High-quality observations for improved seasonal predictions. Implications for the wind energy sector

Abstract

Multiple initiatives are being implemented to mitigate and, in the worst-case scenarios, adapt society to climate change. The vast majority consider renewable energies key to accomplishing a necessary transition from fossil fuels to clean energies. The electricity system, in particular, is facing a significant transformation, being it more dependent on renewable production and, subsequently, on meteorological factors like wind speed or solar radiation. The prediction of anomalies of meteorological variables is well- established and trusted from minutes to days ahead, and so is the amount of renewable generation. Beyond those time scales, seasonal predictions start to produce beneficial results in anticipating the amount of generation months in advance, but their quality is still far from that offered by weather forecasts. In this regard, the climate community is advancing towards better seasonal predictions, both from the perspective of climate modelling and its post-processing.

To further increase the value of seasonal predictions, climate services have recently appeared to make climate information ---sometimes deemed challenging to digest--- more understandable and practical for non-experienced users. Climate services facilitate the integration of seasonal predictions into the renewable industry. The wind power industry, for example, employs seasonal predictions not only to advance the future availability of the wind resource but also to schedule maintenance activities in wind farms. A better understanding of the opportunities of seasonal predictions allows wind energy users to identify gaps and report specific needs. This PhD thesis looks into those user needs to improve the quality of seasonal predictions for wind speed.

More specifically, the enhancement of seasonal predictions is achieved from the perspective of wind observations. We first focus on wind records measured at tall meteorological towers, a non-standard type of climate data widely used within the wind industry. We identify, retrieve and collect climate records from 222 tall tower locations distributed worldwide. After unifying the data format and performing an exhaustive quality control, specifically designed for this type of wind data, we release the dataset under the name of The Tall Tower Dataset. The data collection is made publicly accessible through a data web portal. We later explore reanalysis datasets to quantify how they differ from the true observed wind speeds. We consider five global reanalyses and describe their agreements and discrepancies in representing surface wind speeds. By comparing reanalysis data against winds from The Tall Tower Dataset, we conclude that representativeness errors in reanalyses can be large sometimes, to the extent not to trust gridded estimates in specific areas. We also conclude that ERA5 shows the closest wind speed estimates to those observed at the tall towers.

Once wind observations are characterised, and their quality is ensured to be sufficiently high to produce robust results, they are used to enhance seasonal predictions. The hybrid seasonal forecasts provided in this work allow predicting near-surface wind speeds at a point scale ---e.g. wind farm location. Those forecasts rely on the information of the large-scale atmospheric circulation, summarised in the state of the four main Euro- Atlantic Teleconnections. In general, hybrid predictions show skill at lead times two and three, while dynamical predictions do not. Another aspect that is improved is the skill

assessment of seasonal predictions. We illustrate the strong dependency of the score estimates, namely the Brier Score, on the choice of the observational reference. This has implications in, for example, the selection of the best prediction system among a set of possible candidates. To solve this issue, we consider two methodologies already proposed in the literature and apply them to seasonal predictions for wind speed. We evaluate their strengths and weaknesses to end up recommending the use of the observation-error-corrected scoring rules.

Les prediccions estacionals són una eina força útil per tal d’anticipar variacions climàtiques futures i prendre les accions oportunes per tal de mitigar els seus possibles efectes negatius millor. La industria eólica és un usuari potencial d’aquestes prediccions: la quantitat de producció renovable futura depèn fortament de les anomalies de velocitat de vent. Malauradament, les prediccions estacionals de què disposem no tenen una qualitat equiparable a la dels pronòstics meteorològics, i la seva naturalesa probabilística les fa difícils d’entendre per a usuaris inexperts. Aquesta tesi doctoral té com a objectiu principal la millora de les prediccions estacionals de vent, focalitzant-se en aquelles necessitats específiques que els usuaris de la indústria eólica han reportat. Aquest objectiu s’assoleix mitjançat un ús adequat d’observacions de vent d’alta qualitat.

Primerament, ens centrem en dades de vent mesurades en torres meteorològiques altes, un tipus de dades climàtiques que no són estàndard però que s'utilitzen àmpliament dins de la indústria eòlica. Identifiquem, col·lectem i netegem dades climàtiques de 222 torres altes localitzades arreu del món, publicant-les sota el nom de The Tall Tower Dataset. També s’avaluen dades de reanàlisi, a fi de quantificar la diferència d'aquestes respecte de les dades reals observades. Comparant les dades de reanàlisi amb les dades del Tall Tower Dataset, concloem que els errors de representativitat són grans en els reanàlisi, fins al punt de no poder confiar en aquestes estimacions de vent en àrees específiques.

Aquestes observacions són seguidament utilitzades per tal de millorar les prediccions estacionals. Les prediccions estacionals híbrides proporcionades en aquest treball permeten predir la velocitat del vent a prop de la superfície a escala puntual (per exemple, a un parc eòlic), i permeten millorar la qualitat de les corresponents prediccions dinàmiques. Un altre aspecte que s'ha millorat és l'avaluació de l'skill de les prediccions estacionals. S’Il·lustra la forta dependència de les estimacions dels valors d'skill amb l'elecció de la referència observacional, un efecte no desitjable en diversos contextos. Per solventar-ho, considerem dos metodologies que ja han estat proposades en la literatura i les apliquem a les prediccions estacionals de velocitat de vent, avaluant els seus punts forts i febles.