Bioclimate map of Sardinia (Italy)

Abstract

Bioclimatology deals with the interrelation between climate and living organisms, in particular, plants and plant communities, considering the main climate variables that are relevant for species distribution. In this context spatial interpolation of monthly temperature and precipitation data using 203 rain gauges and 68 temperature gauges for Sardinia (Italy) was undertaken. As interpolation technique, we used regression kriging which combines multiple linear regression (MLR) with ordinary kriging of the residuals. MLR procedures include as independent variables: altitude, latitude, longitude, coast distance and a topographic factor of relative elevation. Elevation data were obtained from digital elevation model at 40 m resolution. Following the approach of the Worldwide Bioclimatic Classification System, a bioclimatic diagnosis of the entire territory was derived using map algebra calculations of the bioclimatic indices proposed by Rivas-Martínez et al. [(2011). Worldwide Bioclimatic classification system. Global Geobotany, 1, 1–638]. Two macrobioclimates (Mediterranean pluviseasonal oceanic and Temperate oceanic), one macrobioclimatic variant (Submediterranean), and four classes of continentality (from weak semihyperoceanic to weak semicontinental), eight thermotypic horizons (from lower thermomediterranean to upper supratemperate) and seven ombrotypic horizons (from lower dry to lower hyperhumid) were identified, resulting in a combination of 43 isobioclimates. The resulting map represents a useful environmental stratum, for regional planning, ecological modeling and biodiversity conservation.

Keywords:

• climate classification
• ecoregion
• Mediterranean region
• phytoclimate
• spatial interpolation

1. Introduction

Many disciplines use spatial information about climate variables for their studies, from hydrology to ecological modeling. In ecological studies, climate is considered the main factor that influences biological systems, affecting both the spatial distribution of living organisms and ecosystem functions and processes, for example, biomass production or wildfire regime (e.g. Bajocco, De Angelis, Rosati, & Ricotta, 2009; Bajocco, Rosati, & Ricotta, 2010). As a consequence, in hierarchical land classification and mapping (Klijn & Udo de Haes, 1994), bioclimate classification is considered a mandatory environmental layer that has to be taken into account in an analysis (e.g. Bailey, 2004; Blasi et al., 2000; Smiraglia et al., 2013). Hence, accurate bioclimatic maps are important tools for land management and biodiversity conservation (Blasi, Carranza, Frondoni, & Rosati, 2008; Metzger et al., 2013); for instance, if we are not able to detect the relationship between climate and biota, we are not well prepared to understand landscape change. Bioclimatology deals with the interrelation between climate and the distribution of living organisms, in particular, plants and plant communities (phytoclimatology), considering the main variables that are relevant for their distribution (Gavilán, 2005; Rivas-Martinez & Loidi., 1999). Such information is essential to understand vegetation distribution and structure (e.g. Box, 1996; Ríos-Cornejo, del Río, & Penas, 2012), to perform habitat modeling (Guisan & Zimmerman, 2000) and to analyze vegetation dynamics, subsequently allowing the reconstruction of vegetation series and potential natural vegetation (Blasi, Filibeck, Frondoni, Rosati, & Smiraglia, 2004; Blasi et al., 2012; Farris, Filibeck, Marignani, & Rosati, 2010). Furthermore, bioclimatic models, focusing on climatic thresholds of species distributions, can predict the responses of living organisms to climate change (Huntley, Berry, Cramer, & McDonald, 1995; Pearson & Dawson, 2003; Walther, Berger, & Sykes, 2005) because integrated and biologically relevant combinations are better suited to exploring potential ecosystem response to changing climates than raw climate data (Torregrosa, Taylor, Flint & Flint, 2013).

Many bioclimatic units, based on different classification methods and numerical criteria, have been defined by different authors, ranging from local to global application. Most were developed up to middle of the twentieth century, before large datasets and powerful computing resources became widely available (e.g. Gaussen, 1954; Holdridge, 1967; Köppen, 1936; de Martonne, 1937; Thornthwaite, 1948). More recently, biome or ecological classifications based on bioclimatic features have received greater attention (e.g. Metzger et al., 2013; Peel, Finlayson & McMahon, 2007; Prentice et al., 1992), although they generally provide only a coarse spatial resolution, distinguishing ∼10–30 classes globally. In addition, many recently published bioclimatic maps were produced at small scale or were the result of local analysis, lacking reference to a worldwide classification system (e.g. Pellicone, Caloiero, Coletta, & Veltri, 2014; Vondráková, Vávra, & Voženílek, 2013).

