The elaboration of conservation strategies at regional scale, dealing with the potential effects of climate change on the abundance and distribution of tree species, should be supported by models produced at the appropriate scale. We used a bioclimatic model aimed at analysing the large-scale effects of climate change on the abundance and distribution of tree species with respect to their chorological and ecological characteristics. Abundance data for 16 species, sampled in 912 plots, distributed on a 3x3 km grid were used. A climatic model provided high resolution current climatic surfaces and a climatic scenario for 2080 was obtained using the A1FI emission scenario of HadCM3 GCM. A deterministic Regression Tree Analysis (RTA) and Multiple Linear Regression (MLR) were applied in order to define the realised niche of the species in relation to the chosen environmental variables. The comparison between RMSE values showed that RTA always outperforms MLR, in terms of predicting species distribution. Zonal species were better predicted than rare species (extrazonal or with specific habitat requirements). Climate change is expected to determine a general increase of the average potential altitude. Only the Mediterranean species are likely to be favoured by the predicted climate change, while for the two other chorological types (Sub-Mediterranean and Eurosiberian) the response seems to be species-specific, depending on the ecological characteristic of each species: the more thermophilous and xerophilous species should benefit from the predicted drought in terms of area and mean abundance, while mesophilous species should suffer a strong reduction.
The effects of climate change on ecosystems are already noticeable (
Will the expected increase in average altitude of species’ range in the area considered be one of the main effects of climate change?
Do species belonging to the same chorological type respond in a similar way?
Is it possible to explain the responses of the species using their Ellenberg’s temperature and moisture scores (
The study area consists of two Italian regions (Lazio and Abruzzo) located in Central Italy between the Tyrrhenian and Adriatic coasts and covers about 28000 km2 (
Forests, following the FAO forest definition (
The data source for this study was the IN.DE.FO. of Italy for the study area (
where
We analysed sixteen species that are the most abundant in the study area:
Very high resolution (30 arc-seconds) current climatic surfaces were obtained by interpolating climatic data (average of the 1961-1991 period) from 210 stations for precipitation and 151 for temperature through universal kriging (
Climatic data which influence plant survival and growth were chosen: mean annual temperature, mean minimum temperature of the coldest month, mean maximum temperature of the hottest month, annual, summer and winter precipitations. Moreover, a moisture index was used based on the following formula:
where
where
The annual potential solar radiation was calculated using a specific module implemented in the Grass software GIS (
A slope map and a simplified geological map (scale 1:100.000, Servizio Geologico d’Italia, APAT) were also used. The geological map was a surrogate for pedological information. ArcGIS 9.0 was used to collect, store and manipulate IV as well as environmental data in raster format. The complete list of environmental variables is shown in
A Regression Tree (RT) and a Stepwise Multiple Linear Regression (MLR) models were compared in order to quantify the relationship between environmental factors and the species abundance. The theory of MLR is quite well known and will not be reported here (interested readers should refer to
We randomly put aside 30% of the data (test set), and trained the models on the remaining 70% (training set). The RTA and the MLR were evaluated on the test set using the mean RMSE between the current and predicted IVs for each species. The best model was evaluated by calculating both the correlation between current and predicted IV as well as the so-called Omission Accuracy (OA) measure. To calculate this last measure, we categorized the observed and predicted IVs into binary variable indicating if the values were over a certain threshold. The threshold was set at 2 IV since the values below were characterised by a high uncertainty. For each species, the OA formula is (SP/(SP+SO)*100), where SP is the number of areas with observed values over the threshold where the species was correctly predicted as present and SO is the number of areas with observed values over the threshold where the species was incorrectly predicted as absent. Because of the sampling approach we used, classification accuracy was not calculated for the test set; indeed, the absence of a species at a given sampling plot does not imply its absence over the 3 x 3 km grid.
