Northerly wind trends along the Portuguese marine coast since 1950

1 downloads 0 Views 6MB Size Report
Abstract. Wind is a marine coastal factor that is little understood but has a strong interaction with biological productivity. In this study, northerly wind trends in ...
Northerly wind trends along the Portuguese marine coast since 1950

Francisco Leitão, Paulo Relvas, Fernando Cánovas, Vânia Baptista & Alexandra Teodósio Theoretical and Applied Climatology ISSN 0177-798X Theor Appl Climatol DOI 10.1007/s00704-018-2466-9

1 23

Your article is protected by copyright and all rights are held exclusively by Springer-Verlag GmbH Austria, part of Springer Nature. This eoffprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to self-archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com”.

1 23

Author's personal copy Theoretical and Applied Climatology https://doi.org/10.1007/s00704-018-2466-9

ORIGINAL PAPER

Northerly wind trends along the Portuguese marine coast since 1950 Francisco Leitão 1

&

Paulo Relvas 1 & Fernando Cánovas 1 & Vânia Baptista 1 & Alexandra Teodósio 1

Received: 28 September 2017 / Accepted: 19 March 2018 # Springer-Verlag GmbH Austria, part of Springer Nature 2018

Abstract Wind is a marine coastal factor that is little understood but has a strong interaction with biological productivity. In this study, northerly wind trends in three regions of the Portuguese coast (Northwestern: NW, Southwestern: SW, and Southern: S) were analyzed. Two datasets with long-term (ICOADS: 1960–2010) and short-term data (Satellite: 1989–2010) were used to complement one another. The study revealed the northerly wind yearly data to be non-stationary and highly variable between years. Overall, the northerly wind intensity increased throughout the 1960s regardless of the area and dataset. Between 1960 and 2010, the northerly wind increased at a linear rate of 0.24, 0.09, and 0.15 m s-1 per decade in the NW, SW, and S coastal regions, respectively. The rate was higher in recent decades (1988–2009), with the wind intensity increasing by 0.4, 0.3, and 0.3 ms-1 per decade in the NW, SW, and S regions, respectively. Analyses of the sudden shifts showed significant increases in northerly wind intensities after 2003, 2004, and 1998 in the NW, SW, and S coast, respectively. Exceptions were found for autumn (September for short-term data), when a decrease in northerly winds was observed in recent decades, regardless of the area, and for summer, when no changes in wind trends were recorded in the NW and SW. The long-term data also showed a major increase in northerly winds in winter (January and February), which is the recruitment season for many small and medium-sized pelagic fish. The increase in the intensity of the northerly winds over the past two decades and the past half-century occurred at a higher rate than was estimated by the IPCC for the next century.

1 Introduction At the global scale, the prediction of wind field variability at different time scales is important for several reasons. Wind speed has a significant impact on power production (Tobin et al. 2015), storm forecasting (Powell et al. 1991), and wave field evaluation (Caires and Sterl 2004). The characterization and understanding of wind variability have attracted a great deal of interest recently, particularly for the installation of offshore wind power farms in many parts of the world, but not in Portugal. Nonetheless, 16% of the energy produced in Portugal originates from renewable energy, and 20% originates from wind energy from a continental source (PORDATA). Electronic supplementary material The online version of this article (https://doi.org/10.1007/s00704-018-2466-9) contains supplementary material, which is available to authorized users. * Francisco Leitão [email protected] 1

Centro de Ciências do Mar, Universidade do Algarve, Campus de Gambelas, 8005-139 Faro, Portugal

In the Iberian Peninsula, most wind studies consider continental wind behavior by using limited time-series data (Lorente-Plazas et al. 2015a, b). In contrast, little attention has been paid to wind at sea (Carvalho et al. 2013; Sánchez et al. 2007). Oceanographic wind studies performed for the Iberian Peninsula on long-term wind using observational or reanalysis datasets and on wind variability are scarce. Coastal oceanography studies along the southern Iberian Peninsula have usually been based on anemometer wind measurements either from coastal meteorological stations, which were either land-based (Fiúza et al. 1982), or more recently, from ocean buoys (Sánchez et al. 2006). Although these series generally cover short-medium period of time and have great temporal resolution, the scarcity of land-based stations and ocean buoys precludes investigations of the spatial wind field (Carvalho et al. 2013). To address these concerns, satellite-driven products (e.g., Cross-Calibrated Multi-Platform Ocean Surface Wind Vectors (CCMP)) and conventional ship and buoy data (e.g., International Comprehensive Ocean-Atmosphere Data Set (ICOADs)) have been used extensively in oceanographic studies to provide additional environmental information for a given spatial area (Freeman et al. 2016; Reis et al. 2006). ICOADS is the most widely used freely available collection

Author's personal copy F. Leitão et al.

of surface marine observations and provides data for constructing gridded analyses of meteorological variables (Freeman et al. 2016). Moreover, both the satellite and ICOADS datasets evolved by releasing new products that allow the correction of observed data (e.g., satellite-driven products began to include Cross-Calibrated Multi-Platform Ocean Surface Wind Vectors, and ICOADS released new filtered/ corrected data). In fact, some of the few published studies to compare multi-source ocean surface wind data in the Iberian Peninsula were derived from several satellite products and the Cross-Calibrated Multi-Platform (CCMP) project with buoys, and they showed that the wind data trends are similar (Carvalho et al. 2013). Accordingly, most available scientific studies regarding oceanographic features (e.g., sea surface temperature, salinity, and upwelling), wind, and particular wind components (such as v-wind or meridional wind) have been one of the less studied marine coastal features. Such reasons are related with the spatial distribution of wind, that distributes at a large-scale, with the surface wind values at the mesoscale (regional scale) usually demonstrating high temporal oscillation from the intra-annual to multi-centennial time scale and anomalous events of high or low wind speeds (Caires and Sterl 2004). The surface wind field can loosely be considered a local response to the large-scale circulation. The large variability and vectorial nature of this variable not only introduce additional complexity to its diagnosis and prediction but also provide an important topic of study in related disciplines, such as fisheries oceanography (Cushing 1975; Houde 2008; Cury and Roy 1989). Several synthesis are consistent with the hypothesis that wind-driven (wind direction and intensity) turbulent mixing, pending sea bottom topography and coastline configuration, affects the variability in the survival of young fish larvae and fisheries (Peterman and Bradford 1987; O’Brien et al. 2012) with the magnitude of the wind impacts determined by the timing of interactions between specific stages of the marine life cycle and local environmental conditions (Leitão et al. 2016). The Portuguese coast is part of the northern branch of the eastern boundary upwelling system of the Canary Current, which is also occasionally called the Portugal Current at this latitude. In this area, most of the oceanographic characteristics, such as coastal upwelling events, filaments and eddies, and circulation patterns are closely related to wind variability (Fiúza 1983; García-Lafuente and Ruiz 2007). The southern Portuguese coastline is zonally oriented (West-East), and the west coast is meridionally oriented (North-South). Thus, the impact of wind on Portuguese coastal dynamics varies based on the orientation of the coast and/or the cross-shore component of the Ekman transport induced by alongshore windstress. In coastal areas, the alongshore wind components and Coriolis deflection generate a net transport of surface water perpendicular to the coastline (cross-shore Ekman transport),