Developed by Rivas-Martinez and co-authors, the Worldwide Bioclimatic Classification Systems (WBCSs) can be applied at different scales and has recently been updated (Rivas-Martínez, Rivas-Sáenz, & Penas, 2011). Formulated in the context of vegetation science, this classification was widely adopted in geobotany, landscape ecology (e.g. Fascetti, Pirone, & Rosati, 2013; Gavilán, Fernández-González, & Blasi, 1998; Rosati et al., 2010) and climate change studies (Torregrosa et al., 2013). Unfortunately, several applications of this classification are based on empiric expert based mapping procedures, failing with respect to reproducibility and updatability (e.g. Cano, Cano-Ortiz, Del Río Gonzalez, Alatorre Cobos, & Veloz, 2012; Mesquita & Sousa, 2009; Rivas-Martínez, 2001).

In this context, we performed a bioclimatic classification of Sardinia island (Italy), based on a high resolution (40 m grid) spatial interpolation of monthly temperatures and precipitation. Because climatic information is limited to meteorological stations, that are uneven and relatively rare, we considered the choice of interpolation techniques important. In the literature, the use of geographic information systems (GIS) and multiple regression analysis on temperature and precipitation data has been demonstrated to significantly improve the mapping of climatic variables, allowing the development of an objective bioclimatic classification of a territory (e.g. Ninyerola, Pons, & Roure, 2000; 2007). The choice of interpolation methods follows the general approach suggested by Ninyerola et al. (2000, 2007) that shows good results in this specific geographic context (Boi, Fiori, & Canu, 2011). Spatial interpolation of raw monthly data was considered a better choice, rather than direct interpolation of bioclimatic indices (e.g. Mesquita & Sousa, 2009), taking into account the different spatial density between temperature and precipitation stations and how the two climatic parameters have a distinctly different spatial variability and behavior.

2. Study area

Sardinia island (Italy) is located in the middle of West Mediterranean Sea, between 38° 51′ N and 41° 15′ N latitude and between 8° 8′ E and 9° 50′ E longitude, covering approximately 24,000 km² with a coastline of about 1900 km. A complex orographic pattern characterizes the island with plain, hilly and mountainous landscapes placed on different geological substrata and the presence of a wide variety of biotopes. The highest mountain is Gennargentu Massif (1834 m) in the Central-eastern region.

Due to its geographic position, the climate is typically Mediterranean, with dry and hot summers and relatively rainy and mild winters. Rainfall ranges from 411 to more than 1215 mm in the inner mountainous regions. Measured mean annual temperature ranges from 11.6°C to 18.0°C.

In the plain and hilly areas, land use is dominated by agriculture areas, covering about half of the territory. In the inner and roughland areas, maquis and woodlands (mainly Quercus ilex, Q. suber and Q. pubescens s.l. stands), combined with ovine pastures, are prevalent (Bacchetta et al., 2009; Farris, Secchi, Rosati, & Filigheddu, 2013).

3. Methods

Data from 203 rain gauges and 68 temperature gauges, over a 30-year period (1971–2000), were homogenized and processed using the climatic database of ‘Dipartimento Meteoclimatico of ARPA Sardegna’. According to World Meteorological Organization (WMO, 1967), a minimum climatic dataset of meteorological records is considered to be at least of 30 years.

The density of stations is one in 117 km² for precipitation and one in 348 km² for temperature with a mean distance of 20 km; altitude ranges between 0 and 1060 m and sea distance ranges from 28 m to about 48 km.

To obtain objective air temperature and precipitation maps, we performed a spatial interpolation of a set of homogeneous monthly climatic variables (mean minimum temperature, mean maximum temperature and mean precipitation).

For each climatic variable, the interpolation method adopted for this study is regression kriging (RK) (Hengl, 2007), which is a two-step procedure: in the first step, multiple linear regression (MLR) was applied and, in the second step, MLR residuals were spatially interpolated using ordinary kriging (OK).

As independent geographic variables, MLR procedures include altitude, latitude, longitude, sea distance and, just for temperature interpolation, relative elevation (H_r), a topographic factor related to thermal inversion phenomenon. Elevation data, sea distance and H_r were obtained from a digital elevation model at 40 m resolution. H_r was calculated as the difference between the elevation at the grid point and the minimum elevation within a 3 km radius around the point as developed in Boi et al. (2011). Grid points with low relative elevation should be associated with thermal inversion phenomena. Raster maps have a spatial resolution of 40 m and were projected in to the UTM 32 WGS84 system.

