3. ASPECTOS SOCIORRELIGIOSOS
3.2. RELIQUIAS Y SANTOS LUGARES
The comparison of the correlations obtained for the extreme indices characterising the low wind
speed values (q10 and f bq10clim) and those for
the high wind speed values (q90 and f aq90clim)
shows that the former produce noisier correlation maps than those extreme indices for the higher
wind speed values. The differences between
these two different sets of indices are evidence of the asymmetry in the predictability of the climatological wind speed distribution relative to
the six-hourly wind speed values. The 10th and
90th climatological percentiles that have been
used for the creation of f bq10climand f aq90clim
do not match the probability density function of ERA-Interim and this translates into differences in their associated indices.
The extreme indices considered exhibit similar positive and significant correlation in both oceanic and tropical regions over land such
as northern South America, northeastern
Brazil, western Africa and southeastern Asia. Extra-tropical regions also show positive and significant correlation, as it has been found for
North America. The four extreme indicators
show similar correlation to the mean wind speed, although they are generally smaller for the extremes than for the mean wind speed. Some exceptions are found in the tropical Atlantic and central South America where the
q10 and f bq10clim show higher correlations
than the seasonal mean and also in northeastern
Brazil where q90 and f aq90clim display higher
correlations than the mean wind. This result is in agreement with previous studies (Hamilton et al.
2012;Pepler et al. 2015;Bhend et al. 2017) that
have shown that the predictability of the extreme indices derived from the seasonal forecasts is a consequence of the inherent potential skill of the mean.
6.5 Conclusions
Seasonal forecasts tailored to the wind energy sector represent an innovation in the use of climate information to better manage the future
variability of wind energy resources. The
minimum level of quality required for climate predictions will highly depend on the decision to be made, the vulnerability of the sector at a particular time and location, and the variability
of the wind resource. Wind energy users
have traditionally employed a simple approach that is based on an estimate of retrospective
climatological information. Instead, seasonal
forecasts can better support the balance between wind energy demand and supply. Here it has been shown that the ECMWF System 4 seasonal forecasts of near-surface wind speed have skill in several regions where there is substantial installed wind power. Although the most skilful regions are concentrated in the tropics, there are some extra-tropical areas where the seasonal predictions could provide an added value to the reference climatology.
The relative merits of different techniques for the statistical bias adjustment of ensemble forecasts to address different aspects of the
Figure 57 – Pearson correlation of the 10-m wind speed forecasts of a) mean wind speed value, b)10th percentile, c)90th percentile, d) fraction of time under the10th climatological percentile and e) fraction of time above the90th climatological percentile from ECMWF System 4 and ERA-Interim reanalysis in
winter (DJF). The predictions have been initialised the 1st of November for the period of 1981-2012. Hatched regions indicate correlation significantly (one-tailed t-test at the 95% confidence level) larger than zero.
6.5. Conclusions
forecast error have been illustrated. The most important gain in forecast quality for seasonal predictions comes through the increase in their skill and reliability, the latter being a critical aspect of the forecasts from the user perspective because it guarantees the trustworthiness of the probabilistic predictions. This work reveals that calibration is necessary because it produces an improvement in both skill and reliability, making this technique essential for seasonal predictions to be usable.
However, bias adjustment methods are not able to improve the misrepresentation of the physical mechanisms leading to low levels of skill. For that reason, a reconstruction approach based on empirical and dynamical forecasts has been
applied. As it has been illustrated in chapter
4, ENSO teleconnections with wind speed are different in the seasonal predictions and in the observations for northern North America. When the seasonal predictions of 10-m wind speed are reconstructed from the empirical relationship between the Niño-3.4 index with the 10-m wind speed combined with the seasonal predictions of the Niño-3.4 index, an improvement of the correlation values in comparison with that for
the raw predictions is found. This result
indicates that the misrepresentation of the ENSO teleconnections to the 10-m wind speed is responsible for the low skill in that region. Hence, in the case that wind energy users are interested in the seasonal predictions of 10- m wind speed in that area, the reconstruction methodology generates more useful information than that produced directly from the seasonal
forecast system.
Future work will be based on the combination of bias adjustment approaches with reconstruction
methodologies, which take into account
information of the large-scale circulation,
to improve the skill of the seasonal forecasts of wind speed in relevant regions for the wind industry. The combination of bias adjustment methods with large scale circulation patterns such as ENSO is referred to as process- conditioned bias adjustment, and it has been
recently used by Manzanas and Gutiérrez
(2018) to adjust the systematic errors affecting the ECMWF System 4 seasonal forecasts of precipitation in northwestern Peru.
The description of the current capabilities of the seasonal prediction systems to produce accurate wind speed seasonal forecasts requires that the information about the mean wind speed predictability is complemented with information
about the tails of its distribution (i.e. low
and high wind speed values in a season). The potential skill of the extreme wind speed indicators is very similar to that illustrated for the mean wind speed, which shows that the predictability of the wind speed is mainly related to the seasonal forecast systems capacity to reproduce the variability of the mean wind speed. Bias adjustment is not essential for the wind speed indices based on climatological thresholds
( f aq90climand f bq10clim) because these indices
are defined from the 10thand 90thclimatological
percentiles computed from the predicted and
reference distributions separately. However,
absolute thresholds (q10 and q90) require a bias adjustment stage to reduce the systematic errors. Two different strategies could be followed to
overcome this problem. On one hand, the
application of a bias adjustment approach such as the quantile mapping, which corrects the full wind speed six-hourly probability distribution,
and then to compute the 10thand 90thpercentiles
based on the bias adjusted wind speed. On
the other hand, the implementation of bias
adjustment over the wind speed extreme indices, although this approach has the problem that
the consistency between the 10th and 90th
percentiles derived from the six-hourly wind speed distribution cannot be guaranteed. These considerations should be taken into account in the case the seasonal predictions of these indicators have to be delivered in an operational way, and they should be further explored in the future.