carrying the upper layer either away from the coast (upwelling) or toward the coast (downwelling). Therefore, northerly wind is responsible for upwelling along the western coast of Portugal, while along the southern coast of Portugal, northerly wind induces drift currents (Vila-Concejo et al. 2003) that affect the vertical mixing of water. In fact, the marine coastal transition zone off the coast of Iberia is characterized by a marked seasonality that is related to the largescale wind climatology. Both satellite imagery and in situ observations reveal a poleward surface current as a persistent feature of the winter circulation, when winds relax or favor downwelling (Frouin et al. 1990; Haynes and Barton 1990). However, wind in the Iberian Peninsula blows toward the equator along the western meridional margins during a considerable portion of the year and drives a seaward flow in the upper layers, which can be explained by Ekman dynamics and the upwelling of cold subsurface waters along the coast. A well-defined upwelling season between March and September in the western Portuguese coast (Fiúza et al. 1982) results from strong northerly winds associated with the northward displacement of the Azores high-pressure cell and the weakening of the Icelandic low. Moreover, in the Iberian Peninsula, strong westerlies and storms occur in winters that have a negative North Atlantic Oscillation (NAO) index, whereas upwelling-favorable winds characterize winters that have a positive NAO index (Trigo et al. 2002). In addition to this known winter effect, recent research has shown that the NAO index influences the summer climate in this area (Folland et al. 2009). Moreover, the Iberian marine wind circulation is affected by coastal low-level jets that are a low-tropospheric wind feature. Such winds are the result of a pressure gradient produced by a sharp contrast between high temperatures over land and lower temperatures over the sea. Consequently, when this gradient is high, such as in summer when contrast between the cold ocean and the warm land increases, these winds impact the coastal parallel winds on the ocean, generating upwelling currents, sharpening the temperature gradient close to the coast and giving rise to strong baroclinic structures on the coast (Soares et al. 2014). Alves and Miranda (2013) showed that between 1989 and 2008, there were a decreasing number of coastal upwelling events in the northern part of the western Iberian coast and an increasing number of coastal upwelling events in the southern part of the western Iberian coast. These opposite trends in the mean number of coastal upwelling days are in good agreement with the trends in the meridional wind component off of the western Iberian coast, computed using ERA-40, ERAInterim, and satellite scatterometer data, which all indicated a weakening of the northerly wind intensity in the north and an intensification in the south. The wind speed changes estimated for the Portuguese marine Iberian coast (Intergovernmental Panel on Climate Change (IPCC)) are projected using the HadCM3 global circulation model (Miranda et al. 2006;

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950

Andrade et al. 2007). Despite the deficiencies found by Portuguese modelers in HadCM3 A2 wind simulations, which are shown to have a low degree of confidence, the total wind strength is projected to decline at a low rate of 0.01 m2 s-2 per year over the next hundred years (Reis et al. 2006). For Scenario HadCRM3 B2, no changes in wind strength are projected. However, time series studies showing the evolution of meridional wind intensity over previous periods (a halfcentury or decades) and future projections are missing. Such studies are important because they can help explain the observed variations in the behavior of the coastal system, including recruitment and fish landings. On the Portuguese coast, the meridional wind component (v-wind) is a major issue, as explained above, due to its direct and indirect effects on biotic communities. The assessment of climate variability at the regional scale is mainly based on the application of downscaling approaches that employ large-scale atmospheric circulation information to obtain estimations of variables at the regional/local scale by identifying the main statistical associations between the spatial scales (statistical downscaling) or using limited area models (dynamical downscaling). However, the transfer of information between spatial scales involves many sources of uncertainty that propagate from global to more regional scales in the downscaling process (García-Bustamante et al. 2012). Therefore, earth system models (such as those used by the IPCC for wind) operate at a large scale, and global models are uncertain at a regional scale. Moreover, global circular models (IPCC) do not differentiate between wind components, such as meridional wind that affects the coastal upwelling dynamics of the western Portuguese coast. The detection, estimation, and prediction of meridional wind trends and associated statistical and physical significance are important aspects of marine climate research. To date, time series analyses of meridional wind along the Portuguese Iberian coast are lacking. In this study, meridional wind trends were studied for the period between 1960 and 2010 to evaluate the inter-annual, seasonal, and monthly trends and identify sudden shifts in meridional wind regimes. Local environmental conditions are the result of the interaction between large-scale atmospheric and oceanographic conditions and the morphology at each geographical point. Therefore, the meridional wind trends were individually studied for three distinct climatic/oceanographic areas of the Portuguese coast. In the present research, we discuss our analysis of the wind time series in the context of oceanographic and fisheries research.

2 Methods 2.1 Area description The study of meridional northerly winds was evaluated individually for three distinct sub-areas of the Portuguese coast.

The division of the coast in sub-areas was performed according to the marine oceanographic condition of each area (Cunha 2001; Bettencourt et al. 2004), and additionally it also matches many fisheries resources division adapted from ICES (International Council for the Exploration of the Sea), to area IXa stocks. Therefore, the coast was divided into the Northwestern (NW), Center or Southwestern (SW), and South-Algarve (S) Atlantic coast of Portugal (Fig. 1).

2.2 Data acquisition Monthly meridional wind data, hereafter v-wind (that is a wind blows in the North-South direction; units: m/s), were obtained from two independent data archives: (i) International Comprehensive Ocean-Atmosphere Data Set, hereafter defined as ICOADS data (Link: http://rda.ucar.edu/ datasets/ds540.0/#!imma_subset.php) and (ii) from PO. DAAC hereafter defined as Satellite data (Atlas et al. 2011; L i n k : ht t p: / / p o d aa c . j pl . n as a . g o v / d a t a s e t / C C M P _ MEASURES_ATLAS_L4_OW_L3_5A_MONTHLY_ WIND_VECTORS_FLK?ids=&values=). The monthly ICOADS data set comprised a 50-year period (1960 to 2010) and was used to evaluate long-term changes in v-wind regimes. Satellite data comprised 22-year period (1988–2009) and was used for confronting ICOADS data: in what extend the two databases shown similar trends? Positive v-wind values means that wind blows from the south while negative v-wind values means that wind blows from the north. This means that a decrease (increase) in negative v-wind means an increase (decrease) of the northerly wind intensity, while a decrease (increase) in positive v-wind means a decrease (increase) in the southerly wind intensity. The monthly ICOADS data were downloaded in a tabular format and can be used for statistical time series analyses (http://rda.ucar.edu/datasets/ds540.0/#!imma_subset.php; Woodruff et al. 2011), allowing the collection of data by each observation. The period covered by ICOADS dataset comprised a more extended period (extends from 1960 to 2010), with good representation over the coastal areas (as depicted in Fig. 1), therefore allowing for comparison of recent v-wind records (satellite data) with historical ones (ICOADS data) over the past 50 years. Satellite monthly data are available from summary imagery data obtained at 4 × 4 km spatial resolution accessible on the NASA Ocean Color Giovanni website, (http://gdata1.sci.gsfc. nasa.gov) and includes monthly information between 1988 and 2009. Monthly satellite data were obtained for the Portuguese coast, using Marine Geospatial Ecology Tools (MGET) (Roberts et al. 2010). Satellite information comprised raster files and therefore raster coverages were required to conjugate into tabular data. Raster values contained within each polygon/region (see Fig. 1) for a given period were averaged by regions (NW, SW and S) using the Bisectpolyrst^

Author's personal copy F. Leitão et al.

Fig. 1 The coast divided into the Northwestern (NW), Center or Southwestern (SW), and South-Algarve (S) Atlantic coast of Portugal

tool of Geospatial Modelling Environment (http://www. spatialecology.com/gme/isectpolyrst.htm). All wind data, derived from both ICOADS and Satellite records, included observations until 200 m depth (200 m bathymetric), the Portuguese continental shelf break limit (Fig. 1).

2.3 Time series trends analyses If one is looking for common trends in multiple time series that were measured monthly, then the main part of the variation may be related to seasonal fluctuation. It may be an option to remove the seasonal pattern in each time series and focus on the remaining information instead (Zuur et al. 2007). Several techniques can be used to remove seasonality in monthly environmental data, as in the case of v-wind. A clear seasonal pattern in wind data is revealed by the autocorrelation function (Supplement 1, top figure). Herein, to derive general trends in monthly vwind values, time series were decomposed using LOESS smoother to eliminate the strong seasonal signals within these data sets (Cleveland 1979). LOESS (locally weighted smoothing) is a popular tool used in regression analysis that creates a smooth line through a time plot to help identify relationships between variables (time and v-wind) and foresee trends. LOESS is a polynomial model that combines the simplicity of linear least squares regression with

the flexibility of nonlinear regression (k-nearest-neighbor). LOESS includes fitting simple models to localized subsets of the data to build up a function that describes the deterministic part of the variation point by point in the data. Therefore, LOESS does not require the user to specify a global function to fit a model to the data but instead requires functions that are changed to fit each segment of the data. This smoother technique works in a similar manner to the running-line smoother except weighting factors for the regression are used within each segment of the data (Zuur et al. 2007). The output of this technique is a long-term trend (a long-term smoother trend is estimated), a seasonal component, and residual information for each time series (Zuur et al. 2007). The trend estimated by LOESS represents the original monthly series with seasonal variations removed and 12 months’ periodicity, thus allowing putative trends in the data to be captured. The trend estimated by LOESS was plotted and later analyzed using the dynamic factor analysis (DFA) time series technique. DFA is a multivariate smoothing technique that can be used for non-stationary time series analysis comparisons (Zuur et al. 2003a, 2007). DFA applies the same principle as factor analysis, in which the axes are restricted to be latent smoothing functions over time (Zuur et al. 2003a). DFA was used to compare yearly and monthly (after LOESS application) v-wind time series across