For each climate parameter, a stepwise regression procedure was carried out in order to select the best MLR model and exclude non-significant predicting variables. As an example, temperature and precipitation regressions were obtained according to the equations reported in Table 1. These refer to precipitation (for June and for the whole year), maximum temperature (for January and October) and minimum temperature (for February and August) and outline the differences in terms of predictors and the regression coefficients used.

Table 1. Regression equations and statistics.

Regression equations		F	p
Pjune = −112.518 – 0.00002Long + 0.000031Lat + 0.011Alt + 0.00016Dist	.67	104.67	<.001
Pyear = 565.277 + 0.491 Alt − 0.002287Dist	.61	160.87	<.001
= 42.976 + 0.0000057Long − 0.0000071Lat − 0.0060Alt	.87	145.03	<.001
= 45.687 + 0.0000063Long − 0.0000065Lat − 0.00636Alt + 1.493Dist_01	.81	71.75	<.001
= 29.785 − 0.0000040Lat − 0.00579 Alt − 2.757Dist_01 + 0.14126Hr	.84	86.67	<.001
= 55.197 − 0.0000066Lat − 0.00451 Alt − 3.1185Dist_01 + 0.14735Hr	.70	39.76	<.001

Notes: Long = longitude, Lat = latitude, Alt = altitude, Dist = distance from the sea (m), Dist_01 = distance with 0.1 exponent, H_r = relative elevation,  = adjusted R² which takes into account the number of predictors, F = the model F-statistics and p = the corresponding p-value (the significant level).

The residuals of MRL were then interpolated by means of OK using a spherical model. The final maps, in this case monthly temperature and monthly and annual precipitations, were obtained by summing the two layers (MLR and OK layers). In order to estimate the prediction error for RK and compare it with other interpolation methods already used by ARPA Sardegna in climate mapping (i.e. MLR, OK), a validation procedure was carried out for each climate variable. As validation method, we adopted leave-one-out cross-validation where each sampling point is considered. The root-mean-square prediction error (RMSE) was calculated by comparing estimated values with actual observations at validation points.

For monthly precipitation, RMSE values ranged between 2.3 mm for July (29% of its mean climatic value) and 12.8 mm (14%) for November; RMSE for annual precipitation was 80.7 mm (12% of mean value). RMSE values for Regression Kriking were significantly lower than other methods considered. For monthly minimum temperature, the RMSE ranged between 0.75 °C (9%) for April and 1.06 °C (6%) for July, while for maximum temperature, the error ranged between 0.70°C for May and 0.98 °C for July (3% for both). These values were significantly lower than OK but just slightly better than MLR.

On the monthly average temperature and precipitation maps, we calculated the bioclimatic indices that have been developed as tools to explain the spatial distribution of vegetation units by the combination of different climatic factors, following the scheme of WBCS. The WBCS is based on easily calculated bioclimatic indices and numerical thresholds closely related to plant distributions. We use WBCS for this study because it works hierarchically and it is considered to well reflect the pattern of precipitation and temperature that differentiate plant communities. It integrates up to 26 climate parameters to define the categories of bioclimate. In particular, three primary indices are used: continentality, a measure of oceanic influence and temperature fluctuations; ombrotype, a measure of aridity; thermotype, a synthetic measure of temperature regime. In the WBCS, five macrobioclimates are recognized worldwide (Polar, Boreal, Temperate, Mediterranean, Tropical), with 27 bioclimates and five variants. Bioclimatic belts and horizons are defined, within every bioclimate, considering variations in temperature and rainfall amount (i.e. thermotypes and ombrotypes).

Following this approach, a bioclimatic classification of the entire territory was derived by map algebra calculations of the bioclimatic indices necessary to define the bioclimatic horizons of each grid point in the map. In particular, continentality (I_c), ombrotype (I_o), summer aridity (, , ) and thermicity (I_t) indices were obtained as reported in Rivas-Martínez, S., Rivas-Sáenz, S., and Penas-Merino (2011). Bioclimatic indices and definitions used in this paper are summarized in Table 2.

Table 2. List of bioclimatic indices and definitions.

