https://orcid.org/0000-0002-2165-5239
Figures
Abstract
Climate change over the Amazon basin has the potential to cause major hydrological and ecological impacts over the region’s extensive wetlands. To investigate this the Joint UK Land Environment Simulator (JULES) land surface model is first extended to include riverine inundation. Potential impacts of future climate change on Amazon basin wetlands are then evaluated by driving this updated JULES model with modelled meteorology projections from six different climate simulations reaching approximately 4°C global warming at the end of the 21st Century. The projected changes in inundation extent and seasonality are assessed over four major wetland regions. The simulations project, on average, a significant decrease in total Amazon basin inundated area of 11% (range: -36% to +9%) by the 2090s. This considerable spread is primarily driven by disparity in simulated precipitation changes, ultimately driven by sea surface temperature differences. The wetter contemporary climate simulations simulate the greatest drying by the end of this Century, resulting in the largest wetland area reductions. The largest qualitative disagreement is over the western Iquitos wetland, with inundated area changes ranging from a very large reduction of -53% to an increase of 12%. A new wetland classification scheme is developed to summarise the projected changes in wetland seasonality. The largest drops in simulated wetland season length occur over the Central/East Manaus and West Iquitos wetland regions, with reductions of up to 10 and 8 months respectively. Such significant changes in future inundation are likely to have a major impact on regional wetland hydrology and their ecosystems.
Citation: Gedney N, Rudorff C, Betts RA (2024) Future amazon basin wetland hydrology under projected climate change. PLOS Water 3(9): e0000225. https://doi.org/10.1371/journal.pwat.0000225
Editor: Ritesh Kumar, Wetlands International South Asia, INDIA
Received: October 10, 2023; Accepted: June 26, 2024; Published: September 30, 2024
Copyright: © 2024 Gedney et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The JULES simulation data can be accessed at doi:10.5281/zenodo.10459886. The MERIT DEM [45] is accessible online (https://hydro.iis.utokyo. ac.jp/~yamadai/MERIT_DEM/, last access: 14 December 2023). The HESS [19,20] product is available online (https://daac.ornl.gov/LBA/guides/LC07_Amazon_Wetlands.html, last access: 15 May 2024). The GIEMS [10,40] product can be acquired by contacting Catherine Prigent (catherine.prigent@obspm.fr). The SWAMPS-GLWD [41] product can be acquired by contacting Zhen Zhang (zhenzhang@itpcas.ac.cn). The WAD2M [43] product is available online (https://zenodo.org/records/3998454, last access: 15 May 2024). The WFDEI [48] dataset is available from the ftp site at IIASA, Vienna (ftp.iiasa.ac.at, username: rfdata password: forceDATA, then use: cwd/WFDEI, last access: 15 May 2024). The HELIX [37] data can be acquired by contacting John Ceasar (john.ceasar@metoffice.gov.uk). Upon publication the JULES simulation data can be accessed at doi: 10.5281/zenodo.10459886.
Funding: NG and RAB acknowledge support from the Newton Fund through the Met Office Climate Science for Service Partnership Brazil (CSSP Brazil). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
Globally wetlands cover an area of approximately 12 million km2 [1], which is about 9% of total, ice-free land. They play a key role in both hydrological and biogeochemical cycles and are one of the world’s most biodiverse ecosystems [2]. Wetlands reduce soil erosion and mitigate floods and droughts [2]. They provide a range of ecosystem services including fish and rice production, and wood and peat supplies [3]. Wetlands also store a significant amount of carbon, with the highest carbon density of all terrestrial ecosystems, making them an important sink for atmospheric carbon dioxide (CO2) [4]. Due to their anoxic environment wetlands are a significant natural source of methane (CH4) [5].
Climate change is likely to impact wetlands predominantly through changes in hydrology and temperature [3], which combined with increasing atmospheric CO2 [6], can alter water table depths, vegetation composition and productivity, respiration and thereby modify their CO2 and CH4 emissions. As such, climate change has the potential to cause local impacts which may feed back onto climate change [6]. How wetlands respond to climate may therefore affect flooding, ecology and the anthropogenic emission reductions required to meet the Paris Agreement climate targets [6,7]. In spite of their significance, there is considerable uncertainty both in future climate change projections and how wetlands are likely to respond to them [3,6]. Moreover, the quantification of their present-day extent remains uncertain over large areas of the globe, especially in regions with dense vegetation cover. Estimates of global wetland area have increased with time as better products have become available, although it is likely that the real wetland area continues to be underestimated [8].