The results obtained through the final model were used to produce maps of both current and future potential distribution under a scenario of changed climate assuming unlimited dispersal capacity of the species. The predicted potential distribution was obtained by replacing the climate variables with those obtained from the HadCM3 GCM; predicted values were then imported into ArcGIS to produce the corresponding maps.
The effects of climate change on species distribution were evaluated by calculating the percentage changes in the potential area as well as on the mean IV and altitude.
The comparison between RMSE values corresponding to both models for all of the analysed species shows that RTA always outperformed MLR, in terms of predicting species distribution for the training set (
The tree diagrams produced by RTA analysis are shown in
According to the estimated model,
RTA had a better performance in modelling the spatial distribution of all the analysed species compared to MLR (
Using results from RTA, a significant correlation between number of sample plots where the species was recorded and model accuracy was found (
The main chorological component was represented by Sub-Mediterranean deciduous species (8 species), showing on average the best accuracy values (
The deciduous tree species are more abundant than the sclerophyllous evergreen ones in terms of species number and spatial distribution. This result depends on the fact that the study area, despite its Mediterranean location, is characterized mainly by a temperate climate because of the effects of the Apennine mountains, which intercept the humid winds from the sea, thus causing a rainfall increase (
Other examples of tree species expansion due to human intervention are those of
A general increase of the average potential altitude is predicted applying the HADCM3 climate scenario for year 2080. However if area and abundance are analysed, four different trends can be identified: species that could be advantaged by the predicted climate change (
According to the chorological types, the two Mediterranean species (
RTA has proven to be efficient for the identification of the environmental variables driving the current distribution and abundance of tree species at a regional scale. Only Mediterranean species are likely to be favoured by the predicted climate change, while for the two other chorological types (Sub-Mediterranean and Eurosiberian) the response seems to be species-specific depending on the autoecological characteristic of each species. A more complete picture of the potential effects of climate change would be obtained by analysing the whole Italian peninsula considered as a biogeographical unit and including a detailed land use map as a measure of fragmentation and anthropization of the territory, influencing the expected shift of species.
This work has been carried out in the framework of CONECOFOR (
The study area.
Comparison between the RMSE value of the regression tree analysis (black diamonds) and that of the multiple linear regression (black triangles) for all the species.
Tree diagrams for A -
Current and future predicted potential distribution of
Current and future predicted potential distribution of
List of the selected environmental variables.
Environmental variable | Acronym |
---|---|
Mean Annual Temperature | MAT |
Mean Maximum Temperature of the Hottest Month | MTH |
Mean Minimum Temperature of the Coldest Month | MTC |
Annual Precipitation | AP |
Summer Precipitation | SP |
Winter Precipitation | WP |
Moisture Index | Mi |
Slope | Slo |
Geology: - Alluvial sediments - Arenaceous rocks - Carbonated rocks - Clayey formations - Sands - Volcanic rocks | - |
Validation of the regression tree model. N is the number of sample plots. R is the correlation between measured and predicted IV. OA_Ve and OA_Va are the omission accuracy for the whole data set (verification) and for the 30 % of the validation data set respectively. Abbreviations for chorology: M, Mediterranean; SM, Sub-Mediterranean; E, Eurosiberian.