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950

study areas and between datasets. Therefore, DFA allows the N time series to be modeled as a linear combination of M common trends. These common trends represent the joint signal in a group or all the series. Only the period covered by the two different datasets (matching years: 1988 to 2009) was used in DFA. Six time series (N = 6) were initially introduced in the DFA model (ICOADS-NW, ICOADS-SW, ICOADS-S, Satellite-NW, Satellite-SW, and Satellite-S) to evaluate the number of common trends that are required to describe each individual time series. The DFA was made independently for yearly observational data and monthly detrended data (after LOESS application). The DFA trend for a structural time series model can be described as follows: Time series = M common trends + noise, where the fitted values and trends are the values obtained by the Kalman smoothing/filter algorithm (Zuur et al. 2003). Validation of the DFA was carried out considering (i) the observation of residual plots (a trend in the residuals means not satisfactory); (ii) DFA canonical values of the time series correlation with DFA common trend(s), or (iii) a DFA model showing overfitting signs (observed and fitted/estimated values matched). Then, individual DFAs were carried out to assess differences among vwind time series/datasets in each region. All DFA models started with setting the number of common trends. The DFAs started with the simplest model characterized by trend plus noise. Then, the number of trends can be increased in order to determine whether the M-time series are better explained by one or more trends. Different DFA models (single or M-common trends) were fitted by using a diagonal covariance matrix, and the Akaike’s information criterion (AIC) was used as a measure of goodness-of-fit and to compare models (Zuur et al. 2003a, b). The DFA model with the smallest AIC value was taken to be the optimal model (here, we only present the Bbest^ DFA models; see Supplement 2). Canonical correlation values of DFA were used to establish a relationship between individual time series and common DFA-estimated trend(s). Absolute values higher than 0.5 indicate a relationship between an individual time series and a DFA common trend. In summary, if the different time series are correlated with DFA-estimated common trends (canonical correlations values > 0.05), then the time series are similar. Before the DFAs, v-wind values were standardized (normalized to center all variables around zero) before running the models as advised by Zuur et al. (2003a, b). As revealed preliminarily by data exploitation procedures, all yearly and monthly (LOESS data) plots indicated that the data used in DFA are normally distributed. However, although normality of data is generally beneficial in DFA models, it is not a standard prerequisite (Zuur et al. 2003a). For this study, the

Brodgar software package (Highland Statistics Ltd, http://www.brodgar.com) was used to fit univariate time series DFA models. Linear regression model was also fitted to v-wind time series data to investigate and quantify (slope) the occurrence of inter-annual linear trends trend in time series. Therefore, the slope of the linear fitted model was used as a proxy of the trend tendency (upward or downward) and to quantify the rate of change in time series data. The statistical significance of the linear model was assessed via a student t test (P value < 0.05). The null hypothesis was formulated as no trend that describes an unchanging-wind. Following climatological studies in the extra tropics regions, inter-annual variability in seasonal v-wind trends yearly data was grouped as follows: Winter (December to February), Spring (March to May), Summer (June to August), and Autumn (September to November).

2.4 Detecting abrupt changes The yearly and monthly v-wind data were evaluated for regimes or sudden shifts. Discontinuities in the yearly time series were detected using the Regime Shift Analyses Index—RSI (Rodionov 2005a, b) using a National Oceanic and Atmospheric Administration software application (http://www.beringclimate.noaa.gov/ regimes/). Due to the high availability of data for evaluating and detecting sudden shifts in long-term data, ICOADS datasets were used. There are two parameters that control the magnitude and scale of the regime shifts to be detected: the significance level and the cut-off length. The significance level is the level at which the null hypothesis, in which the mean values of the two regimes are equal, is rejected by the two-tailed Student’s t test. The lower the significance level, the larger the magnitude of the shift detected. If a regime shift is identified, the difference between the mean values of the old and new regimes is statistically significant for at least one of the given levels. The cut-off is similar to the 100% cut-off point in filtering. All of the regimes longer than the cut-off length are detected. The shorter the cut-off length, the shorter the regimes that are selected, and vice-versa. For each time series, the regime shift index (RSI), the mean value of the regime with equal and unequal weights, regime length, and confidence levels for the shifts and the weights of the outliers were calculated. Different significance levels, cut-off lengths, and Huber parameters were set to better understand their mutual effects on regime detection. A high inter-annual variability was expected in non-stationary v-wind values, and therefore a P-significance of 10% (P = 0.1) was used.

Author's personal copy F. Leitão et al.

Moreover, after the initial data exploitation, models were fitted with a cut-off length of 10, and the Huber parameter (Huber 2005) that controls the weights assigned to the outliers was set to 3 (i.e., all v-wind values that were less than three standard deviations had equal weights). Therefore, this parameter affects the average value of the regimes (no significant outliers—P values < 0.05— were highlighted by the sudden shifts in the yearly analyses—Regime Shift Index—for any area, which evidences a good selection of the model parameters regarding Huber’s weight). The setting of the RSI parameters provided statistically significant models. The preliminary analysis of stationary wind data was based on time series plots as well as auto-correlation function analyses (ACF) and partial-ACF (PACF). ACF allowed the detection of patterns in the time series for which it was calculated and indicated the presence of a trend by studying the shape. The auto-correlation function is simply determined by using a Pearson correlation of a time series with itself after applying a shift (lag) of k-years (Zuur et al. 2007). This function estimates the length of the cycle rather than considering seasonality (monthly patterns; Supplement 1, upper figure). However, fewer points were used to calculate the length of the cycle for larger time lags k; therefore, it was better to limit the interpretation of the auto-correlation to only the first 40% of the time lags (see Zuur et al. 2007). Overall, the v-wind values were non-stationary, and correlogram analyses showed that the time series had few significant spikes at small lags and decreased sharply after early lags (Supplement 1, middle and lower figures). The Kolmogrov-Zurbenko Adaptative (KZA) filter method was used to detect discontinuities (break points) in the monthly v-wind time series (Zurbenko 1986). The KZA method is based on an iterated moving average series that adjusts the length of the window according to the rate of change of the variable. This filter method removes noise and seasonality from time series, enabling the separation of the low-frequency component from the original sign. This procedure included two steps: (i) estimation of the KZ filter based on an iterated moving average procedure for smooth data (Zurbenko et al. 1996), and (ii) calculation of a variant adaptive KZ, allowing the detection of abrupt changes in the times series (Yang and Zurbenko 2010). KZA filtering analysis was conducted using the BKZA^ library (version 3.0.0) in the R software (R Development Core Team 2016). Three iterations (K-component of KZA = 3) were used to estimate the KZA filter, which resembled a Gaussian-shaped filter. Periodicity was set to 12 months. ACF and PACF analyses were also used to derive the window size of the KZA filter (m-component of KZA, that is, setting the Bsmooth^ component in the number

of months to smooth) based on information on time series lags; if lags derived from ACF and PACF were detected before month 13 of the time series, then 1 year (12 months) represented a suitable starting point the KZA analyses. After preliminary data analysis, the final KZ and KZA models included a smooth window of 4, 4 and 6 years for the NW, SW, and S study areas, respectively.