Dynamic representation of wetland inundation within climate models is therefore a key challenge in improving the simulation of hydrology [9] and biogenic gas fluxes over continental regions [6]. Many applications would benefit from the treatment of the surface waters as interactive components of climate models and not simple static parameterizations [10–12]. Floodplain models vary from simple statistical techniques [12] to detailed 2-D floodplain flows [13]. Simpler models tend to be used at the large scale as detailed topography is not explicitly resolved in these global and regional models [14].
The Amazon basin contains some of the largest wetlands in the world. Seasonal inundation is a significant hydrological feature here [15], with the timing, duration and extent of flooding directly driving ecosystem functions and biodiversity [16] where many species are specifically adapted to the conditions [17]. For example, fish stocks and diversity have been shown to be related to flooded forest extent, with stocks decreasing in drought years with less flooding and higher fish mortality [16,18]. Saunois et al. [17] classify the Amazon basin wetlands into 14 different major wetland types. These classifications include vegetation type, fertility state and whether wetlands are dominated by riverine (fluvial) flooding, or interfluvial wetlands which are inundated by local precipitation and runoff. Reis et al. [15] use a wetland classification system over the Amazon basin which incorporates both the timing and duration of the major wetlands as an aid to help with regional wetland management. Reis et al. [15] characterise the hydrological differences of the major interfluvial Amazon basin wetland complexes which are in the Roraima, Iquitos and Llanos de Moxos regions (in the North, West and South of the basin respectively), and the central/eastern Amazon main-stem floodplain in the Manaus region. These wetlands differ in their land cover: Roraima is a mix of mainly forest, woodland and shrub; Iquitos is predominantly forest; Llanos de Moxos is largely herbaceous vegetation and forest; and Manaus is mainly open water and forest [19,20].
There have been multiple studies estimating Amazon basin inundation extent and seasonality, including use of multiple datasets at various resolutions, optical sensors, synthetic aperture radar (SAR) and hydrological models [21]. There is considerable spread even within these approaches. For example, SAR estimates range between 0.4–2.8 and 5.1–6.3x105km2 for the minimum and maximum Amazon basin inundation respectively [21]. Uncertainty in contemporary Amazon basin inundation extent impacts on the uncertainty in estimates of its present-day wetland CH4 emissions [6], which at between 44 to 52 TgCH4yr-1, is roughly 25% of global wetland emissions [22,23]. Detailed knowledge of their contemporary spatial and temporal extent is therefore important for estimating the current greenhouse gas budget [6].
Statistically significant trends in recent historical hydrology have already been observed [24], with increases in precipitation and river discharge in the northern and western parts of the basin and decreases in river flow in the South and East. These changing stream flows have been attributed variously to climate change, land-use change and associated water vapour recycling and evapotranspiration [25,26]. Glacial melting in the Andes can also affect river levels in basins fed by meltwater [27].
Projections of future precipitation over the Amazon basin are highly uncertain [28,29] but generally simulate a decrease in rainfall over the eastern part of the basin and increase over the western region [30–32]. By 2100, CMIP5 (Coupled Model Intercomparison Project 5) RCP8.5 (Representative Concentration Pathway greenhouse gas concentration scenario 8.5) simulations generally show a reduction in precipitation all year round in the southern and eastern Amazon basin, with a substantial relative reduction mainly in the dry and transition seasons [33]. CMIP6 simulations show a stronger consensus over long-term Amazon basin rainfall reduction than CMIP5, particularly over the southern and eastern regions [34]. There are associated projected reductions in river discharge, but with uncertainties in their magnitude [31].
