to 3%, relative to a
simulation at the time step of 1 minute. Satellite data are available in a time step of 15 and 30 minutes, which in turn
allows generating TMY in the same time-step, without any need for processing. More detailed data (1 to 10 minutes)
can be created by several approaches (Table 3):
1. Fusion of local measurements with satellite data allows adaptation of satellite data to the time step of the
1962 T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969

local measurements if they are available [11]. However, this method can be subjective, as it depends on the
quality, consistency and length of available ground measurements. By this method, the original
representation of high-frequency daily dynamics is altered in case of variable cloudiness. This method also
reduces coherence between GHI and DNI and in statistical distribution of values.
2. If only satellite data is available, time interpolation of cloud index from 15-minute (30-minute) to a shorter
time step. The disadvantage of this method is that it is not capable to represent events with broken clouds.
Therefore it is not suitable for time steps shorter than ten minutes.
3. Statistical post-processing of cloud index considers time and spatial variability of cloud index [12],
separately considered for GHI and DNI. Polo et al. [13] introduce synthetically generated variability into
data. The “intensity” of this variability is controlled by the value of cloud index and evaluated on the basis of
ground measurements in the region. The results well represent the measurements, but the method cannot be
applied in a global scale, because the generated variability functions are based on site-specific statistics.
4. Calculation of cloud movement vectors from consecutive images (method originally used in forecasting and
nowcasting) is used for simulation of cloud position in a finer time resolution. Such methods are
computationally demanding and are rarely used. This method does not resolve variability from small broken
clouds, which are below grid resolution of satellite data (typically 3 to 5 km).

Table 3. Summary of methods for high-frequency time resolution


Method Advantages Disadvantages
1 Use of original 15 (or No data manipulation is needed 15-minute (30-minute) time step is the limit
30) minute time step Preserves natural values in the original
available in satellite data satellite data set
2 Interpolation of 15- Simple and objective method Smoothing daily profiles in case of fast cloud changes, no
minute and 30-minute Interpolation to 10-minute time step is a additional variability
GHI and DNI into 10- good compromise, May not work well for time steps lower than 10 minute
minute values
3 Cloud index post- Potential benefits also for high frequency Computationally very demanding, little explored,
processing simulation (1 to 10 minute time step) geographically dependent
4 Fusion of ground May work only if long history of measured Subjective, needs expert knowledge
measurements with data exist (at least 3 to 4 years) Site-specific (not universal approach)
satellite data May not preserve consistency of all parameters and criteria
(as it is applied for each parameter independently)

The best practice for the CSP is to use a TMY with a maximum time step of 10 minutes, and strong weighting
applied to DNI. If local measurements are available, a sophisticated fusion technique can combine them with
satellite data. For general applications, the most robust, simple and objective approach is interpolation of GHI and
DNI data, even though it does not allow generating real 10-minute data intermittency. The transformation of 15- or
30-minute to 10-minute values does not have significant impact on overall statistics, especially in regions with low
occurrence of clouds.
15-minute time step may be found as a good compromise between a need for higher resolution data and increased
data volume, as it preserves the qualities of the original satellite time series. Moreover the coherence of solar
radiation with other meteorological parameters is rarely considered when high frequency radiation is introduced.
Some methods may result in significant inconsistencies between radiation and meteo data, thus having negative
impact on accuracy of energy simulation.

4. TMY approaches

Determined by its construction, the traditional TMY data set represents average (normal) year, by ignoring
extreme months. Such data product is often also called TMY P50.
The financial viability of a major energy project must be based on a bankable study. Such study evaluates
financial risks for the worst cases, in addition to the case of average (or "normal") years. The solar resource
assessment study has to highlight the probabilities of exceedance of minimum annual productions, with a confidence
T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969 1963

level of 90%, 95% or 99% (depending on the requirements of financial institutions). These probabilities are
statistically calculated from the available meteorological datasets, and correspond to the minimum threshold of
yearly solar radiation, which will be exceeded at least 90, 95 or 99% of the lifetime of the project. There remains a
relatively low probability that the absolute worst case can happen, and that the solar resource of a year is below this
threshold. This leads to creation TMY data sets representing less favourable solar irradiance conditions, which are
expected with different probabilities of occurrence, for example for P75, P90, P95 or other Pxx cases. The definition
of such conditions remains vague, and a wider consensus between professionals does not exist. The most typically
TMY P90 data set is required, describing low sunlight conditions that may occur once in ten years. The longterm
annual average value of GHI and DNI of such TMY is close to the P90 value, i.e. value which should be exceeded
for 90% of cases. The annual irradiation values at P90 are derived as combination of (i) uncertainty of model
estimate and (ii) uncertainty from interannual variability for one year (see Section 4.2).

