Deforestation in Amazon is expected to decrease evapotranspiration (ET)
and to increase soil moisture and river discharge under prevailing
energy-limited conditions. The magnitude and sign of the response
of ET to deforestation depend both on the magnitude and regional patterns
of land-cover change (LCC), as well as on climate change and CO
The Amazon Basin provides a range of ecosystem services. The rivers
are used for navigation and hydropower; the forest is an important
global sink and store of carbon, and a store of biodiversity; evaporation
provides a water vapour source for rainfall downwind. When analysing
changes to this ecosystem, it is important to take an integrated approach
because each of these services may be affected by, or may affect,
the others. Currently, two major changes are taking place simultaneously
in Amazonia: deforestation and climate change. From the middle 1970s,
southern Amazonia has experienced widespread deforestation
Here, we focus on future changes to the river hydrology of the Amazon Basin. For different deforestation scenarios, we model the changes
in river flow from grid-based drainage and runoff estimated by different
land surface models (LSMs) driven by forcing data derived from general
circulation model (GCM) output. Because of the long transit times
of water moving from soil to the mouth of the Amazon, to simulate
discharge requires LSMs to be coupled to a river routing scheme
The climate of the Amazon Basin is notoriously difficult to model
and there is a wide between-GCM variation in the estimated precipitation
and its changes
Equally, several LSMs exist and, to a greater or lesser extent, they
all incorporate existing process knowledge into their parameterizations
Panel
A third level of uncertainty stems from the land-cover change (LCC)
scenarios used. Observed historical deforestation rates in Amazonia
are substantially different from the rates projected in the SSPs (Shared
Socioeconomic Pathways) of global scenarios (Representative Concentration
Pathways, RCPs) used for the last CMIP (Coupled Model Intercomparison
Project) assessment. This disparity questions the realism of the globally
projected rates of deforestation when considering the regional scales
Here, we apply for the first time three grid-based LSMs forced by three different GCM climate projections and more realistic regional LCC scenarios, combining the effect of uncertainty in GCM forcing data, LSM structure and LCC future scenarios, and allowing us to estimate the magnitude of likely future hydrological changes due to deforestation and climate change and their uncertainty. In particular, we discuss the relative contribution of GCM uncertainty, LSM uncertainty and LCC scenario uncertainties in future projections of runoff and evapotranspiration (ET) fluxes, with a special focus on the more vulnerable southern Amazon catchments.
The time frame studied includes a present period representing current
climate conditions (1970–2008) and 21st century projections (2009–2100).
Although the domain used in the simulations described
below includes the whole Amazon Basin (Fig.
Models used in this study.
Flow chart methodological approach for present and future simulation
processes (CC indicates climate change, LCC indicates land-cover change). Abbreviations
of the LCC scenarios are explained in Table
We used three LSMs, namely LPJmL-DGVM, which simulates daily water
budgets interactively with changes in vegetation physiology, and ORCHIDEE
and INLAND-DGVM, which operate with a 30 min time step (see Table
First, we performed an historical simulation (1850–2008) where we forced
the LSMs with pre-industrial land cover and the Princeton global climate
Using HIST as initial conditions, multiple future simulations with
each LSM forced by three GCMs (see Sect.
The projections of future climate (2009–2100) were obtained from simulations
of three GCMs from CMIP3 under the SRES A2 scenario for which sub-daily
outputs were available for driving LSMs, including the Parallel Climate
Model (PCM), the Community Climate System Model (CCSM3) and the Hadley
Centre Coupled Model (UKMO-HadCM3) (Table
List of the different simulations performed with the three LSMs (ORCHIDEE, INLAND-DGVM and LPJmL-DGVM) with or without climate change (CC) and land-cover change (LCC).
List of the GCMs participating in CMIP3 used in this study with their approximate atmospheric horizontal resolution.
The Amazon Basin is located in the countries of Bolivia, Brazil, Colombia, Ecuador, Peru and Venezuela. Each country in the basin has its own socioeconomic and institutional context-specific aspects to be taken into consideration when building scenarios in order to avoid oversimplifications. Our methodological choice was to generate new updated scenarios only for Brazil and Bolivia, the most important deforestation hotspots in the basin. The Brazilian portion of the basin covers approximately 50 % of the area, being also where most of the deforestation hotspots have been located in the previous decades. Bolivia has also been facing an intensive deforestation process for agricultural expansion around the Santa Cruz area. For the other countries, existing spatial projections were used.