[35] use 24 IPCC AR4 [36] climate simulations to estimate inundation changes using an empirical relationship between modelled discharge and inundation. The majority of these predict a future increase in western Amazon basin inundation, with less consensus over the east of the basin. Sorribas et al. [31], using a subset of CMIP5 simulations to drive a regionally calibrated hydrology model, show a statistically significant increase in future inundated area in the West of the basin and potential decreases elsewhere.
This study seeks to further understand the potential impact of future climate change and its uncertainty on the wetlands in the Amazon basin, while addressing the need for a dynamic inundation model which can be fully integrated into the UK community Earth system model (UKESM). The 6 climate simulations [37] utilised in this study are atmospheric-only whose sea surface temperatures (SSTs) and sea ice patterns are taken from a sub-set of CMIP5 simulations. These have the advantage that, not only were they carried out at a higher resolution than the CMIP5 simulations making them more suitable to regional scale impact assessments, but they also allow us to explore the regional hydrological uncertainty due to SST patterns.
Here the Joint UK Land Environment Simulator (JULES) land surface model [38,39] (Section A and Section B in S1 Text), which includes vegetation and carbon dynamics, is adapted to include a parameterisation of fluvial inundation caused by rivers overflowing their banks or “overbanking”. This parameterisation (summarised in the Materials and Methods Section, and described in detail and validated in Section C in S1 Text) is combined with the inter-fluvial (groundwater) inundation model already in JULES (described and assessed in [22 SI]). It is first driven with contemporary observed meteorology and validated against EO-based inundation products. JULES is then forced with the six climate projections, which have a range of changes in precipitation, to investigate the hydrological response to projected climate change. The analysis here focusses on the impact of climate change on the major wetlands in the Roraima, Iquitos and Llanos de Moxos and the Manaus regions (Fig 1).
Open water in both low and high water seasons (blue), seasonally flooded trees (green) and shrubs and herbaceous vegetation (yellow). Non-wetland (dark grey) areas and areas above 500m elevation in the Amazon basin (light grey). The major wetland regions discussed in the text are outlined by rectangles: North–Roraima, West–Iquitos, South–Llanos de Moxos, Central/East: Manaus. The coastline is from the cartopy python module which is from the Natural Earth “coastline” shapefile collection https://www.naturalearthdata.com/downloads/10m-physical-vectors/10m-coastline/.
2. Materials and methods
2.1 Inundation products used in model validation
The global, but relatively coarse scale, remote sensing based products GIEMS (Global Inundation Extent from Multi-Satellites) [10,40], SWAMPS-GLWD (Surface WAter Microwave Product Series) [41] and WAD2M (Wetland Area and Dynamics for Methane Modeling) [5] are utilised here. GIEMS covers the period of 1993–2007 and uses multiple satellite observations including visible and near infrared, passive and active microwave measurements, which enables improved detection of flooding even under dense vegetation canopies. SWAMPS-GLWD combines remote sensing data with the mapping Global Lakes and Wetlands Dataset [42], excludes permanent open water (water bodies, rivers, snow/ice) and covers the period 2000 to 2012. WAD2M has been developed to address some of the known issues of its predecessor SWAMPS-GLWD and covers the period 2000–2017. A MODIS-based product is used to exclude freshwater (lakes, ponds, reservoirs, streams and rivers) extent for SWAMPS-GLWD whereas a higher resolution Landsat-based product [43] is used to remove freshwater for WAD2M. In the WAD2M dataset production, freshwater is defined as the area where surface water (determined from satellite imagery) persists for more than 6 months. The Hess regional dataset [19,20] is also used in this study. It is a higher resolution product, and as such is more likely to detect inundation than the other coarser resolution satellite-based products [19]. It only covers a two short observation time periods however: October to November 1995, and May to July 1996.
For increased consistency across all datasets, the freshwater terms are added back into the SWAMP-GLWD and WAD2M datasets and are referred to as SWAMPS-GLWD+fw and WAD2M+fw respectively. Since there is no temporal information on the “permanent” inland water removed from WAD2M dataset, here it is assumed to occur over the whole calendar year and so this freshwater term is added to every month.