Index	Name and definition	Formula
Ic	Continentality Index	Ic = Tmax − Tmin
Io	Ombrothermic Index
	Summer Ombrothermic Index of July + August
	Summer Ombrothermic Index of June + July + August
	Summer Ombrothermic Index of May + June + July + August
It	Thermicity Index	It = (T + m + M)*100
M	Average maximum temperature of the warmest month
m	Average minimum temperature of the coldest month
Pp	Positive annual precipitation
T	Average annual temperature
Tmax	Average temperature of the warmest month
Tmin	Average temperature of the coldest month
Tp	Positive average annual temperature

The territory was assigned to Temperate or Mediterranean macrobioclimates according to the thresholds of summer aridity expressed by the value of the ombrothermic index of the warmest two-month period (July + August) of the summer quarter ().

Regions of the study area with <2 were classified Mediterranean, otherwise they were considered Temperate. In addition, for the Mediterranean area, the thresholds reported by Rivas-Martinez et al. (2011) were used, who suggest that the values of the annual ombrothermic index (I_o) can be compensated by the summer ombrothermic values (, , ) in order to eventually shift from Mediterranean to Temperate macrobioclimate. Otherwise, the Submediterranean variant of the Temperate macrobioclimate was used when at least one month of the summer quarter had average precipitation 2.8 times less than the average temperature (I_osi P < 2.8 T).

Continentality was calculated as the range between the average temperature values of the warmest (T_max) and coldest (T_min) months of the year.

Ombrotype was calculated as the ratio between the yearly positive precipitation in millimeters (P_p) and the yearly positive temperature in Celsius degrees (T_p). The yearly positive precipitation index is defined as the total average precipitation of those months whose average temperature is higher than 0°C. Yearly positive temperature is the sum of the monthly average temperatures of those months whose average temperature is higher than 0°C.

Thermotypes were assigned based on thresholds for the thermicity index (I_t), compensated thermicity index (I_tc) and positive temperature index (T_p). However, none of the cells in the study area required compensation. The thermicity index was calculated as the sum of the yearly average temperature (T), the average minimum temperature (m) of the coldest month of the year and the average maximum temperature of the coldest month of the year (M). Finally, for each macrobioclimate, isobioclimates were identified as unique categorical combinations of the three main indices (I_c, I_t, I_o). The final map of isobioclimates was drawn at a scale of 1:250.000.

4. Results and conclusions

Two macrobioclimates (Mediterranean pluviseasonal oceanic and Temperate oceanic), one variant of Temperate (Submediterranean), four classes of continentality (from weak semihyperoceanic to weak subcontinental), eight thermotypic horizons (from lower thermomediterranean to upper supratemperate) and seven ombrothermic horizons (from lower dry to lower hyperhumid) were identified and mapped, resulting in a combination of 43 isobioclimates (Main Map).

Mediterranean types are widespread and occupy an area of about 99.1%. In the western and southern sectors (along the coast line), oceanic types (i.e. lower values of the continentality index) are common. The euoceanic (I_c value between 12 and 17) is the dominant type, whereas the semicontinental type characterizes only the inner area. Thermicity index ranges between 101 and 427, allowing the presence of eight horizons, mainly correlated with altitude, from lower thermomediterranean to lower supramediterranean. The ombrothermic index ranges from lower dry (minimum value 2.5) to lower hyperhumid (maximum value 13.4). Temperate macrobioclimate is confined in to the main mountain relief, at higher altitudes and where orographic rainfall values reduce the summer dry period and intensity. Temperate areas appear buffered by Submediterranean types, which represent the transition zones toward the Mediterranean macrobioclimate. Temperate horizons range from lower mesotemperate, lower subhumid to upper supratemperate, lower hyperhumid.

Climate variability and inhomogeneous orography of Sardinia require high detailed ecological studies at an appropriate scale. Detailed bioclimatic maps of 40 m spatial resolution are very helpful to manage both animal and plant biodiversity, because insular high levels of biodiversity are often generated by population, habitat and related processes acting at very local scales. As an example, the half of the Sardinian endemic plants are narrow endemics (Bacchetta et al., 2012), with scattered and/or fragmented populations of few individuals occupying few square meters (Farris, Fenu, & Bacchetta, 2012).

The resulting map could represent a useful environmental tool for regional planning, ecological modeling, biodiversity conservation and for applied studies of climate change.

Software

All spatial analyses were performed and processed using ESRI ArcGIS Desktop version 10. The statistical analyses were conducted using the R statistical computing environment.