3 Results The monthly average v-wind intensities were − 2.14, − 2.97, and − 1.45 m/s−1 for the long-term ICOADS data (between 1960 and 2009) and − 2.58, − 3.48, and − 2.21 m/s−1 for short-term satellite data (between 1988 and 2009) along the NW, SW, and S coasts, respectively. A high inter-annual variability was also observed in the v-wind values regardless of the geographical area (Figs. 2 and 3, Table 1). The annual LOESS trends (obtained from average monthly LOESS values) showed a decline in the v-wind intensity in latter decades, independent of the area. All of the LOESS time series for matched years were correlated with a single common DFA (Dynamic Factorial Analysis) trend, which could indicate a similarity between the v-wind time series between areas and databases (Supplement 2). The common DFA trend for the Iberian Peninsula revealed a decline in v-wind trends, that is, an increase in the northerly wind intensity. LOESS long-term annual trends (ICOADS dataset) showed that the v-wind values dropped below the mean after 1987 in the NW coast (Fig. 2). A sudden shift in v-wind yearly values in the NW coast was observed in 2003, showing a significant decline in the RSI (Regime Shift Index), from an average of − 1.99 (1960–2002) to − 3.11 m/s−1 (2003–2009) (Fig. 4, Table 2). The monthly v-wind long-term data analyses (KZA) in the NW coast showed discontinuities in two different periods: a gradual decline in 1987 and a sudden decline later in 2005 (Fig. 5). In the SW coast, the LOESS long-term v-wind annual values (ICOADS dataset) declined gradually until 1978, with wind values dropping below the mean after 1973 (Fig. 2). After 1978, the v-wind time series showed an upward trend until 1998 (the maximum value of the time series). The time series continuously declined, reaching values below the mean after 2003, until reaching the lowest v-wind values recorded from the time series (Figs. 2 and 3). Annual sudden shifts (Fig. 4, Table 2) were observed in the SW coast after 2003, with the RSI showing a significant decrease from an average of − 2.84 (1960–2003) to − 3.88 m/s−1 (2004–2010). The monthly data analyses (KZA) of the SW coast showed a smooth discontinuity in v-wind between 1972 and 1984. However, the most significant shift in the monthly SW coast

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950

Fig. 2 LOESS long-term annual trends (ICOADS dataset) showing the v-wind values dropped below the mean after 1987 in the NW coast

Author's personal copy F. Leitão et al.

Fig. 3 The time series showing continuous decline, reaching values below the mean after 2003, until reaching the lowest v-wind values recorded from the time series

time series (KZA) was observed after 2005, when the values of the v-wind suddenly decreased (Fig. 5). On the Southern coast, the v-wind annual time series longterm LOESS values (ICOADS dataset) smoothly declined until 1972 and then gradually increased until the v-wind values dropped below the mean in 1995 (Fig. 2). A sudden shift in yearly data (Fig. 4, Table 2) was observed on the Southern coast in 1998, showing a significant decrease from an average of − 1.31 (1960–1997) to − 1.88 m/s−1 (1998– 2010) in the RSI. The KZA analyses showed a discontinuity in the v-wind trend on the Southern coast after 1968, when the v-wind gradually declined. Subsequently, the v-wind values reached a steady state, although they began to decline more significantly in the early-mid-1990s. A sudden shift in

the v-wind was observed in the KZA analyses (ICOADS dataset) on the Southern coast for 2003 (Fig. 5). The inter-annual and half-decadal v-wind oscillation were characterized by (i) a non-stationarity in mean v-wind values and a slight linear downward trend for long-term v-wind time series (1960–2009), regardless of the area (Table 1), and (ii) a downward trend for short-term data regardless of the area (represented by a dotted line with linear fitting, as plotted in Fig. 3). The continuous and random increases and decreases in half-decadal anomalies illustrate a non-stationarity of the time series around the mean (Supplement 1). Seasonal annual graphs of v-wind time series (Fig. 6) were characterized by (i) a predominant northerly wind regime in the Northwestern, Southwestern, and South regions (average v-

− 2.08 (± 0.11) − 2.61 (± 0.50) − 2.37 (± 0.48) − 2.37 (± 0.52) − 3.16 (± 0.61)

− 1.40 (± 1.08) − 1.99 (± 0.35) − 1.87 (± 0.58) − 2.22 (± 0.82) − 1.52 (± 0.65) − 2.08 (± 0.69) − 2.54 (± 0.90) − 2.23 (± 0.36) − 2.08 (± 0.99) − 3.29 (± 1.11)

− 1.46 (± 0.60) − 1.34 (± 0.40) − 1.40 (± 0.45) − 1.51 (± 0.51) − 1.60 (± 0.37)

− 3.30 − 4.14 − 3.82 − 3.90

− 4.86

− 3.58 − 3.57 − 3.94 − 3.13 − 3.50

− 4.89

− 2.10 (± 0.53)

− 1.10 (± 0.34) − 1.17 (± 0.16) − 1.61 (± 0.39) − 1.18 (± 0.36)

ICOADS

ICOADS

− 3.17

Mean

Min

− 3.48 − 4.09 − 4.27

IXaS-Algarve

IXaCS

SATELITE

ICOADS

SATELITE

ICOADS 0.27 − 1.43 − 1.07 − 0.54 − 1.08 − 0.86 − 1.36 − 1.68 − 0.83 − 1.68

Max

Mean

IXaCN

− 2.63 (± 0.45)

−2.76 − 2.49 − 2.55 − 2.70 − 2.82 − 2.83 − 3.71 − 2.69 − 3.74 − 4.86

ICOADS

− 1.82 (± 0.44) − 2.19 (± 0.24) − 2.11 (± 0.37) − 2.06 (± 0.30)

SATELITE

− 1.98 − 1.91 − 1.54 − 1.71 − 2.00

SATELITE

Min

− 1.40

− 0.40 − 0.59 − 0.67 − 0.79 − 1.35

− 0.91 − 1.09 − 0.50

− 0.71

ICOADS

Max

− 2.19 − 3.34 − 2.92 − 3.19 − 3.66

SATELITE

− 1.85

− 1.38 − 1.96 − 1.52 − 1.79

SATELITE

− 2.73 (± 0.36) − 2.88 (± 0.50) − 2.64 (± 0.63) − 3.16 (± 0.74) − 3.43 (± 0.41) − 2.75 (± 0.58) − 2.94 (± 0.52) − 2.40 (± 0.59) − 2.63 (± 0.45) − 3.89 (± 0.71)

ICOADS

Mean

IXaCS

− 2.73

− 1.95 − 1.95 − 1.84 − 2.26 − 2.33

− 1.36 − 2.19 − 1.49

− 1.69

− 2.13 − 2.36 − 2.27 − 2.07 − 2.49 − 2.01 − 2.45 − 1.59 − 2.26 − 2.79

ICOADS

ICOADS

Min

− 3.10 (± 0.20) − 3.59 (± 0.46) − 3.27 (± 0.47) − 3.24 (± 0.34) − 3.98 (± 0.69)

SATELITE

Max

− 3.21

− 2.25 − 2.59 − 2.58 − 2.63

SATELITE

− 2.90 − 2.98 − 2.46 − 2.88 − 2.98

SATELITE

Half-decadal mean, maximum (Max), and minimum (Min) values of the v-wind (northerly wind) for each study area: Northwestern (NW), Southwestern (NW), and Southern (S-Algarve)

1960–1964 1965–1969 1970–1974 1975–1979 1980–1984 1985–1989 1990–1994 1995–1999 2000–2004 2005–2009

Years

Table 1

Author's personal copy

Northerly wind trends along the Portuguese marine coast since 1950

Author's personal copy F. Leitão et al. Table 2 Regime shift index (RSI) mean values and confidence levels (P value < 0.05 indicates a significant shift) of the regime analyses for each study area: Northwestern (NW), Southwestern (NW), and Southern (SAlgarve) Years

IXaCN RSI

1998 2003 2004

Fig. 4 Annual sudden shifts v-wind values

wind values were lower than zero regardless of the season and area); (ii) a non-stationarity and a slight linear downward trend in v-wind long-term values (ICOADS: 1960–2009) in all seasons regardless of the area; (iii) a downward linear trend for short-term v-wind values (Satellite: 1988–2009) in all seasons with the exception of summer in the NW; (iv) a decline in the vwind for short-term data in all areas, most pronounced in autumn and winter for both the NW and SW coasts (Fig. 6, Table 3); and (v) a greater slope value of the v-wind downward linear trend (v-wind intensity) in the short term (Satellite: 1988– 2009) than in the long term (ICOADS: 1960–2009) (Table 3). A decline was observed in the yearly inter-annual longterm values (ICOADS: 1960–2009) on the NW coast for most of the months with the exception of December (Fig. 7, Table 3). For long-term data, January, February, and September were the months with the highest inter-annual variability, declining on the NW coast at rates higher than 0.05 m/ s−1 per year. Along the NW coast, the short-term inter-annual values declined in January, May, November, and December (increased intensity of v-wind or northerly wind) but displayed an upward trend in March, June, and September; the absolute values of the upward or downward trend rates were greater than 0.05 m/s−1 per year for the latter months. On the SW coast, the inter-annual monthly values showed a decline for most of the months, regardless of the database