2.2 JULES hydrology summary
Fluvial inundation is parameterized using estimates of river depth from the modelled river flow [44] and is described in detail in Section C in S1 Text. The water surface elevation is estimated based on the statistical relationship between hydraulics and geomorphology of stream channels. This is combined with the hypsometric curve (calculated from the Multi-Error-Removed Improved-Terrain (MERIT) DEM [45]; https://hydro.iis.u-tokyo.ac.jp/~yamadai/MERIT_DEM/, last access: 14 December 2023) to give the river inundation extent in each grid box. Here the term fluvial inundation fraction is defined as the fraction of surface river water in the grid box, and therefore by definition includes the river width. Including the river width within the definition is likely to over-estimate the overbank inundation however there are many uncertainties in fluvial modelling at this scale, including estimations of river channel dimensions, accuracy of the digital elevation model (DEM) [14] and uncertainties in observed inundation extent [21] to calibrate against. The scheme does not explicitly model the movement of water between river channel and floodplain and of the infiltration of flood water into the soil.
In addition to the river inundation, the inter-fluvial inundation is calculated within each grid box by combining topographic data and simulated soil moisture. The JULES model simulates the vertical profile of soil moisture within each grid box, which is dependent on many factors, including climate, soil properties, variation of hydraulic properties with depth and vegetation cover. (See Section A in S1 Text for more details of the JULES model setup and ancillary data). The resulting simulated grid box mean water table depth is combined with the grid box statistical distribution of topographic index [46] to produce a probability distribution function of water table within that grid box and, therefore determine the fraction of groundwater inundation (i.e. the fraction of the grid box where the water table is at/above the soil surface) [9,22].
The parameterisation of fluvial inundation is combined with that from groundwater by assuming that both are likely to preferentially occur in the same parts of a grid box (because high water tables also usually occur in regions with low topography). Hence within each grid box the maximum of the two is taken to give the total grid box inundation fraction.
2.3 Contemporary JULES simulations using observed meteorology
The JULES model setup (described in detail in Section A in S1 Text) includes the CO2 physiological effect whereby elevated atmospheric CO2 reduces stomatal opening and transpiration [47] and increases water use efficiency (WUE). JULES is first driven by the meteorological dataset WFDEI [48] (WATCH Forcing Data methodology applied to ERA-Interim data). WFDEI combines monthly mean observations with ERA-Interim reanalysis output data, producing a dataset with sub-daily temporal variability which is bias-corrected to observed monthly means. It contains sub-daily, near-surface air temperature, wind speed and humidity, and short‐wave and long‐wave radiation, and precipitation. WFDEI uses (GPCC Global Precipitation Climatology Centre) monthly mean precipitation gauge station data and applies a gauge correction. This approach has been compared against satellite products and gauge station data [48]. WFDEI is at 0.5 degree resolution and covers the time period 1979–2012. JULES is spun up by running it multiple times over the 1979–1988 time period until the soil moisture, temperature and river routing stores have reached a steady state. The resulting simulation is analysed for the period 1990 to 1999.
2.4 JULES simulations using earth system model meteorology
An ensemble of climate projections designed to explore a range of regional climate outcomes used in previous studies under the HELIX research programme (High-End cLimate Impacts and eXtremes [37]; www.helixclimate.eu, last access: 22 December 2023) is utilised here. These have been generated by driving the same global atmosphere model (HadGEM3-A) [49–51] with SSTs and sea ice cover produced by an earlier set of climate models (IPSL-CM5A-LR, GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-MR, MIROC-ESM-CHEM and ACCESS1-0). These were forced by the very high greenhouse gas concentrations scenario RCP8.5 which led to approximately 4°C global warming by the end of the 21st Century. The different SST patterns drive different regional precipitation responses, allowing the implications of uncertainties in future precipitation to be explored. The HadGEM3-A atmosphere-only simulations were at a higher spatial resolution than the original climate models, providing greater spatial detail. The output is then re-gridded onto a 0.5 degrees grid and used to drive JULES. The modelled meteorology produced by the six experiments are referred to as: HadGEM3-IPSL-MR, HadGEM3-GFDL, HadGEM3-HadGEM2-ES, HadGEM3-IPSL-MR, HadGEM3-MIROC and HadGEM3-ACCESS and the same nomenclature is used for the respective JULES simulations driven by those climates. The JULES simulations are run from 1979 to 2098 after repeating the first ten years until soil moisture, temperature and river flow have been spun-up.
