The impact of accumulated deforestation in the Amazon on the regional climate
The selection of deforested cells to be compared against reference neighbors comprised areas within the moist forest biome domain with a remaining forest cover in 2021 <80%, whereas forest cover in reference cells was >80%. We grouped deforested cells by their extent of remaining forest cover in 2021, as follows: ≤40%, 40-60%, and 60-80% remaining forest cover. Our sample selection contained only deforested cells that were touching boundaries with reference cells within a 3 × 3 window and vice versa, assuming these geographically close areas share similar climate backgrounds. Additionally, we ensured deforested and reference cells had nearly unaltered forest cover during the 2013-2021 period, to highlight the role of historic deforestation, and also pairs of deforested and reference neighboring cells shared the same Köppen climate zone. The final selection of grid cells for this analysis is shown on the map of Fig. 1.
For each selected pair of deforested versus reference neighboring cells, we computed their differences in the average of climate variables during 2013-2021. Subsequently, we compared the distribution of these differences with that of reference versus reference neighboring cells. While the distribution of climate differences between deforested versus reference neighbors is assumed to have a major influence of accumulated forest loss, the distribution of climate differences between reference versus reference neighbors is assumed to have a major influence of background factors other than deforestation. We only selected pairs of reference versus reference neighbors in which the difference in their forest cover was negative or equal to zero, since forest cover differences between deforested versus reference neighbors were necessarily negative (lower forest cover in the deforested cell). The average difference in forest cover between reference versus reference neighbors was -7 percentage points (pp), and reached an average of -62 pp for our most deforested group (forest cover ≤40%) versus reference.
Differences in land surface temperature (LST) between reference versus reference neighboring cells had a balanced distribution of positive and negative values (Fig. 2), with a median difference of +0.27 °C in the dry season. The distribution of differences between deforested versus reference cells, however, shows a larger proportion of positive values (higher LST in the deforested cells), with a median of +1.14 °C in the dry season, reaching values of +4 °C or greater for the most deforested group (≤40% forest cover). Likewise, evapotranspiration (ET) in all periods showed more negative differences as the remaining forest decreased, indicating lower ET in deforested regions compared to the reference. The median difference of ET in the dry season for comparisons between deforested cells with forest cover ≤40% versus reference was -45 mm, in contrast to -4.7 mm for the reference versus reference comparisons.
For the dry season rainfall, a higher frequency of negative differences (lower rainfall) was found for deforested versus reference comparisons, especially for the deforestation groups with ≤40% or 40-60% forest cover. For instance, the proportion of comparisons with negative differences in dry season rainfall gradually increased from 53% in reference versus reference, to 59% in deforested cells with 60-80% forest cover versus reference, 63% in cells with 40-60% forest cover versus reference, and 82% in cells with forest cover ≤40% versus reference. Additionally, a higher frequency of negative differences in the number of rainy days (i.e., fewer rainy days) was found for deforested cells with forest cover ≤40% or 40-60% versus reference. The median difference for these deforestation groups against reference ranged from -8 to -11 days, in contrast to -1 day for reference versus reference comparisons.
We tested if the distribution of climate differences between deforested versus reference neighbors showed significant differences against the distribution observed between reference versus reference neighbors, based on the Two-Sample Kolmogorov-Smirnov (KS) test at the 0.05 level. We found significant differences for most climate variables and deforestation groups against reference (Supplementary Table 1), except for the annual and wet season rainfall, in which all or most deforestation groups showed non-significant differences. For the wet season ET, dry season ET, maximum dry spell duration, and the number of rainy days, significant differences were only found for the deforested groups with ≤40% or 40-60% forest cover. Similar patterns were found with the Welch's t-test for differences in the mean of distributions (Supplementary Table 1), including significant differences at the 0.05 level in all deforestation groups for the dry season rainfall and the number of rainy days, and non-significant differences in all groups for the maximum dry spell duration.
To further evaluate the consistency of our statistical comparisons, we extended our spatial representation of climate differences related to factors other than deforestation by resampling reference versus reference comparisons across the whole study area. We selected reference cells that were neighboring each other but not necessarily neighboring deforested cells (the set of dark green and faded green cells in Fig. 1), deriving one thousand random subsamples of 150 pairwise comparisons between reference versus reference neighbors. We then reapplied the statistical tests to compare the distribution of climate differences between deforested versus reference neighbors against the distribution of each reference versus reference subsample, computing the 95% confidence interval of p-values for these one thousand comparisons. Statistical tests with this approach were mostly in line with the significant differences found in previous comparisons (Supplementary Table 1), but diverged mainly by indicating significant differences in dry season rainfall only for the most deforested group (forest cover ≤40%), and additional significant differences in dry season ET for the group with 60-80% forest cover.