The scenario process followed the “story and simulation” (SAS) approach
largely adopted in environmental scenarios
To feed the spatial model, only some selected elements of the storylines
were used – mainly concerning the natural resources theme:
(a) deforestation rates; (b) secondary vegetation dynamics; (c) roads
and protected areas network; and (d) law enforcement. The quantification
process for the Brazilian Amazon is described in
Based on these elements, future maps of forest area were then simulated
using the LuccME (Land use and cover change Modeling Environment;
LuccME
For the Brazilian Amazon, annual spatially explicit deforestation
maps from 2002 to 2013, provided by the PRODES (Program for the Estimation
of Deforestation in the Brazilian Amazon) system
To generate basin-wide LCC projections, the annual spatially explicit
results for Brazil and Bolivia were combined with the existing “business-as-usual”
projection (defined by the continuation of the current trend) to the
other countries, based on historical deforestation trends, as part
of the EU-funded ROBIN (Role Of Biodiversity In climate change mitigatioN)
project
In this paper, we explore the effects of Scenario A and Scenario C
contrasting storylines. Scenario A storyline quantification produced
low forest loss (LODEF), whilst Scenario C was quantified into a high (HIDEF)
and extreme (EXDEF) forest area loss for the Brazilian Amazon
(Table
LCC scenarios used in this study.
Changes in mean annual (mm yr
We selected two 20-year periods, 2040–2059 and 2080–2099, for LSM
output analysis. The impact of future climate change alone was estimated
for each LSM by the difference between the results of NODEF and HIST
in precipitation, ET, runoff and river discharge (Table
The spread in the ensemble mean variation (LSMs and GCM forcings)
was measured by the interquartile range (IQR). The consistency of
the variations in precipitation, ET and runoff were estimated by the
signs of the first (
We quantify the relative contribution of GCMs, LSMs and LCC scenarios
to uncertainty using an analysis of variance (ANOVA) framework as
in, e.g.
By the end of the 21st century, GCM-mean annual temperature increases
by 3.3
We focus on the period corresponding to the end of the dry season,
from August to October (ASO). Lower precipitation during this period
could have critical effects on the vegetation and hydrology
Maps indicate spatial change in
Interannual variation of forest area (10
Forest area decrease (%) over the different catchments of the Amazon Basin between each of the three LCC scenarios in 2099 and the NODEF scenario in 2009 (the abbreviations of the catchments are indicated in Table S1).
Maps indicate spatial change in
The total area of Amazonian forest prescribed in 2009 is 5.27 million km
The 8.5 % average increase of GCM-estimated annual precipitation
(190 mm yr
During ASO, the end of the dry season in the south-eastern catchments,
reduced precipitation causes a consistent decrease in ET, e.g. in
the Xingu catchment by up to 8 % (10 mm month
The same as Fig.
ET changes (mm yr
Deforestation and climate change led to a consistent decrease in
annual ET in the Amazon Basin by the end of the century of up to
2.6 % (30 mm yr
During ASO in the south-eastern catchments, ensemble-mean ET consistently
decreases by up to 11 % and runoff increases by up to 27 % in the
EXDEF scenario (Figs.
Impact of deforestation combined with climate change on ET (mm month
Seasonal river discharge (m
Deforestation-induced ET variations during the dry season are driven
by soil moisture changes which limit ET from dry soils
The amplitudes of the seasonal cycle of precipitation are different
between the GCM forcings. In the UKMO-HadCM3 model, the seasonal amplitude
is lower than in CCSM3 and PCM (compare Fig.
Relative change (%) of the first deciles (i.e. low flow, left panels)
and the last deciles (i.e. high flow, right panels) of river discharge due to
climate change (grey) and deforestation combined to climate change (three LCC
scenarios) of
The increase of runoff simulated over the catchments translates into
an increase of river discharge through the routing schemes of ORCHIDEE
and LPJmL-DGVM (INLAND-DGVM does not simulate river discharge). Because
of the small effects of deforestation and climate change on the water
budget of the entire Amazon Basin, changes in river discharge simulated
by the LPJmL-DGVM, which is already dry in regions affected by deforestation
(see above), are negligible for all the catchments (Fig.