2.5 Wetland classification
Reis et al. [15] apply a classification system to a downscaled version of GIEMS, called GIEMS-D15 [52] and use a clustering approach on a 1km resolution grid. The [15] wetland classification is adapted here to produce one that is suitable for a coarse (0.5°) resolution, statistical inundation model. For each grid box the wetland season is defined as the time when the wetland area is above a specific threshold of the maximum of the long term, contemporary (1990s), monthly mean inundation in that grid box. Here 75% is chosen as the threshold value as it enables clear differentiation between the different regions at the 0.5° resolution. First any inundation simulated by JULES that is not observed in either the GIEMS or SWAMPS-GLWD+fw inundation products is masked out before applying this classification. This is done by masking out grid boxes where both these products have less than 10% inundation in that 0.5° grid box. (Only GIEMS and SWAMPS-GLWD+fw are used for this detailed seasonal analysis because of the seasonal uncertainty of “permanent” inland waters in WAD2M+fw, and because Hess only covers two seasons).
2.6 Wetland classification—regional averages
The regional wetland season is defined as when the total regional inundation extent is at or above a specific threshold of the multi-annual mean monthly maximum in the 1990s. By applying this threshold to the total regional inundation area rather than each grid box, this enables the calculation of effective regional wetland season onsets and endings, without the complications of meaning calendar months over different grid boxes. (For example, the mean of December and February onsets should logically be January and not July. This also allows us to analyse changes when the wetland season is a whole calendar year in one time period but not the other). This is applied to the total wetland area in the region (i.e. there is no individual grid box 0.1 threshold masking).
3. Results
3.1 Contemporary JULES inundation using observed meteorology
The simulated JULES inundation produced from observation-based contemporary driving data WFDEI [48] (see Materials and Methods) is compared against different observational datasets for the “low water” (October to November), “high water” (May to July) and annual mean 1990s time periods (Fig 2). (The overall precipitation mean and variability over the 1990s and the different time periods of the EO inundation datasets are comparable to that over the full WFDEI time series—see Section D in S1 Text for more information. Hence the overall simulated 1990s inundation is typical of the contemporary period and comparison between the long term mean inundation datasets and JULES simulation is valid).
Left to Right: Observation-based datasets GIEMS (1993–2007), SWAMPS-GLWD+fw (2000–2012) and WAD2M+fw (2000–2017), Hess (May->July 1996, Oct->Nov 1995), and JULES model simulation under WFDEI (1990–1999). Rows 1, 2, 3: Long term means: Annual (except for Hess which is mean of May->July 1996 and Oct->Nov 1995), May-> July, Oct->Nov. The major wetland regions are outlined by rectangles in the top row (see Fig 1). See main text for observation-based dataset details. Values range between 0 and 1 which correspond to a grid box with zero and 100% inundation respectively. This Amazon basin outline was constructed by Authors based on the 0.5 degree resolution ascii data downloaded from the TRIP data website (http://hydro.iis.u-tokyo.ac.jp/~taikan/TRIPDATA/#Data/). This TRIP dataset is republished under a CC BY license, with permission from Taikan Oki.
A very strong seasonal signal can be seen in all the EO-based products, with a much greater inundated area in May to July than October to November. There are considerable differences between all the EO-based inundation products however. WAD2M+fw and Hess generally have more extensive spatial inundation coverage than GIEMS and SWAMPS-GLWD+fw over all regions and seasons, but lower inundation on the Amazon main-stem in the Manaus region.
JULES picks up the major contemporary inundated regions in the Amazon basin (regions shown in Fig 1), successfully simulating the fluvial floodplain of the Amazon main-stem river and the major inter-fluvial wetlands in the Iquitos, Llanos de Moxos and Roraima (western, southern and northern wetland) regions. For all the time periods shown the JULES inundated extent is in-line with the EO products in the Manaus central/eastern region. JULES produces a high inundation fraction over Iquitos which covers a slightly more extensive area than SWAMPS-GLWD+fw and Hess. The JULES flooded extent is generally within the product range for both the Llanos de Moxos and Roraima wetland regions, apart from for the most northern part of Roraima where JULES simulates significant (groundwater) inundation not in any of the products. This is likely to be related to the occurrence of soils specific to this region [53] which are not well described by the soil hydraulic property formulations used (see Section A in S1 Text).