0 − 5.595 0

IXaCS

IXaS-Algarve

Conf

RSI

Conf



0



0.030 –

0 − 5.956

– 0.017

RSI

Conf

− 6.881

0.003

0 0

– –

(Fig. 8, Table 3). The inter-annual long-term monthly values also declined on the SW coast, except in April, August, September, and October. The inter-annual rate of increase was extremely low for the later months (less than 0.009 m/ s−1 per year). January and February showed the most pronounced decline in inter-annual long-term values on the SW coast; for the later months, the absolute value of the rate was above 0.03 m/s−1 per year. In contrast, March, April, June, and September showed an increase in yearly short-term values on the SW and NW coasts (higher than 0.047 m/s−1 per year; Fig. 8, Table 3). The inter-annual values showed a decline for long-term data on the southern coast, except in December, when the trend intensity was extremely low (rate of increase < 0.002 m/s−1 per year; Fig. 9, Table 3). Inter-annual increases in v-wind values along the southern coast were verified for March, April, June, and September (rate of increase > 0.026 m/s−1 per year), using short-term data and comparing it to ICOADS.

4 Discussion Based on the results from regional circulation models, changes in climate may differ by region, resulting in the heterogeneous response of environmental variables (Reis et al. 2006), as studied here for meridional winds along the Portuguese mainland coast. A better understanding of the variability of regional/local wind trends would greatly benefit applications that rely on the prediction of wind-related variables. Both long- (1960–2010) and short-term (1988–2009) records of v-wind data over past decades showed a negative trend in all of the areas of Portugal that were studied (NW, SW, and South-Algarve), which indicates an increase in the northerly wind intensity because a negative v-wind is indicative of northerly wind. Independent of the area studied, both the inter-annual and seasonal rates of decline were lower than those of specific months, where the v-wind intensity decreased more substantially. Individual monthly v-wind analyses are better evaluated and contextualized within the context of the field of fisheries oceanography. In a region of wind-

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950

Fig. 5 The NW coast showing discontinuities in two different periods

induced coastal upwelling such as the Portuguese Iberian coast, the existing relationship between the recruitment success of fish, namely, pelagic fish, and the intensity of upwelling is likely to be dome-shaped (Cury and Roy 1989). For instance, many small and medium pelagic fish are known to depend on the upwelling-rich waters that are associated with northerly winds in northwestern and southwestern Iberia (Santos et al. 2001, Borges et al. 2003; Leitão et al. 2014; Leitão 2015b). Recruitment can be affected by northerly winds during specific periods. Such could be the case in autumn (October, November, and December), winter (between January and March), and the early months of spring (March and April), when there is an increase in the magnitude of the wind from the north independent of the area in recent years. The later seasons/months correspond to the peak of recruitment for several species of fish, such as sardine, mackerel, and sea horse mackerel, when negative interactions with larvae phases have been found (Chicharo et al. 2003; Santos et al.

2007; Leitão et al. 2014; Leitão 2015b). These pelagic fish account for 63% of the Portuguese landings between 1950 and 2010 (Leitão 2015a). Consequently, the persistence of this trend or the absence of an optimal range of wind conditions could significantly affect coastal fisheries during such critical periods (Cury and Roy 1989) and should be accounted for in fisheries management (Baptista and Leitão 2014; Leitão et al. 2014). The effect of wind on fish recruitment and landings is indirect and occurs mainly (but not exclusively) through a wind-induced upwelling process. We are aware that the upwelling process that results in the seaward transport of cold upwelled water is also a function of oceanographic conditions in the water column, particularly stratification and pycnocline depth. As a result, the same upwelling-favorable wind may result in different upwelling patterns and intensities. Regardless, many widely used upwelling indices are based on the strength of upwelling-favorable winds (e.g., PFEL upwelling indices https://www.pfeg.noaa.gov/products/PFEL/ modeled/indices/upwelling/upwelling.html) and their application has produced valuable results (Santos et al. 2001). Thus, the assessment and analysis of regional wind trends have been revealed to be an essential component of basic research to support further studies, mainly related to fisheries oceanography. Additionally, the most cost-effective management options are those that operate at regional scales (Baptista and Leitão 2014). Therefore, it is necessary to consider the local background environmental influences, such as v-wind, and adapt fishing effort practices when identifying conservation measures to ensure the sustainable harvest of fish management units. There are numerous caveats that should be kept in mind when analyzing time series trends. Long- and short-term estimates based on observations are subject to differing sampling networks. The wind record from ICOADS that spans 1960 to the present clearly demonstrates a trend of long-term gradual decline. It is intuitive that a rate of decrease or increase in meridional wind depends on the date used for time series analyses. However, as shown using satellite data, there are short-term negative trends that cumulatively cover 22 years that are embedded within this series, where the recorded rate of decline is higher than that found in the long-term analyses. An estimation of the correlation between non-stationary variables can show false dependencies, which lead to inaccurate conclusions about the development of v-wind trends. In a stationary process, the statistical parameters do not change over time (Challis and Kitney 1991). Meridional wind lacked a large spatial inter-annual autocorrelation, indicating a nonstationary climate system regardless of the area. Therefore, if a given variable has a deterministic trend, a trend-adjusted series should be used, i.e., DFA. In fact, DFA revealed that the

Author's personal copy F. Leitão et al.

Fig. 6 Seasonal annual graphs of v-wind time series

overall v-wind trends were similar, which allowed comparisons between the short- and long-term data without a dependence on specific periods for analyses. Both data sources (including station, ship, and satellite observations) were subject to assorted errors that were random, systematic, or external, such as changing instruments, observation times, or observational environments. Despite these errors, trend analysis showed congruence between the satellite and ICOADS datasets (observed values). However, overall statistical models handle the basic dilemma of how to balance realism (the qualitative aspect), precision (the quantitative aspect), and generality (the aspect of the universality of applicability) differently, by using only a single approach (Levins 1966). Herein, the trend analysis was restricted, and considering the differences between satellite and ICOADS dataset observations (observed values) was beyond the scope of this study. The linear fitting followed the same trend in the NW and SW coastal regions for most months; in March, April, June,

and September, the yearly short-term northerly wind values increased meaningfully (decline in negative v-wind values). However, some of these trends were not significant, although the trend slope information was used for result/pattern analyses (qualitative analyses were suitable for the scope of the work). The non-stationary nature of the v-wind time series analyses showed that wind demonstrated long-term trends between the mid-1950s and 2010s. However, anomalous positive/negative episodes of v-wind intensity are identified inter-annually, over half of a decade or over decades. Analyses of yearly sudden shifts reveal an abrupt decline in the v-wind intensity in recent decades regardless of the area. In fact, in South-Algarve, a clear shift in vwind has occurred since the late 1990s (1998), while in both the NW and SW, these sudden drops in v-wind occurred only a half-decade later, in 2003 and 2004, respectively. The analyses of annual sudden shifts also reveal periods in which the rate of decline of the v-wind is

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950 Table 3 Slope values for seasonal and monthly linear regression analyses of northerly wind (v-wind) for both long-term (ICOADS) and short-term (Satellite) data for each study area: Northwestern (NW), Southwestern (NW), and Southern (S-Algarve) Satellite: 1988–2019

ICOADS: 1960–2019

IXaCN

IXaCS

IXaS

IXaCN

January

− 0.126

− 0.088

− 0.086

− 0.080 − 0.033 − 0.035

February March

− 0.071 0.079

0.000 0.048

0.068 − 0.051 − 0.048 − 0.043 0.032 − 0.005 − 0.017 − 0.012

April

0.030

0.051

0.040 − 0.020

May June

− 0.142 0.074

− 0.142 0.051

July August

0.016 − 0.007 − 0.083 − 0.019 − 0.019 − 0.024 − 0.049 − 0.047 − 0.025 − 0.026 0.007 − 0.007