4.1. TMY P50: average year

The most common approaches are based on the concatenation of months or days selected from multiyear time
series into one TMY dataset. Due to short and often non-continuous availability of solar measurements for location
of interest, the TMY approaches based solely on the use of measured data are not considered for solar energy
applications. Multiyear (10 years and more) time series are used for TMY, and such time series can be typically only
obtained from the modelled data (satellite models and meteorological models). In preparing TMY, selection of
individual months is typically controlled by one or more criteria of similarity of statistical characteristics:
1. In the simplest approach the sole criterion for searching the most suitable month is the minimum deviation
between DNI (GHI) monthly average and the longterm average for each particular month in the data
history. This method is repeated for each month in the year; the selected months are concatenated to the
final TMY, see e.g. [11]. The disadvantage is that the method may be applied only to one parameter (e.g.
DNI), and although the selected months in TMY are close to the long-term average, the distribution of
(sub)hourly values (histogram) in the chosen months may not well represent the distribution of values in
the original data. This method does not allow creating P90 or Pxx data sets.
2. Range of methods is based on the stepwise exclusion of the individual months, e.g. TMY3 method by
NREL [3]. In the first step, months with too different Cumulative Distribution Function (CDF) of daily
values of the selected parameter (e.g. DNI) are excluded. In comparison, the statistics by Filkenstein and
Schafer [14] is used, integrated for all weighted meteorological parameters, where weights control
influence of individual meteorological parameters. In the second step, only preselected subset of five
months is considered, and the further exclusion is based on the persistency criteria – searching for months
with presence of specific weather situations for several days in a row identified in original data. Finally
twelve months are selected, which have monthly averages close to the averages represented in time series.
This method does not allow creating P90 or Pxx data sets.
3. Method based on searching in a moving window over the time series to find sections of 365-days in the data
where the yearly value of DNI is close to the long-term average [15]. Time step of the moving window is
one day. This first stage typically results in finding several windows (each considering a continuous
sequence of 365 days) in the time series, for which the annual DNI value is close to the long-term average.
In the second stage, another criterion can be defined to choose the most appropriate window: e.g. one can
look at the best-possible match of monthly averages in the TMY compared to long-term monthly averages
calculated from the original time series, or it can be based on the analysis of cumulative Distribution
Function (CDF). The method description in the original paper by Hoyer-Click et al. [15] does not provide
details about how such criteria should be implemented. The advantage of this method is that it can preserve
some extreme situations as it practically selects the complete series of 365 days. The disadvantage is that
hourly and daily representation of the data for some months in a selected year can deviate from the average
month (having different sequence of hourly and daily values). The original method focuses only on DNI,
which can result in choosing a sequence of 365 days, where GHI does not adequately represent the
longterm average. The original method did not consider creating P90 or Pxx cases. This method can be
1964 T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969

extended by implementing additional criteria: more parameters can be considered (with different weights),
match of monthly averages and distribution function and it can be also used for TMY for Pxx cases.
4. Another optional method is to select a candidate month and to fit the statistical characteristics of that month
to those of original time series by manual replacing the individual days [15]. The disadvantage is
subjectivity of this method, and it becomes even more complex if more than one meteorological parameter
is to be considered at the same time.
5. Methods based on the comparison of normalized daily residuals of individual parameters from their
longterm averages may serve as a good alternative to the stepwise methods ([16, 17] and others). But a
discrepancy in yearly and monthly averages exists in the results, and should be corrected by rescaling or by
changing weightings for the most important factors.