Fig 3 shows the total seasonal inundation extent over the Amazon basin. The seasonality simulated by JULES is greater than in GIEMS, SWAMPS-GLWD+fw and WAD2M+fw but less than Hess. (Note that the assumption that the freshwater term added to WAD2M+fw here occurs for a full calendar year is likely to have impacted the inundated area seasonality). The total inundation simulated by JULES is within the observational estimates (Fig 3) and is also comparable to other hydrological models which simulate minimum and maximum inundated areas ranges of 0.7–1.9 and 4.5–6.0x105km2 respectively [21].
Observation-based datasets GIEMS (green), SWAMPS-GLWD (blue dashed), SWAMPS-GLWD+fw (blue solid), WAD2M (red dashed), WAD2M+fw (red solid) and Hess (orange), and JULES simulation under WFDEI (excluding rivers–dashed black and including rivers–solid black). Time periods for monthly averages as in Fig 2 caption. Dashed lines indicate where freshwater sources are not included in a dataset.
The new simple wetland classification scheme developed (see Materials and Methods) is applied to show how the hydro-period (the seasonal occurrence of flooding) varies throughout the Amazon basin (Fig 4). The central and eastern Amazon main-stem and western wetland Iquitos regions generally show relatively little seasonality with their season length for the majority of the year. The northern and southern regions have shorter wetland seasons and different timings due to rainfall seasonality differences primarily dependent on latitude and driven by the movement of the ITCZ. For the EO-based products the northern wetland season typically starts between May to June, and ends between July and August for GIEMS and later for SWAMPS-GLWD+fw. The most northern part of this region has the latest onset and earliest ending, and therefore shortest duration. The southern region season typically runs from January-March to March-May. After the masking process (see Materials and Methods) JULES generally successfully predicts the onset and length of these four major wetlands. However JULES simulates a slightly longer inundation length in the Llanos de Moxos and Roraima regions, due to the simulated wetland seasons both starting earlier and ending later.
The classification uses the 75% threshold level. Left to right: Month of start, end and wetland season length. Top to bottom rows: GIEMS and SWAMPS-GLWD+fw EO products (using their respective long term monthly means) and JULES simulation with WFDEI forcing 1990–1999. The major wetland regions are outlined by rectangles in the first panel (see Fig 1).
3.2 Contemporary JULES simulations using earth system model meteorology
3.2.1 Annual mean hydrology.
The HadGEM3-driven JULES simulations for the 1990s are compared with the JULES simulation forced with meteorology from the WFDEI re-analysis (Fig 5). There is considerable spread in precipitation between the HadGEM3 runs with a general pattern from wetter to drier simulations in the following order: HadGEM3-ACCESS, HadGEM3-HadGEM2-ES, HadGEM3-MIROC, HadGEM3-GFDL, HadGEM3-IPSL-LR and HadGEM3-IPSL-MR. Since the only difference between the setup of the HadGEM3 simulations is the prescribed SSTs (and sea ice), this highlights the important role of SST patterns on simulated rainfall patterns over the Amazon basin. This importance is seen in the impact of ENSO on observed basin rainfall variability [54].
Top row: WFDEI precipitation and WFDEI-forced JULES inundation. Other rows: Climate model simulated precipitation and JULES inundation when forced with simulated meteorology. Rows 2–7 in order from wettest to driest. (Specific simulation is stated in each panel). Inundation is shown for the Amazon basin only. The major wetland regions are outlined by rectangles in the top row (see Fig 1). This Amazon basin outline was constructed by Authors based on the 0.5 degree resolution ascii data downloaded from the TRIP data website (http://hydro.iis.u-tokyo.ac.jp/~taikan/TRIPDATA/#Data/). This TRIP dataset is republished under a CC BY license, with permission from Taikan Oki.