Species | Chorology type | N | R | OA_Ve | OA_Va |
---|---|---|---|---|---|
|
M | 102 | 0.77 | 0.79 | 0.68 |
|
M | 43 | 0.63 | 0.77 | 0.67 |
Mean | - | - | 0.70 | 0.78 | 0.68 |
|
SM | 106 | 0.72 | 0.67 | 0.65 |
|
SM | 85 | 0.63 | 0.85 | 0.74 |
|
SM | 54 | 0.53 | 0.63 | 0.61 |
|
SM | 85 | 0.78 | 0.89 | 0.85 |
|
SM | 141 | 0.63 | 0.76 | 0.68 |
|
SM | 177 | 0.74 | 0.81 | 0.75 |
|
SM | 287 | 0.81 | 0.97 | 0.92 |
|
SM | 41 | 0.60 | 0.82 | 0.71 |
|
SM | 361 | 0.80 | 0.98 | 0.86 |
Mean | - | - | 0.69 | 0.82 | 0.75 |
|
E | 88 | 0.68 | 0.84 | 0.75 |
|
E | 73 | 0.62 | 0.60 | 0.64 |
|
E | 263 | 0.90 | 0.99 | 0.95 |
|
E | 59 | 0.57 | 0.79 | 0.69 |
|
E | 50 | 0.76 | 0.70 | 0.72 |
Mean | - | - | 0.71 | 0.78 | 0.75 |
Current, future and percentage of change of the potential area (km2) and average IV and altitude (m); the percentage changes are due to the HadCM3 model outputs; Ellenberg’s scores for temperature (T) and moisture (M). In the last columns, X means the species has a wide spectrum for that parameter. Abbreviations for chorology: M, Mediterranean; SM, Sub-Mediterranean; E, Eurosiberian.
Group | Species | Chorology type | Area (km2) | IV | Altitude (m) | T | M | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Current | Future | % | Current | Future | % | Current | Future | % | |||||
Group 1: +IV +Area |
|
SM | 24417 | 24832 | 1.7 | 46 | 67.3 | 46.3 | 408.3 | 576.1 | 41.1 | 8 | 4 |
|
M | 9423 | 16453 | 74.6 | 37 | 37.9 | 2.4 | 212.4 | 342 | 61 | 9 | 3 | |
|
SM | 24732 | 27947 | 13 | 60.7 | 81.7 | 34.6 | 414.3 | 541.5 | 30.7 | 8 | 3 | |
|
M | 3006 | 5282 | 75.7 | 40.7 | 48.1 | 18.2 | 75.2 | 79.3 | 5.5 | 8 | 3 | |
|
E | 2682 | 2926 | 9.1 | 24.9 | 33.4 | 34.1 | 163.2 | 216.6 | 32.7 | 7 | X | |
Group 2: -IV -Area |
|
SM | 3393 | 1863 | -45.1 | 13.4 | 11.6 | -13.4 | 698.6 | 1159 | 65.9 | 5 | 6 |
|
E | 2115 | 584 | -72.4 | 12.7 | 10.6 | -16.5 | 972.7 | 1297.6 | 33.4 | 5 | 6 | |
|
SM | 5175 | 5051 | -2.4 | 60.9 | 38.4 | -36.9 | 578.2 | 932.6 | 61.3 | 8 | X | |
|
E | 8946 | 6352 | -29 | 83.5 | 81 | -3 | 1144.2 | 1330.7 | 16.3 | 5 | 5 | |
|
SM | 7497 | 4183 | -44.2 | 24.8 | 23 | -7.3 | 736.5 | 1082.7 | 47 | 8 | 4 | |
|
E | 5634 | 3747 | -33.5 | 28.9 | 21.7 | -24.9 | 238.2 | 248.2 | 4.2 | 6 | 6 | |
Group 3: -IV +Area |
|
SM | 6255 | 7400 | 18.3 | 25.3 | 18.5 | -26.9 | 330.5 | 358.3 | 8.4 | 7 | 5 |
|
E | 3645 | 6616 | 81.5 | 14.7 | 8.1 | -44.9 | 442.7 | 862.8 | 94.9 | 6 | X | |
Group 4: +IV -Area |
|
SM | 1242 | 532 | -57.2 | 18.2 | 30.9 | 69.8 | 616.8 | 1244.1 | 101.7 | 7 | 3 |
|
SM | 7101 | 5681 | -20 | 13.4 | 13.7 | 2.2 | 468.6 | 380 | -18.9 | 8 | 3 | |
|
SM | 3645 | 2278 | -37.5 | 12.8 | 21.7 | 69.5 | 63.7 | 79.8 | 25.3 | 6 | 6 |