4.2. TMY P90 and Pxx: year with less favourable solar irradiance conditions

Most of historical methods for creation of classical TMY do not allow for evaluating risks of low-irradiation
years and insufficient energy production, nor (a fortiori) the associated financial risks. Financial institutions
typically require a risk assessment described by the probability exceedance at 90% confidentiality, or "P90". Some
institutions are even more cautious and require P95, P98 or even P99. This has prompted the industry to obtain
TMY based preferably on P50 but also on P90.
It is in the interest of project developers to use most accurate satellite data, to run ground-measurement
campaigns at potential development sites, and to benefit from correlation of in situ measured data (short term) with
the satellite-modelled time series (long term).
Evaluation of uncertainty is essential. To operate with data at the same confidence level, uncertainty is assumed
here at 90% probability of exceedance. In creating TMY for P90 or Pxx case two aspects should be considered:
1. Uncertainty introduced by measurements and models:
x Uncertainty of the satellite-based solar model;
x Uncertainty of the ground instrument(s) [17];
x Uncertainty of site-adaptation approach, if applied.
2. Uncertainty due to interannual variability
x Considering weather fluctuation that may occur in any single year (this uncertainty is used for TMY Pxx
data generation);
x Considering weather cycles within a period of 20 to 25 years of power plant operation.
If no measurements are available, only uncertainty of the satellite model is considered. If local measurements are
also available, the uncertainty of the instruments and of the site-adaptation method is to be included [18]. If high
accuracy instruments are used, and well maintained, contribution of uncertainty of ground measurement to the
overall uncertainty is low.
The second element of uncertainty relates to the interannual variability of solar resource. Interannual variability
does not follow the Gaussian distribution, however due to limited data availability and limited geographical
knowledge, this simplification has to be accepted. In the generation of TMY for Pxx case, interannual variability for
any single year is assumed, to consider the statistically worst possible situation given by year-by-year weather
variation. It is to be noted that extreme volcano events (emitting large volume of ashes into stratosphere) are not
covered in the data history of last 20 years.
Methods for production of TMY P90 or Pxx cases are not many. Wey et al. [11] advocate an analysis of
percentiles characterizing interannual variability, for the selection of typical months, by analogy with the method
used for conventional TMY (TMY P50). All selected months are then concatenated into P90 TMY. However, this
approach is limited to one parameter only. They combine the measured and calculated data by the fusion method,
but in considering P90 they do not take into account uncertainty of estimate. This method also neglects the fact that
sum of monthly P90 values does not produce yearly P90 value.
T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969 1965

5. SolarGIS approach

5.1. SolarGIS method for P50

TMY generation results in reduction of data information value. SolarGIS method is designed to preserve climate
statistics in the TMY data as close as possible to original multiyear time series. The TMY P50 is generated with an
objective to match the P50 yearly DNI and GHI long-term averages. SolarGIS method jointly evaluates for each
month, in the original time series, two criteria that are described by a set of calculated similarity indexes: (i)
Similarity of monthly averages and (ii) Cumulative Distributed Function (CDF) generated from time series:
x Several parameters (most often DNI, GHI, DIF and TEMP) are evaluated in parallel with different
weighting using integrated similarity index.
x Months with best similarity index are then used for construction of TMY. The result is a selection of
months having good match of means and CDFs, especially for GHI and DNI (Fig. 1). The influence of
CDFs and means as well as individual parameters may be controlled by weighting.
x The annual average of such generated TMY has a maximum deviation within ±1.5 % from the longterm
average (usually below 1%). This deviation is removed in the final step of TMY construction, when the
annual averages of DNI and GHI are corrected by rescaling to gain zero difference (when compared to
multiyear time series).
Selection of the most appropriate month for each meteorological parameter results in a good statistical similarity
between its cumulative frequencies and a set of months available in the original climate dataset. Fig. 1 shows that a
TMY P50 defines a solar resource that is very close to the median (50% probability of exceedance of DNI and/or
GHI values). There is a good agreement between the measured average frequencies, calculated average frequencies
(on 20 years) and the frequencies of the resulting TMY P50. Interannual variability of the cumulative frequency
curves of DNI is obvious. It is to be reminded that the reduction of the full time series to one typical year also
removes this important information on the interannual variability. This method keeps all criteria under control and
yields the most representative data. It was extended also for creating TMY for P90 and other Pxx scenarios.

Fig. 1. Cumulative frequency curves of DNI (left) and GHI (right) (in W/m2) calculated from 20 individual years.
Red: multiyear time series; blue: TMY P50. Source data: PSA Almeria

5.2. SolarGIS method for P90

SolarGIS method for P90 (or any Pxx case) considers combined uncertainty of the estimate and interannual
variability of annual values. The method is looking for match of P90 yearly values (DNI and GHI) in TMY with
those calculated from Time Series and P90 uncertainties. The method can be employed (i) only on the basis of
modelled radiation data (obtained by the satellite model), or preferably (ii) using satellite data site-adapted by local
measurements.
1966 T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969

The principles of SolarGIS method for P90 (Pxx) are:


x Annual longterm P50 values of DNI and GHI are corrected (downwards) by their uncertainty to derive
annual P90 (or Pxx) values of DNI and GHI;
x Original data for individual months from full time series are ranked in descending order of total monthly
DNI (or GHI in case of TMY prepared for PV);
x Penultimate month from the list is selected to represent first guess of a candidate for TMY P90 (Pxx);
x Difference between preliminary-constructed TMY and derived annual P90 values is minimized by replacing
the individual months by their neighbors in the ranked set.
This procedure is iteratively repeated until convergence of TMY annual average and required P90 (Pxx) values is
reached. The selection of months is done similarly to TMY P50 (Pxx): the weights are used to prioritize some
parameters, but the main criterion is fit of resulting TMY annual value with the annual value in TMY Pxx derived in
first step. The coherence of DNI and GHI is preserved.