All of the HadGEM3 simulations produce too little contemporary precipitation in the North East of the Amazon basin, although the extent of this varies greatly (Fig 5). (Indeed HadGEM3 underestimates precipitation in the East of the basin and overestimates it in the West even when observed SSTs are prescribed [51]). This underestimate over the Roraima and Manaus regions contributes to all these JULES simulations showing less inundation than under WFDEI, with the largest modelled flooded area discrepancies occurring when JULES is driven by climate simulations with the greatest precipitation biases (HadGEM3-GFDL, HadGEM3-IPSL-LR and HadGEM3-IPSL-MR).
Over the South and West of the Amazon basin HadGEM3-ACCESS and HadGEM3-HadGEM2-ES tend to over-estimate rainfall, resulting in more inundation simulated by JULES than under the WFDEI climate over both the Llanos de Moxos and Iquitos regions (Fig 5). HadGEM3-MIROC and HadGEM3-GFDL simulate higher and comparable rainfall to that observed over the Llanos de Moxos and Iquitos regions respectively, resulting in similar differences in inundation when JULES is driven by these climate simulations and by the WFDEI reanalysis. The inundation simulated under HadGEM3-IPSL-LR and HadGEM3-IPSL-MR climates is lower than that under WFDEI over Iquitos but of comparable magnitude over Llanos de Moxos.
Overall the regional precipitation simulations which are closest to that observed are the wetter simulations in the North and Central/East regions and the drier simulations in the South and West. As such there is no one simulation that consistently produces the best overall precipitation. In spite of these differences between the HELIX simulations, all the land surface hydrological responses to precipitation variability [55] are within observation estimates at the Amazon basin scale (see Section E in S1 Text).
3.2.2 Seasonal hydrology.
The seasonality in the hydrological cycle is now analysed (Fig 6). The monthly rainfall peaks in the calendar are related to the seasonal movement of the Inter Tropical Convergence Zone (ITCZ), with the peak occurring latest in the calendar year in the north of the Amazon basin and earliest in the south. The timings of the wet and dry seasons are generally well simulated. Over the northern Roraima region the rainfall is generally successfully reproduced by the wetter models, although all models are too dry in the latter part of the dry season. In western Iquitos region the simulated rainfall seasonal cycle is generally too large. In the Central/East Manaus region the wetter models produce sufficient precipitation for a majority of year although none of the models simulate enough rainfall in the height of the rainy season. In the southern Llanos de Moxos region the drier models better reproduce the observed rainfall throughout the year.
Rows top to bottom: Precipitation, evaporation, runoff, inundation fraction (all 1990s), and inundation fraction (2089–2098). Precipitation projections from the climate models and the respective hydrological responses simulated by JULES simulations are all distinguished using climate model names as follows: HadGEM3-ACCESS (purple), HadGEM3-HadGEM2-ES (dark blue), HadGEM3-MIROC (cyan), HadGEM3-GFDL (green), HadGEM3-IPSL LR (orange), HadGEM3-IPSL MR (red). Ensemble means are shown with black dotted lines. Thick black lines refer to the 1990s WFDEI precipitation and JULES simulated hydrology driven by the WFDEI climate. Thin black and grey lines are GIEMS and SWAMPS-GLWD+fw inundation respectively. Regions are as defined in Fig 1.
For both runoff and inundation, overall seasonal differences between JULES simulations driven by WFDEI and the HadGEM3 ensemble generally follow those in the WFDEI and HadGEM3 precipitation. There are differences in evaporation seasonality simulated over Roraima and Manaus regions between the observational and modelled data forced simulations which do not follow those in precipitation. As such they are likely to be related to other forcing differences. There is also generally greater evaporation and lower runoff and inundation in the JULES simulations driven by HadGEM3 than by WFDEI, even when there is a dry precipitation bias. These discrepancies could be caused by the HadGEM3 climate simulations having higher surface radiation resulting in more energy available for evaporation, and therefore less runoff. This is consistent with HadGEM3 under-estimating outgoing top of the atmosphere shortwave radiation [56]. Less intense rainfall could also reduce surface runoff and therefore increase the water available for evaporation.