Differences between deforested versus reference neighboring cells for LST and ET significantly increased in intensity according to the accumulated deforested area, considering both absolute (Supplementary Fig. 1) and relative climate differences (Fig. 3). Absolute climate differences for dry season rainfall were not significantly different among deforestation groups, based on the Welch's t-test at the 0.05 level, but the deforestation group with 60-80% forest cover was significantly different from the others when considering relative differences. Differences for ET, rainfall, and the number of rainy days had a larger dispersion from the mean for deforested cells with 60-80% forest cover versus reference neighbors (Fig. 3), showing more extreme positive and negative differences against reference neighbors.
The time series from which climate averages were computed (2013-2021) includes two years with extreme drought conditions in the Amazon, 2015 and 2016, mainly driven by the El Niño and anomalous warming in the North Atlantic Ocean. To ensure that larger climate differences between deforested versus reference cells, compared to reference versus reference cells, were not biased by interannual climate variability, we reapplied the original comparisons of climate differences for two additional time intervals: the average of extreme drought years (2015 and 2016) and non-drought years (2013 and 2018). The years of 2013 and 2018 were chosen as a control against drought years for presenting some of the lowest proportions of water deficit anomalies in the Amazon since 2013.
Results from KS and Welch's t-tests for extreme drought and non-drought years showed very similar patterns when these two intervals are compared against each other, or compared with the whole time series (Supplementary Table 2), including significant differences at the 0.05 level in dry season rainfall and the number of rainy days for most deforestation groups. Extreme drought years most remarkably diverged from the other periods by presenting non-significant differences in dry season rainfall in the deforestation group with 60-80% forest cover. We also compared these different time intervals in terms of the magnitude of climate differences in the same deforestation group versus reference comparisons (Supplementary Fig. 2). Non-significant differences at the 0.01 level were found with this method within all deforestation groups and for all climate variables based on Welch's t-test, indicating similarities in the mean intensity of climate differences between deforested versus reference neighbors. Altogether, these findings converge to indicate that climate differences between deforested and reference regions, including those related to dry season rainfall and the number of rainy days, show a consistent link to deforestation and not a bias associated with extreme drought periods included in our dataset.
Among our samples, there is a wide amplitude of differences in forest cover between deforested and reference neighboring cells, which contributes to the observed variation of magnitude in climate differences (as illustrated in Figs. 2 and 3). To normalize estimates of deforestation impacts on climate among our samples, we divided each climate difference between deforested versus reference neighbors by their corresponding difference in forest cover. Additionally, to further minimize the influence of factors other than deforestation in these estimates, we removed outliers, averaged climate differences from the same deforested cell versus its reference neighbors, and averaged climate differences among our samples (see "Methods" section for additional detail). We assumed this central value is more likely linked to deforestation, while values dispersed from the mean have a greater influence of background factors. We did not estimate the impacts on annual and wet season rainfall, in addition to maximum dry spell duration, due to the unclear evidence of deforestation contribution to changes in these variables.
Our normalized estimates for deforestation impacts on climate (Table 1) represent the likely shift in the average of climate variables from the climate conditions of reference regions, for every 1 percentage point (pp) loss in forest cover. Deforestation impacts in the dry season indicate a mean LST increase of +0.05 °C (95% CI 0.04-0.06) per pp reduction in forest cover, while ET shows a mean reduction of -0.14 mm month ([95% CI -0.23 to -0.06 mm] or -0.15% [-0.25 to -0.06%]), and rainfall show a mean reduction -0.37 mm month ([95% CI -0.56 to -0.18 mm] or -0.41% [-0.59 -to -0.23%]) per pp reduction in forest cover (Table 1). Moreover, our findings indicate a mean reduction of -0.23 rainy days ([95% CI -0.31 to -0.14 days] or -0.18% [-0.24 to -0.11%]) per pp reduction in forest cover. The 95% confidence interval of our estimates shows a large variation in the range of values that would represent the actual mean deforestation impacts on ET and rainfall. Nonetheless, the range of values for annual and dry season ET, dry season rainfall, and the number of rainy days agree in their negative sign, emphasizing that deforestation is associated with reductions in these variables.
