Results and discussion

Grouping of regions

The results of the cluster analysis show with an error of 51.4% that there are four main groups. All these groups experience increments in temperature, and the differences among them are mainly determined by the changes of precipitation during the year. The first split among these groups, and that explained 30% of total variation, was given between those areas that are going to experience decreases of precipitation and those that are going to experience increments. Next, for the areas with decreases of precipitation, a new split that explained 11% of total variation was formed and it was driven mainly by the time in which these decreases of precipitation are going to occur. For the groups with increments of precipitation, a new split that explained 6% of total variation was formed, and it appears to be driven not only by the time in which these increments of precipitation are going to occur, but also for their magnitude.

Figure 5. Results of the cluster analysis.

Following the labels assigned by the cluster analysis, group 2 is characterized for having a decrease of precipitation near to 15% for the December-February period, and is comprised mainly by the North and Center regions of Mexico, Guyana, Suriname, French Guiana, Chile and South region of Argentina (Figure 6). Group 3 is characterized for decreases of precipitation near 9% for the period that extends from June to November, and is comprised by regions such as South region of Mexico, most of Central America, Venezuela, most of Brazil and the interior faces of the Andes mountain region (Figure 6). Group 1 is characterized for increments of precipitation of 5% in a period that goes from September to February, and is comprised by regions such as South of Brazil, Paraguay, and the exterior faces of the Andes mountain region (Figure 6). Group 4 is characterized by increments of precipitation of 16% and 9% for the periods of December to February and March to May, respectively, and it is comprised by regions such as Costa Rica, Panama, Colombia, and the Uruguayan and Argentine Pampa (Figure 6). 

Figure 6. Projected changes in temperature and precipitation by season for the year 2050.

Comparison of methods of modelling of distribution

The comparison of the methods for the modelling of distribution of biomes showed that RF had a better performance with a model error of 10.6 %, while LDA had a model error of 35.7%. The matrix of misclassifications for RF showed that overall, this method had an accuracy above 80% for almost all classes, except for Tropical and Subtropical Dry Broadleaf Forests (71.5%), and Tropical and subtropical moist broadleaf forests (78.2%) (Table 2). On the other hand, the matrix of misclassifications for LDA showed that this method had varying results for the prediction of all classes but really struggled with Tropical and Subtropical Dry Broadleaf Forests (17%) (Table 3). These results agree with the reports from the literature, where Random Forest always shows a greater accuracy than LDA, especially when there is a strong interaction among the predictor variables (Cutler et al., 2007).

Table 2. Matrix of missclassifications for Random Forest.

	Biome	1	2	3	4	5	6	7	8	9	10	11
1	Deserts and Xeric Shrublands	87.8	0.2	1.4	1.3	2	0	0	3.5	2.4	0.7	0.7
2	Flooded Grasslands and Savannas	0	94.7	2.2	0	0	0	0.7	0	1.3	1	0.1
3	Mangroves	0.6	1.5	91.9	0	0	0	0	1.3	2.5	0.3	1.9
4	Mediterranean Forests Woodlands and Scrub	0.8	0	0	95.5	0.8	2.8	0.1	0	0	0	0
5	Montane Grasslands and Shrublands	0.3	0	0	0.8	94	0.7	1.6	0.1	1.3	0	1.2
6	Temperate Broadleaf and Mixed Forests	0	0	0	4.1	0.1	94.6	1.2	0	0	0	0
7	Temperate Grasslands Savannas and Shrublands	0	1	0	0	0.6	3.4	94.3	0	0.4	0.3	0
8	Tropical and Subtropical Coniferous Forests	1.1	0.1	3.5	0	0	0	0	91.7	2.7	0	0.9
9	Tropical and Subtropical Dry Broadleaf Forests	3.5	2.4	7.7	0	1.7	0	0.7	6.8	71.5	3.1	2.6
10	Tropical and Subtropical Grasslands Savannas and Shrublands	1.5	2.4	0	0	0.1	0	0.1	0.1	2.1	88.8	4.9
11	Tropical and Subtropical Moist Broadleaf Forests	0.5	0.5	5.2	0	1.9	0	0	3.3	2.4	8	78.2

 

Table 3. Matrix of missclassifications for Linear Discriminant Analysis.