IXaCS

IXaS

0.009 − 0.025

− 0.110 − 0.011 − 0.005 − 0.006 0.026 − 0.010 − 0.006 − 0.007

0.051 0.048 0.030 − 0.059 0.005 − 0.007 0.009 − 0.006 October − 0.001 − 0.011 − 0.015 − 0.026 November − 0.218 − 0.193 − 0.175 − 0.015 − 0.012 − 0.009 September

December Winter

− 0.132 −0.079 −0.083 0.034 0.021 0.002 − 0.110 − 0.056 − 0.003 − 0.032 − 0.020 − 0.025

Spring Summer Autumn

− 0.011 − 0.015 − 0.012 − 0.012 − 0.005 − 0.014 0.014 − 0.001 − 0.027 − 0.019 − 0.006 − 0.013 − 0.056 − 0.052 − 0.054 − 0.033 − 0.006 − 0.007

Italics values represent significant regression coefficients (P-value < 0.05)

several times greater than in the inter-annual, seasonal, or monthly time frames. Nevertheless, analyses (KZ and KZA filters) of monthly sudden shifts appeared to be quite accurate, detecting periods where shifts in v-wind occurred. For instance, KZ trends showed that in the three areas, v-wind had an oscillatory decreasing trend (KZ filter). However, KZA revealed different v-wind trend variability between regions. The first sudden drops in v-wind, according to KZA analyses, were in 1987, 1972, and 1968 in the NW, SW, and S-Algarve, respectively. This reveals a latitudinal trend in the v-wind decline that started earlier in the south than in the north of Portugal. The KZA data also showed that after the v-wind declined in specific periods, it returned to values similar to those observed before the decline (e.g., the SW after 1972). The annual (RSI) and monthly (KZA) analyses do not correspond in terms of the sudden shifts regardless of the region. As expected, there is a decalage between the KZ and KZA due to the smooth terms and window size. The KZ analyses allowed a clear observation of the trend (decline in vwind) that preceded the sudden shift in the RSI. In fact, the smooth KZ time series line showed good agreement with the RSI shifts. Different analyses provide different results that are open to discussion. Therefore, we should favor the general interpretation of our results, including observation values. In fact, in all of the DFA, KZ, KZA,

and annual shift analyses, there was good agreement between the observed and fitted data. Overall, short-term inter-annual, seasonal, and monthly linear data analyses and sudden shift analyses showed that the v-wind intensity decreased at a higher rate in recent periods than in past periods. Environmental variability trends, if persisting over decades, can lead to climatic variability, and therefore, environmental and climatic variability are closely interdependent. In this present work, climatic variability follows the definition from the Inter-Governmental Panel on Climate Change (IPCC). In this study, climate variability is defined as the Baverage weather,^ i.e., the statistical description in terms of the mean and variability of relevant wind over a period that is classically assumed to be 30 years. In a wider sense, climate is the state of a climatic system (as defined by the World Meteorological Organization). The IPCC predicts an increase in the mean wind strength of 0.01 m s-1 over the next century. However, the rate of decline of vwind (or increased intensification of northerly winds) occurred at a higher rate over the past half-decade than the predicted intensification of total wind strength over the coming century (2007 onwards). Therefore, climatic changes in v-wind are found in this study, particularly in specific months or seasons in which the rate of decline is higher than the inter-annual rate of decline. It is important to understand past v-wind changes at the regional scale to predict future climate scenarios and their effects. Predicting changes in coastal and shelf marine systems, such as those in the Iberian Peninsula, may become crucial since studies indicate that climate affects fisheries (Teixeira et al. 2016). The populations of many species of commercial fish have decreased along the Portuguese coast (Leitão 2015a). This decline in population has raised concerns among scientists, managers, and citizens regarding many endangered fish species, with consequences at the population and ecosystem levels, and the reduction in the overall value of this resource for human consumption (Leitão 2015a). Our results support the hypothesis that in the eastern boundary upwelling systems, global warming results in the strengthening of the northerly coastal v-wind (a vwind decline), which is induced by an increased landsea temperature contrast and consequent atmospheric geostrophic adjustment (Bakun 1990). This northerly wind is upwelling-favorable off of the western coast of Portugal and appears to have caused an intensification of upwelling during previous decades (Relvas et al. 2009a, b). However, the statistical significance of the changes in the meridional upwelling-favorable wind off the coast of Iberia has been questioned (Barton et al. 2013), mainly in fisheries science, where studies on

Author's personal copy F. Leitão et al.

Fig. 7 Decline observed in the yearly inter-annual long-term values on the NW coast for most of the months with the exception of December

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950

Fig. 8 The inter-annual monthly values showing a decline for most of the months, regardless of the database

Author's personal copy F. Leitão et al.

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950 Fig. 9 The inter-annual values showing a decline for long-term data on the southern coast, except in December, when the trend intensity was extremely low

upwelling-rich waters induced by wind events reveal that upwelling or wind may enhance species recruitment (species-specific) depending on the season (Leitão et al 2014; Leitão 2015a; Leitão et al. 2016). Northerly winds drive a seasonal coastal upwelling regime off Western Iberia, intensified during summer months (typically from April until September) and episodically along the entire southern coast of Iberia and almost absent during winter (Relvas et al. 2007). However, the subsurface structure of the coastal ocean (thermohaline distribution along the water column) plays a major role in the upwelling intensification. The same wind at the same latitude blowing over a poorly stratified ocean or with a deep pycnocline would bring to the surface rich and colder waters from deeper levels than if it blows over a highly stratified ocean or with a shallow pycnocline. In fact, the coastal upwelling process is highly sensitive to the wind pattern, the thermohaline structure of the water column, and the coastline and seabed configurations. Due to these regional factors, the adjustment of the upwelling pattern is highly affected by variability in v-wind. Changes in these regional factors will result in the adjustment of the upwelling pattern; moreover, indirect wind also affects sea surface temperature (SST) as there is evidence that different mesoscale upwelling structures in the region experience different SST trends (Relvas et al. 2009a, b). Intensification of v-wind in this study was particularly relevant in autumn. Therefore, v-wind changes associated with variability in upwelling and SST would also be expected to affect biotic communities’ responses to oceanographic drivers in particular seasons. Additionally, the generalized intensification of upwelling remains controversial. There is some evidence that the response of the coastal ocean is not spatially uniform (Relvas et al. 2009a, b). The upwelling pattern off of the southern part of the western Iberian Peninsula, which is associated with northerly wind regimes, underwent a substantial intensification since 1985 in contrast to more regular behavior further north. The intensity of the v-wind is directly related to synoptic patterns, and it must to be related to the position of the high pressure over the Atlantic and the North Atlantic oscillation (NAO). The NAO index operates indirectly and at a broader spatial scale than other environmental variables. The yearly N A O e n t er e d a p o s i t i v e ph a s e i n 2 0 0 6 (h t t p s : / / climatedataguide.ucar.edu/climate-data/hurrell-north-atlanticoscillation-nao-index-station-based). A decline in v-wind has been observed since 2003 in both NW and SW and in 1998 in

South-Algarve. However, on the western Iberian Coast, the NAO is mainly a driver of westerly winds rather than v-winds. Positive values of the NAO index are typically associated with stronger-than-average westerlies over the middle latitudes, more intense weather systems over the North Atlantic, and wetter/milder weather over Western Europe. Based on spatial-temporal analyses of continental wind power resources in the Iberian Peninsula (Chidean et al. 2018) and Southwestern Europe (Jerez et al. 2013), it was found that wind speed patterns are forcefully related to differences between atmospheric pressures in Lisbon and Reykjavik (NAO index). Therefore, in future works, the relationship between the v-wind in the Iberian Peninsula and the NAO could be explored for the sake of clarifying the relationship between large-scale effects (e.g., NAO) and the v-wind field. In a study characterizing the surface continental wind speeds over the Iberian Peninsula, the results indicate that the seasonal variability of the synoptic wind scale is related to inter-annual variability and modulated by local features (Lorente-Plazas et al. 2015b). Additionally, several works on continental wind and the NAO effect noted that a more significant variation in mean wind direction may be the seasonal variation (Alan and McGregor, 2008; Lorente-Plazas et al. 2015b), which seems to agree with the results concerning seasonal v-wind trends found in this study for the Portuguese coast.