In light of the observed significant impact of accumulated deforestation on climate in the Amazon, we evaluated whether the distribution of climate averages in highly deforested regions has shifted towards climatic characteristics of savannas or forest-savanna transitional regions. From our original selection of deforested grid cells, all within the moist forest biome, we selected for this current analysis the cells with forest cover <60% to be considered as our highly deforested regions. In addition, all their immediate reference neighbors were selected as our highly forested regions (forest cover >80%), representative of the climate characteristics of the moist forest biome under minimal influence of deforestation. Different biome types, such as open grasslands and wooded savannas, also occur within the Amazon boundaries. We selected grid cells from the savanna biome and grid cells in transitional regions between savannas and moist forests to compare their 2013-2021 climate averages with those of highly deforested and forested cells.
Regions from the moist forest biome that are highly forested show very distinct climate characteristics from savannas within the Amazon, with significantly different distributions at the 0.05 level for all climate variables, based on the Two-Sample Kolmogorov-Smirnov test, except for wet season ET and maximum dry spell duration. They also show significant differences from the distributions in forest-savanna transitional regions. However, in highly deforested regions, distributions for LST variables and annual ET are significantly similar to transitional regions (Table 2). Additionally, highly deforested regions have non-significant differences for dry season ET and rainfall compared to both highly forested regions and transitional regions, showing mean values in between these two biome categories.
Although the annual rainfall and the number of rainy days in highly deforested regions are not significantly different from highly forested regions, there is a notable shift towards lower values (Fig. 4). During the dry season, LST in highly deforested regions is on average 3 °C higher than in highly forested regions but 3 °C lower than in savannas. In addition, the average dry season ET is 12% lower (-40 mm) than in highly forested regions, but 44% greater (+92 mm) than in savannas, whereas dry season rainfall is 25% (-50 mm) lower than in highly forested regions but 60% greater (+56 mm) than in savannas. Moreover, the number of rainy days in highly deforested regions is, on average, about 11 days lower than in highly forested regions, but 24 days greater than in savannas.
To further evaluate deforestation-related shifts in climate envelopes, we compared the selected highly deforested and forested cells in terms of their conformance with the climate envelope of savannas. We identified the climate envelope in which most savannas in the Amazon are found based on the observed 5th and 95th percentiles of climate averages from the selected cells of this biome (Table 2). For instance, most Amazonian savannas occur where the dry season LST is ≥31 °C, dry season ET is ≤309 mm, and dry season rainfall is ≤192 mm (Table 2). For each climate variable, we then assessed the proportion (%) of cells from highly deforested and forested regions that are within the savanna envelope. This analysis revealed that about 93% of our highly deforested cells were in the savanna envelope for dry season LST (Supplementary Table 3); however, a large proportion of highly forested cells (68%) were also under this envelope. More remarkably, about 67% of highly deforested cells were under the envelope for annual LST, in contrast to only 16% in highly forested cells. The vast majority of highly deforested regions are also in the savanna envelope for annual ET (78% of cells against 45% in highly forested cells) and dry season rainfall (70% against 55%), while the dry season ET and wet season rainfall also showed substantial contrasts between deforested and forested cells.
We also mapped the spatial distribution of areas across the Amazon from the moist forest biome whose climate is under the envelope for savannas. This analysis was performed for four variables (annual LST, annual ET, wet and dry season rainfall) and by distinguishing moist forest areas by their forest cover extent (Fig. 5). Areas in which the four variables are in the savanna envelope, thus showing the largest similarity to the savanna climate, are found in the Brazilian states of Rondônia and mid-eastern Pará. The vast majority of these areas are highly deforested (<60% forest cover), contrasting with the highly forested areas in their surroundings that show less similarity with the savanna envelope. Other regions in which deforested areas show a greater climate similarity to savannas are found in northern Mato Grosso state in Brazil, southern Pará, and in the western Colombian Amazon. When each climate variable is analyzed separately (right panel in Fig. 5), we observed that most of the southern Amazon is under the savanna envelope for dry season rainfall, and a large portion of the western Amazon is under the envelope for annual ET and wet season rainfall. Conversely, annual LST and ET show greater spatial distinctions between highly deforested and forest areas, especially in the eastern and southern Amazon.