	Biome	1	2	3	4	5	6	7	8	9	10	11
1	Deserts and Xeric Shrublands	51.3	1.7	5.4	5.1	3.6	0	0.7	8.8	21.2	0	2.2
2	Flooded Grasslands and Savannas	0.4	72.8	4.7	0	0	0	14.4	0	0	7.1	0.6
3	Mangroves	12.5	0	55	0	0	0	0	3.9	4.6	3.4	20.6
4	Mediterranean Forests Woodlands and Scrub	2.8	0	0	84.2	3.6	8.2	1.2	0	0	0	0
5	Montane Grasslands and Shrublands	0.1	0.4	0	2.2	74.4	8.9	13.3	0	0.5	0.2	0
6	Temperate Broadleaf and Mixed Forests	0	0	0	5.9	11.1	82.8	0.2	0	0	0	0
7	Temperate Grasslands Savannas and Shrublands	0	0	0	0.7	11.1	2.2	86	0	0	0	0
8	Tropical and Subtropical Coniferous Forests	5.8	0	10.4	1.3	0.1	0.1	0	82.3	0	0	0
9	Tropical and Subtropical Dry Broadleaf Forests	9.2	30.7	18.5	0	1.8	0	7	10.9	17	3.8	1.1
10	Tropical and Subtropical Grasslands Savannas and Shrublands	0	22.8	13.3	0.6	0	0	9	0	0.5	51.6	2.2
11	Tropical and Subtropical Moist Broadleaf Forests	1.5	4.4	17.4	0.3	1	0.6	1.1	3.1	0.9	20.6	49.1

 

Based on the Random Forest predictions for the year 2050, it is expected a reduction of the suitable climate habitat for biomes such as Tropical and Subtropical Coniferous forest (-1.14%) and Tropical and Subtropical Moist Broadlead forest (-1.84%). On the other hand, it is predicted that there will be an increment in the suitable climate habitat for biomes such as Tropical and Subtropical Dry Broadleaf forest (+2.8%), Tropical and Subtropical Grasslands Savannas and Shrublands (+0.9%) and Deserts and Xeric Shrublands (+0.76%) (Figure 7)(Table 4). These results agree with previous studies where it has been predicted that there will be a transformation of Tropical Forests in Eastern Amazonia to Savannas (Marengo et al., 2012).

Figure 7. Predictions of change in the distribution of the biomes of the Neotropic.

The changes in the altitudinal ranges of distribution show that there will be a retraction of the lower altitudinal range of those biomes distributed along altitudinal gradients, such as Tropical and Subtropical Coniferous Forests and Montane Grasslands and Shrublands, and the most likely thing is that other biomes will expand unto this freed range. For Coniferous forests, these results agree with reports of previous studies, where it has been predicted that the suitable climate habitat for many species distributed along the Mountain ranges of Central Mexico will shift higher along the altitudinal gradient (Saenz-Romero et al., 2012).


Table 4. Percentages of coverage of area and altitudinal ranges (masl) for each biome for predictions of current climate and for the year 2050. LQ=lower quartile UQ=Upper quartile.

Biome	Curr_Pred (%)	2050_Pred (%)	Curr_LQ	Curr_UQ	2050_LQ	2050_UQ
Deserts and Xeric Shrublands	9.92	10.68	276	1376	293	1429
Flooded Grasslands and Savannas	1.22	1.77	70	131.5	112	328
Mangroves	0.50	0.07	3	15	8	144
Mediterranean Forests Woodlands and Scrub	0.84	0.91	253.5	1385	248	1279.25
Montane Grasslands and Shrublands	4.28	3.59	3376	4327	3486	4403
Temperate Broadleaf and Mixed Forests	1.91	1.78	174	1038.5	165	990
Temperate Grasslands Savannas and Shrublands	7.96	7.08	97	573.75	108	656
Tropical and Subtropical Coniferous Forests	2.68	1.54	1030	2227	1595	2410
Tropical and Subtropical Dry Broadleaf Forests	9.67	12.48	136	526	107	602.5
Tropical and Subtropical Grasslands Savannas and Shrublands	15.70	16.63	135	551	162	514
Tropical and Subtropical Moist Broadleaf Forests	45.30	43.46	106	364	99	382