5 Conclusions Overall, the northerly wind intensity increased since the 1960s (a decline in v-wind indicates an increase in northerly winds since the northerly wind is negative), regardless of the area and dataset in the Portuguese coast. Over the past half century (ICOADS: 1960–2010), the negative northerly wind values increased at a yearly linear rate of 0.24, 0.09, and 0.15 m s-1 per decade in the NW, SW, and South-Algarve coastal regions, respectively. However, the v-wind intensity rate was higher in recent decades (Satellite: 1989–2009), with the negative northerly wind intensity increasing by 0.4, 0.3, and 0.3 ms-1 per decade in the NW, SW, and S regions, respectively. The decline in the v-wind trend indicates an increase in the intensity of the northerly wind and a decrease in the intensity of the southerly wind. Therefore, our results suggest a tendency for the wind to decrease its southerly component and/or gain in its northerly component. Analyses from different data series (annual, seasonal and monthly) provide different information. However, overall, the data showed that the v-wind decline (i) is most pronounced in recent decades, particularly in autumn, regardless of the area; (ii) is less evident in summer in the NW and SW in recent decades; and (iii) does not occur in September according to monthly data, when there was an increase in v-wind. The long-term data also showed (i) an overall decline in the

Author's personal copy F. Leitão et al.

seasonal v-wind values in the NW and South (Algarve) in winter and (ii) the most pronounced decline in the intensity of the v-wind in February (upwelling season for small pelagic species). Attention should be paid to sudden shift analyses that reveal recent periods when the v-wind changes varied at a rate several times higher than in monthly, inter-annual, or seasonal analyses. However, these 10- to 15-year periods are characterized by a return to average values similar to those recorded before the sudden shift, except in the most recent decade (2000–2010). While in the general trends in v-wind are similar in the different areas, the sudden shift patterns reveal that the period of the shifts varied between areas, exception in the last decade and a half (1995–2009), when a v-wind decline was generally observed, independent of the area. Acknowledgements Francisco Leitão (SFRH/BPD/108949/2015) and Vânia Baptista (SFRH/BD/104209/2014) hold scholarships from FCT – Foundation for Science and Technology. This research was partially supported by the European Regional Development Fund (ERDF) through the COMPETE – Operational Competitiveness Programme and national funds through FCT – Foundation for Science and Technology, under the project UID/Multi/04326/2013. This research was supported by CLIMFISH project - A framework for assess vulnerability of coastal fisheries to climate change in Portuguese coast founded by Portugal 2020, n2/SAICT/2017 - SAICT (Projetos de IC&DT).

References Alan L, McGregor J (2008) Seasonal variation of the prevailing wind direction in Britain. Weather 63:365–368 Alves JMR, Miranda PMA (2013) Variability of Iberian upwelling implied by ERA-40 and ERA-interim reanalyses. Tellus 65:19245 Andrade C, Pires HO, Taborda R, Freitas MC (2007) Projecting future changes in wave climate and coastal response in Portugal by the end of the 21st century. J Coast Res 50:253–257 Atlas R, Hoffman RN, Ardizzone J, Leidner SM, Jusem JC, Smith KD, Gombos D (2011) A cross-calibrated, multiplatform ocean surface wind velocity product for meteorological and oceanographic applications. Bull Am Met Soc 92:157–174 Bakun A (1990) Global climate change and intensification of coastal ocean upwelling. Science 247(4939):198–201. https://doi.org/10. 1126/science.247.4939.198 Baptista V, Leitão F (2014) Commercial catch rates of the clam Spisula solida reflect local environmental coastal conditions. J Mar Syst 130:79–89 Barton ED, Field DB, Roy C (2013) Canary current upwelling: more or less? Prog Oceanogr 116:167–178. https://doi.org/10.1016/j. pocean.2013.07.007 Bettencourt A, Bricker SB, Ferreira JG, Franco A, Marques JC, Melo JJ, Nobre A, Ramos L, Reis CS, Salas F, Silva MC, Simas T, Wolff W (2004) Typology and reference conditions for Portuguese Transitional and Coastal Waters Development of guidelines for the application of the European Union Water Framework Directive. Instituto da Agua (INAG) – Institute of Marine Science (IMAR), Lisbon Borges MF, Santos AMP, Crato N, Mendes H, Mota B (2003) Sardine regime shifts off Portugal: a time series analysis of catches and wind conditions. Sci Mar 67(1):235–244

Caires S, Sterl A (2004) 100-year return value estimates for ocean wind speed and significant wave height from the ERA-40 data. J Clim 18: 1032–1048 Carvalho D, Rocha A, Gómez-Gesteira M, Alvarez I, Santos SC (2013) Comparison between CCMP, QuikSCAT and buoy winds along the Iberian Peninsula coast. Remote Sens Environ 137:173–183 Challis RE, Kitney RI (1991) Biomedical signal processing (in four parts). Part 1 time domain methods. Med Biol Eng Comput 29(1): 1–17 Chícharo MA, Esteves E, Santos AMP, Santos A, Peliz Á, Ré P (2003) Are sardine larvae caught during a winter upwelling event off northern Portugal starving? An approach using RNA/DNA ratios. Mar Ecol Prog Ser 257:303–309 Chidean ML, Caamaño AJ, Ramiro-Bargueño J, Casanova-Mateo C, Salcedo-Sanz S (2018) Spatio-temporal analysis of wind resource in the Iberian Peninsula with data-coupled clustering. Renew Sust Energ Rev 81:2684–2694 Cleveland WS (1979) Robust locally weighted regression and smoothing scatterplots. J Am Stat Assoc 74:829–836 Cunha ME (2001) Physical control of biological processes in a coastal upwelling system: comparison of the effects of coastal topography, river run-off and physical oceanography in the northern and southern parts of western Portuguese coastal waters. University of Lisbon, Lisbon, Portugal, PhD dissertation Cury P, Roy C (1989) Optimal environmental window and pelagic fish recruitment success in upwelling areas. Can J Fish Aquat Sci 46: 670–680 Cushing DH (1975) Marine ecology and fisheries. Cambridge University Press, Cambridge Fiúza AFG (1983) Upwelling patterns off Portugal. In: Suess E, Thiede J (eds) Coastal upwelling: its sediment record, part A. plenum, New York, pp 85–98 Fiúza AFG, Macedo ME, Guerreiro MR (1982) Climatological space and time variation of the Portuguese coastal upwelling. Oceanol Acta 5: 31–40 Folland CK, Knight J, Linderholm HW, Fereday D, Ineson S, Hurrell JW (2009) The summer North Atlantic oscillation: past, present, and future. J Clim 22:1082−1103 Freeman E, Woodruff SD, Worley SJ, Lubker SJ, Kent EC, Angel WE, Berry DI, Brohan P, Eastman R, Gates L, Gloeden W, Ji Z, Lawrimore J, Rayner NA, Rosenhagen G, Smith SR (2016) ICOADS release 3.0: a major update to the historical marine climate record. Int J Climatol 37:2211–2232. https://doi.org/10.1002/joc. 4775 Frouin R, Fiúza AFG, Ambar I, Boyd TJ (1990) Observations of a poleward surface current off the coasts of Portugal and Spain during winter. J Geophys Res 95:679–691 García-Bustamante E, González-Rouco JF, Navarro J, Xoplaki E, Jiménez PA, Montávez JP (2012) North Atlantic atmospheric circulation and surface wind in the northeast of the Iberian Peninsula: uncertainty and long term downscaled variability. Clim Dyn 38: 141–160 García-Lafuente J, Ruiz J (2007) The Gulf of Cádiz pelagic ecosystem: a review. Prog Oceanogr 74:228–251 Haynes R, Barton ED (1990) A poleward flow along the Atlantic coast of the Iberian Peninsula. J Geophys Res 95:11425–11442 Houde E (2008) Emerging from Hjort’s shadow. J Northwest Atlantic Fish Sci 41:53–70 Huber PJ (2005) Robust estimation of a location parameter. Annals Math Stat 35:73–101 Jerez S, Trigo RM, Vicente-Serrano SM, Pozo-Vázquez D, LorentePlazas R, Lorenzo-Lacruz J, Santos-Alamillos F, Montávez JP (2013) The impact of the North Atlantic Oscillation on renewable energy resources in southwestern Europe. J Appl Meteor Climatol 52:2204–2225