The discharge extremes of the southern rivers (Madeira and Tapajós)
are affected by deforestation (Fig.
Although with high uncertainties, greenhouse-gas-induced climate change
will probably enhance the water cycle in Amazonia, increasing annual
precipitation, ET and runoff by the end of the century. The three
LSMs used in this study simulate an increase of ET, despite the physiological
(anti-transpirant) effect of increased CO
It has been suggested that a reduction in the area of Amazonian forest,
such as that produced by the EXDEF scenario, will push much of Amazonia
into a permanently drier climate regime
Generally, the resulting increase of runoff after deforestation is
consistent with other studies, such as LCC simulations with LSMs at
the global scale
The ET decrease and runoff increase projected for southern catchments
(Madeira and Tapajós) by the extreme deforestation scenario applied
here (EXDEF) balances the climate change effect on low flows. Climate
change alone increases the seasonal amplitude of discharge and high-flow
values. In contrast, deforestation balances this effect by reducing
the risk of decrease in low flows in the Madeira and Tapajós in all
LCC scenarios; this is related to the decrease of ET during the dry
season. The low-flow increase of the Tapajós is consistent with higher
future discharge during the dry season, in the Jamanxim subcatchment
(lower Tapajós)
ET and runoff projections are associated with large uncertainties, although
the projected sign of change is usually robust regarding deforestation.
Large climate change uncertainty, consistent with previous studies
Contributions of GCMs, LCC scenarios, LSMs and interactions between
each to total uncertainty in
To further distinguish and quantify the uncertainties which originate
from the GCMs, LSMs and the LCC scenarios, we used ANOVA. We found
that the main uncertainty source is different for ET (Fig.
Regarding runoff (Fig.
In summary, at the subcatchment scale, the magnitude of the changes in
ET first depends on GCMs and then on the behaviour of each LSM (water-limited
versus energy-limited models) in the southern catchments. Conversely,
uncertainty in runoff changes in the Amazon Basin (OBI) is first attributable
to LSMs, particularly in the south-eastern catchments, and then to
GCMs. The uncertainty attributable to LCC is low in these catchments,
suggesting some robustness in the response of the hydrology to the
deforestation. Thus, our study emphasizes the uncertainty associated
with the choice of the LSMs and their inherent (energy-limited or water-limited) parameterizations in the estimation of deforestation impacts on runoff.
Over large river basins like the Amazon, these models have the disadvantage
of being rather poorly constrained in their parameterizations of both
vegetation functioning
The construction of new land-cover change scenarios for Amazonia indicates that, by the end of this century, the total forested area of the Amazon Basin will have decreased by 7 % in the best case to 34 % in the most severe scenario. The most severe forest clearing occurs in southern Amazonia where the Madeira, Xingu and Tapajós catchments experience a 50 % decrease in forest area. With a multi-model approach, we found that the replacement of the forests by pasture and crops should only slightly decrease annual evapotranspiration by up to 2.5 % and enhance runoff by up to 2.2 %, for the most severe scenario of the Amazon Basin, compared to simulations with climate change only.
The south-eastern catchments, however, are more vulnerable at the
end of the dry season. Compared to forest, crops and pastures fail
to sustain their evaporation in a high drought stress context. Given
the combination of decreased rainfall due to future climate change
and the large forest area loss, evapotranspiration may drop by
Biosphere–atmosphere interactions, not accounted for in our study,
are also crucial in estimating the progress of forest dieback, whereby
forest is replaced by savanna vegetation. During the end of the dry
season, we found a strong reduction of ET in south-eastern Amazonia.
Evaporation at this time of year provides a critical source of water
vapour for precipitation, and lower ET can delay the onset of the wet
season
The version of the ORCHIDEE model used for this study is
Trunk.rev1311. The source code of the ORCHIDEE model can be obtained upon request
(see
The authors declare that they have no conflict of interest.
This work was financially supported by the EU-FP7 AMAZALERT (Raising
the alert about critical feedbacks between climate and long-term land-use
change in the Amazon) project (grant agreement no. 282664) and the
European Research Council Synergy grant ERC-2013-SyG-610028 IMBALANCE-P.
We acknowledge the SO HYBAM team which provided their river flow data
sets for the Amazon Basin (