5.3. Example – TMY data for Almeria

Table 4 shows an example of TMY for P50, P75 and P90 for PSA Almeria site (latitude: 37.092°, longitude: -
2.358°). As an example, P90 yearly values are calculated by subtracting the combined uncertainty Uncertcombined
from P50 yearly value:
ܲ90 = ܲ50 െ ܷ݊ܿ݁‫ݐݎ‬௖௢௠௕௜௡௘ௗ

The combined DNI (GHI) uncertainty is calculated from the uncertainty of the SolarGIS model estimate
Uncertestim and from the interannual variability Uncertvariab:

ܷ݊ܿ݁‫ݐݎ‬௖௢௠௕௜௡௘ௗ = ට൫ܷ݊ܿ݁‫ݐݎ‬௘௦௧௜௠ ଶ + ܷ݊ܿ݁‫ݐݎ‬௩௔௥௜௔௕ ଶ ൯

Uncertainty due to interannual variability is derived from multiyear Time Series data and it gives an idea of
weather fluctuation. It is calculated statistically from the unbiased standard deviation of yearly DNI and GHI
summaries, multiplied by 1.28155 for P90 case (E\IRU3FDVH íWKXVLQWHUDQQXDOYDULability represents an
expected range of annual values at 80% probability of occurrence for P90 case, and 50% probability of occurrence
for P75 case. The lower boundaries of interannual variability and represent 90% probability (75% probability) of
exceedance, and they are used for calculation of the P90 (P75) annual values. This calculation assumes normal
distribution of yearly values, which may not work accurately in some climate regions or if extreme volcano
eruptions occur. Interannual variability, which may occurs in any single year, is considered (Table 4).

Table 4. Estimate of combined uncertainty, P75 and P90 values for PSA Almeria
Parameter Yearly TMY Uncertainty (yearly) [%] Yearly TMY value
value [kWh/m2] [kWh/m2]
Model estimate Interannual variability Combined
P50 P75 P90 P75 P90 P75 P90 P75 P90
GHI 1871 1.6 3.0 1.4 2.7 2.1 4.1 1831 1795
DNI 2090 3.7 7.0 3.2 6.1 4.9 9.3 1988 1896

The derived annual P50, P75 and P90 values are used for creation of TMY data sets using SolarGIS approach. It
is obvious that monthly averages of generated TMY cannot be exactly same as those calculated from multiyear time
series. Fig. 2 compares monthly summaries derived from multiyear time series with TMY data sets generated by
SolarGIS approach for P50, P75 and P90 case.
T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969 1967

Fig. 2. GHI (left) and DNI (right) monthly summaries: time series vs. TMY for P50, P75 and P90

In general P50 follows very well average weather patterns found in the time series, representing the years 1994 to
2013 for PSA Almeria. In 20 years time series rarely a month which fully resembles all statistical characteristics for
all parameters is found, thus smaller deviations between original data and resulting TMY P50 are inherent. They
stem from the nature of the construction process where criteria of similarity of monthly means and monthly CDFs
are used. Higher variability of the TMY P90 (P75) is a result of simplified approach, where only fit of annual Pxx
value is used as only criterion for TMY creation, regardless monthly similarity. This approach resembles real
situations: years with lower annual average (compared to long term annual average) usually have few months with
relatively good solar resource, while low annual value is a result of several months with extremely low sunshine.
SolarGIS method for TMY P50 generates robust and representative data sets. It has been developed for CSP,
CPV and PV industry, and goes beyond traditional concepts. The method aims to preserve long-term averages and
distribution of values for each month, thus eliminating outliers or other specific situations. Maintaining Cumulative
Distribution Function (CDF) of TMY data close to the one identified in the original time series is especially
important for energy simulation in CSP, because high and low values, exceeding certain thresholds, cannot be
always utilized by these technologies. In addition, the method maintains consistency between DNI and GHI
parameters. It allows adapted weighting the parameters (GHI, DNI, DIF and TEMP) to specific needs of individual
solar energy technologies, but is can be also used for constructing a ‘generic’ (cross-solar technology) TMY dataset.
The weights of individual parameters can be selected also with respect to specific geographical conditions, but such
analysis must be supported by a detailed expert examination of the results. Compared to other methods, SolarGIS
method allows for flexibility for different geographic (climate) regions. The method is extended to remove small
residual differences (less than 1.5%) of the yearly means, thus it maintains the same annual long-term average value
for time series and TMY.