Author's personal copy Northerly wind trends along the Portuguese marine coast since 1950 Leitão F (2015a) Landing profiles of Portuguese fisheries: assessing the state of stocks. Fish Manag Ecol 22(2):152–163 Leitão F (2015b) Time series analyses reveal environmental and fisheries controls on Atlantic horse mackerel (Trachurus trachurus) catch rates. Cont Shelf Res 111(B):342–352 Leitão F, Alms V, Erzini K (2014) A multi-model approach to evaluate the role of environmental variability and fishing pressure in sardine fisheries. J Mar Syst 139:128–138 Leitão F, Baptista V, Teodósio MA, Hughes SJ, Vieira V, Chícharo L (2016) The role of environmental and fisheries multi-controls in white seabream (Diplodus sargus) artisanal fisheries in Portuguese coast. Reg Environ Chang 16:163–176 Levins R (1966) The strategy of model building in population biology. Am Sci 54(4):421–431 Lorente-Plazas R, Montávez JP, Jerez S, Gómez-Navarro JJ, JiménezGuerrero P, Jiménez PA (2015a) A 49 year hindcast of surface winds over the Iberian Peninsula. Int J Climatol 35:3007–3023 Lorente-Plazas R, Montávez JP, Jiménez PA, Jerez S, Gómez-Navarro JJ, García-Valero JA, Jiménez-Guerrero P (2015b) Characterization of surface winds over the Iberian Peninsula. Int J Climatol 35:1007– 1026 Miranda PM.A, Valente MA, Tomé AR, Trigo R, Coelho MFES., Aguiar A, and Eduardo BA 2006. O clima de Portugal nos séculos XX e XXI. In Alterações Climáticas em Portugal: Cenários, Impactos e Medidas de Adaptação, pp. 45-113. Ed. by F. D. Santos, and P. Miranda. Projecto SIAM II, Publicações Gravida, Lisboa, Portugal, 454 pp. O’Brien TD, Li WKW, Morán XAG. (Eds). (2012) ICES phytoplankton and microbial plankton status report 2009/2010. ICES Cooperative Research Report No. 313. 196 pp Peterman RM, Bradford MJ (1987) Wind speed and mortality rate of a marine fish, the northern anchovy (Engraulis mordax). Science 235: 354–356 Powell M, Dodge P, Black M (1991) The landfall of hurricane Hugo in the Carolinas-surface wind distribution. Sea Forecast 6(3):379–399 Reis C, Lemos RT, Alagador D (2006) Pescas. In: Santos FD, Miranda P (eds) Alterações Climáticas em Portugal: Cenários, Impactos e Medidas de Adaptação., Projecto SIAM II. Publicações Gravida, Lisboa, pp 345–384 Relvas P, Barton ED, Dubert J, Oliveira PB, Peliz AJ, da Silva JC, Santos AMP (2007) Physical oceanography of the western Iberia ecosystem: latest views and challenges. Prog Oceanogr 74:149–173. https://doi.org/10.1016/j.pocean.2007.04.021 Relvas P, Luís J, Santos AMP (2009a) Importance of the mesoscale in the decadal changes observed in the northern canary upwelling system. Geophys Res Lett 36(22):L22601. https://doi.org/10.1029/ 2009GL040504 Relvas P, Luis J, Santos AMP (2009b) The importance of the mesoscale in the decadal changes observed in the Northern Canary upwelling system. Geophys Res Lett 36:L22601 Roberts JJ, Best BD, Dunn DC, Treml EA, Halpin PN (2010) Marine geospatial ecology tools: an integrated framework for ecological geoprocessing with ArcGIS, Python, R, MATLAB, and C++. Environ Model Softw 25(10):1197–1207 Rodionov SN (2005a) A brief overview of the regime shift detection methods In: Large-scale disturbances (regime shifts) and recovery in aquatic ecosystems: challenges for management toward sustainability, V. Velikova and N. Chipev (Eds.), UNESCO-ROSTE/BAS Workshop on Regime Shifts, 14–16 June 2005, Varna, Bulgaria, 17–24

Rodionov SN (2005b) Detecting regime shifts in the mean and variance: methods and specific examples. In: Velikova V, Chipev N (Eds.) Large-scale disturbances (regime shifts) and recovery in aquatic ecosystems: challenges for management toward sustainability, UNESCO-ROSTE/BAS Workshop on Regime Shifts, 14-16 June 2005, Varna, Bulgaria, 68–72 Sánchez R, Mason E, Relvas P, da Silva AJ, Peliz AJ (2006) On the inshore circulation in the northern Gulf of Cádiz, southern Portuguese shelf. Deep-Sea Res II 53:1198–1218 Sánchez RF, Relvas P, Pires HO (2007) Comparisons of ocean scatterometer and anemometer winds off the southwestern Iberian Peninsula. Cont Shelf Res 27:155–175 Santos AMP, Borger MF, Groom S (2001) Sardine and horse mackerel recruitment and upwelling off Portugal. ICES J Mar Sci 58:589–596 Santos AMP, Chícharo MA, Santos A, Moita T, Oliveira PB, Peliz Á, Ré P (2007) Physical–biological interactions in the life history of small pelagic fish in the Western Iberia Upwelling Ecosystem. Prog Oceanogr 74(2–3):192–209 Soares PMM, Cardoso RM, Semedo A, Álvaro S, Chinita MJ, Raza R. (2014) Climatology of the Iberia coastal low-level wind jet: weather research forecasting model high-resolution results. Tellus Series ADynamic Meterology and Oceonagraphy. Vol. 66, nr. 22377 ISSN 0280–6495 Teixeira CM, Gamito R, Leitão F, Murta A, Cabra HN, Erzini K, Costa MJ (2016) Environmental influence on commercial fishery landings of small pelagic fish in Portugal. Reg Environ Chang 16:709–716 Tobin I, Vautard R, Balog I, Bréon F-M, Jerez S, Ruti PM, Thais F, Vrac M, Yiou P (2015) Assessing climate change impacts on European wind energy from ENSEMBLES high-resolution climate projections. Clim Chang 128:99–112 Trigo RM, Osborn TJ, Corte-Real J (2002) The North Atlantic Oscillation influence on Europe: climate impacts and associated physical mechanisms. Clim Res 20:9–17 Vila-Concejo A, Ferreiro Ò, Matias A, Dias JMA (2003) The first two years of an inlet: sedimentary dynamics. Cont Shelf Res 23:1425–1445 Woodruff SD., Worley S.J., Lubker S.J. Ji Z., Freeman J.E., Berry D.I., Brohan P., Kent E.C., Reynolds R.W., Smith S.R., Wilkinson C. (2011) ICOADS release 2.5: extensions and enhancements to the surface marine meteorological archive. Int J Climatol, 31(7), 951–967 Yang W, Zurbenko I (2010) Kolmogorov-Zurbenko filters. Wiley Interdiscip Rev Comput Stat 2(1):340–351 Zurbenko IG (1986) The spectral Analysis of Time Series. Elsevier North-Holland, Inc, New York Zurbenko I, Porter PS, Rao ST, Ku JY, Gui R, Eskridge RE (1996) Detecting discontinuities in time series of upper-air data: development and demonstration of an adaptive filter technique. J Clim 9(12):3548–3560 Zuur AF, Frywe RJ, Jolliffe IT, Dekker R, Beukema JJ (2003a) Estimating common trends in multivariate time series using dynamic factor analysis. Environmetrics 15:665–668 Zuur AF, Tuck ID, Bailey N (2003b) Dynamic factor analysis to estimate common trends in fisheries time series. Can J Fish Aquat Sci 60: 542–552 Zuur AF, Ieno EN, Smith GM (2007) Analysing ecological data. Springer, New York