6. Discussion

Several methods exist for TMY construction (Table 5). The methods differ in their capability to represent various
properties of the original time series data. Simple methods may result in disagreement of TMY and long-term data
statistics, and they typically do not preserve consistency of GHI and DNI values. Best methods consider both long-
term averages and distribution of values of more parameters. Most advanced methods allow for adaptation to
specific local conditions or/and needs of the specific solar energy technology, giving preference to selected
parameters. More sophisticated methods require experts capable to critically evaluate and fine-tune the outcomes.
SolarGIS TMY data have uncertainty, which is close to time series. However, it is to be underlined that TMY is
data product with the reduced information value. Therefore the following features of TMY are to be considered:
x Statistical information calculated from TMY may deviate from multiyear time series;
x Extreme events may not be preserved in TMY;
x Energy simulation based on TMY may differ from the simulation using multiyear time series.
1968 T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969

Table 5. Summary of TMY generation algorithms


Method Advantages Disadvantages
1 Minimum Simple method Does not consider Cumulative Distribution
deviation between Considers monthly averages Function (CDF)
monthly averages Only one parameter can be chosen
2 NREL TMY3 Considers CDF, persistency, and in the next step it Can remove important month in the 1st step
considers monthly averages Does not to consider fit to yearly long-term average
Maintains consistent irradiance components Does not have flexibility for different geographic
regions
A mixture of parameters is considered, whereas
solar parameters have low weights
3 Moving window Preserves the character of all data over the whole 365- Some months in a selected year can deviate more
(original method) days window from the long-term average
Consistent irradiance components Only DNI is considered, GHI may significantly
deviate from PXX value
CDF is not considered
4 Composition of Large flexibility of choosing days to construct a month Data manipulation may easy go out of control
months from days according to selected criteria Extremely time consuming
Unlimited choice of criteria Subjective method requiring solid expertise
5 Normalized Similar to method 2, more flexible in choosing criteria If focused on one leading parameter, it may not
residuals Flexible in choice and combination of criteria respect representativeness of other parameters
Rescaling to fit long term yearly averages may
affect harmony of solar components
Flexible but needs knowledgeable expert
6 SolarGIS method Considers several parameters in parallel, priority is given It is based on multiple but objective criteria, tools
to DNI, GHI and TEMP, weighting is flexible are flexible for implementation of knowledge and
CDF and monthly averages are close to time series due to expertise
the use of calculated indexes,
Residual differences are rescaled to long-term yearly
values calculated from time series
Enables fine-tuning depending on the geographic region
Consistent irradiance and meteorological components
10-minute values are interpolated from 15 and 30-minute
original data
Feasible for P50 and also for P90 and Pxx data sets

The advantage of the TMY data product is that it enables to simulate the solar energy system with a single year of
(sub)hourly data, which is a critical element in case that thousands of simulations are needed to optimize a power
plant design. On the other hand, the use of TMY may become dangerous in case of nonlinear energy CSP systems,
where for example energy generation starts only after a minimum intensity of solar radiation is reached.
Due to lack of standardization, the data delivered by various experts may differ in properties and
representativeness. Previous sections indicate that the use of TMY may trigger some difficulties, as it has reduced
information content than original time series.
It is in the opinion of many experts (including the authors) that the best strategy is to use multiyear time-series
(12 to 20 years of data, or more) in order to directly assess the effects of variability of the incident radiation on the
production of energy. Direct use of time series data allows risk analysis, which otherwise cannot be performed
completely, even when using TMY for special Pxx cases. The disadvantage of using multiyear time series is the size
of data files and longer computation time.
T. Cebecauer and M. Suri / Energy Procedia 69 (2015) 1958 – 1969 1969

Table 6. Elements determining quality and uncertainty of TMY


Best practices (lower uncertainty) Standard practices (higher uncertainty)
Length of data 15 or 20 years Less than 15 years
Accuracy of data Best available satellite data, rigorously validated, Satellite data based on less accurate concepts and
optimally site-adapted by local measurements less stringent validation or site-adaptation
Weighting Focus on DNI, GHI (DIF) and TEMP Focus on only one parameter (only DNI, ignoring
could be also WS, RH other solar components)
Other extreme is weighting of too many meteo
parameters with low weight to DNI and GHI.
Method Considers CDF + monthly averages + more parameters Averages or one parameter only
Time step Sub-hourly obtained by: Hourly
* Original satellite resolution (15 minute)
* Time Interpolation (from 15/30 minute to 10-minute)
* Assimilation of satellite and measured data
Type of TMY P50, P90, Pxx P50
Experience Experienced and knowledgeable expert Ad-hoc expert with limited knowledge and tools

References

[1] Stoffel T., Renné D., Myers D., Wilcox, S., Sengupta M., George R., Turchi C., 2010. Best practices handbook for the collection and use of
solar resource data. Technical report NREL/TP-550-47465, National Renewable Energy Lab., Golden, CO.
[2] Hall L., Prairie R., Anderson H., Boes E., 1978. Generation of Typical Meteorological Years for 26 SOLMET stations. Rapport technique
SAND78-1601, Sandia National Lab., Albuquerque, NM.
[3] Marion W., Urban K., 1995. User’s manual for TMY2s derived from the 1961-1990 National Solar Radiation Data Base. Rapport technique,
National Renewable Energy Lab., Golden, CO.
[4] Wilcox S., Marion W., 2008. User’s manual for TMY3 data sets. Technical Report NREL/TP-581-43156, NREL Lab., Golden, CO.
[5] Kalogirou S.A., 2003. Generation of Typical Meteorological Year (TMY-2) for Nicosia, Cyprus. Renew. Energy, 28, 2317-2334.
[6] Faiman D. et al., 2004. The Negev radiation survey. Trans. ASME J. Solar Energy Eng., 126, 906-914.
[7] Perez R., Cebecauer T., Šúri M., 2013. Semi-Empirical Satellite Models. In Kleissl J. (ed.) Solar Energy Forecasting and Resource
Assessment. Academic press. http://dx.doi.org/10.1016/B978-0-12-397177-7.00002-4.
[8] Meyer R., Beyer H.G., Fanslau J., Geuder N., Hammer A., Hirsch T., Hoyer-Klick C., Schmidt N., Schwandt M., 2009. Towards
standardization of CSP yield prognosis. Proc. SolarPACES Conf., Berlin, Germany.
[9] Beyer H.G., Fauter M., Schumann K., Schenk H., Meyer R., 2010. Synthesis of DNI time series with sub-hourly time resolution. Proc.
SolarPACES Conf., Perpignan, France.
[10] Hirsch T., Schenk H., Schmidt N., Meyer R., 2010. Dynamics of oil-based parabolic trough plants—Impact of transient behaviour on energy
yields. Proc. SolarPACES Conf., Perpignan, France.
[11] Wey E., Thomas C., Blanc P., Espinar B., Mouadine M., Bouhamidi M.H., Belkabir Y., 2012. A fusion method for creating sub-hourly DNI-
based TMY from long-tem satellite-based and short-term ground-based irradiation data. Proc. SolarPACES Conf., Marrakesh, Maroc.
[12] Perez R., Kivalov S., Schlemmer J., Hemker K.Jr., Hoff T. 2011. Parameterization of site-specific short-term irradiance variability. Solar
Energy, 85, 1343-1353.
[13] Polo J., Zarzalejo L.F., Marchante R., Navarro A.A., 2011. A simple approach to the synthetic generation of solar irradiance time series with
high temporal resolution. Solar Energy, 85, 1164-1170.
[14] Finkelstein J.M., Schafer R.E., 1971. Improved Goodness-of-Fit Tests. Biometrika, 58, 3, 641–645.
[15] Hoyer-Klick C., Hustig F., Schwandt M., Meyer R., 2009. Characteristic meteorological years from ground and satellite data. Proc.
SolarPACES Conf., Berlin, Germany.
[16] Festa R., Ratto C.F., 1993. Proposal of a numerical procedure to select Reference Years. Solar Energy, 50, 9-17
[17] Janjai S., Deeyai P., 2009. Comparison of methods for generating typical meteorological year using meteorological data from a tropical
environment. Appl. Energy, 86, 528-537.
[18] Suri M., Cebecauer T., 2014. Satellite-based solar resource data: Model validation statistics versus user’s uncertainty. ASES SOLAR 2014
Conference, San Francisco, 7-9 July 2014.

You might also like