The simulations with the precipitation closer to that observed (Figs 5 and 6) generally produce wetland classes closer to JULES simulations forced with observation-based meteorology (Figs 4 and 7). In the simulations where the rainfall is too low in the north and west wetland regions the wetland season is shorter. In the western region this is primarily due to a later wetland season onset. In the driest contemporary simulations much of the northern wetlands are below the threshold for classification and are therefore masked out (see Materials and Methods).
First 3 columns: Wetland season start, end and length for the 1990s. Fourth column shows the difference in season length between 1990s and 2089–98.
3.3 Future JULES projections using earth system model meteorology
3.3.1 Annual mean hydrological change.
Projected local precipitation change rather than local temperature change is the main direct driver in the simulated hydrological shift over the Amazon basin (See Section F in S1 Text). The simulated, decadal mean changes in the major wetlands between 1990s and 2090s are shown in Figs 8, 9 (area means) and 10 (spatial patterns). Although there is a large spread in magnitude between the simulated changes, they all tend to show an east-west dipole, with a larger precipitation reduction in the East of the Amazon basin, and a smaller decrease or small increase in the West. The inter-model agreement in precipitation and inundation is greater in the future than in the contemporary period (Fig 8). This is because overall, regions the largest absolute reductions in precipitation tend to occur in the simulations with the highest contemporary precipitation. As already discussed these correspond to the best and worst long term mean simulated precipitation for the North and Central/East, and South and West regions respectively (Figs 5 and 8). (The annual and seasonal projected changes are compared to [57] and [32] in Section F in S1 Text and are consistent with other ESM projections).
Rows 1–4: Precipitation, evaporation, runoff, wetland area decadal means between 1990–1999 and 2090–2098. The first column shows areal means for the whole Amazon basin region. The remaining regions are defined by the boxes in Fig 1. Precipitation projections from the climate models and the respective hydrological responses simulated by JULES simulations are all distinguished using climate model names as in Fig 6. Experiments are also defined in the first panel. Open circles indicate decadal mean changes at the 5% significance level. Annual means are shown as thin lines. WFDEI precipitation and JULES response: Black “w”. Inundation area estimates from GIEMS, SWAMPS-GLWD+fw, WAD2M+fw and Hess: Grey cross, plus, “D” and star respectively.
As Fig 8 but for percentage decadal mean changes between 1990–1999 and 2090–2098.
Changes in simulated precipitation (mmday-1) and JULES inundation fraction when forced with climate model simulated meteorology. Rows top to bottom in order of wettest to driest. The major wetland regions are outlined by rectangles in the top row (see Fig 1). Inundation is shown for the Amazon basin only. This Amazon basin outline was constructed by Authors based on the 0.5 degree resolution ascii data downloaded from the TRIP data website (http://hydro.iis.u-tokyo.ac.jp/~taikan/TRIPDATA/#Data/). This TRIP dataset is republished under a CC BY license, with permission from Taikan Oki.
Of the wetland regions studied, the southern and western wetlands are projected to have the smallest projected drops in precipitation on average (Figs 8 and 10). In the western region two climate simulations project a statistically significant (at the 5% level) reduction and one a statistically significant increase. Over the South and West the drier (and more realistic) contemporary simulations tend to predict either little change or a small increase in precipitation, whereas the wetter and less realistic precipitation rates of these regions usually simulate statistically significant drying. Hence the drier simulations over the West of the Amazon basin are broadly in line with observed historical trends of wetting here [24].
The overall reduction in precipitation tends to mainly reduce runoff rather than evaporation, which is consistent with observations that show that contemporary evaporation in the wetter tropical regions is typically “energy” rather than water limited [58]. This is especially the case over the western wetland region where there is little projected change in evaporation in spite of the large changes in precipitation in some of the simulations. Over the North and Central/East regions, in the most realistic (wettest) simulations both the annual mean evaporation and runoff reduce over the projection, indicating some water limitation by the end of the 21st Century. (This is consistent with evaporation exceeding precipitation for a limited period of the year—Figs 6 and 11). In the simulations with the most realistic (driest) precipitation over the South there is little projected change in future precipitation and (therefore) runoff to indicate whether the region is predominantly water or energy limited. However in the realistic contemporary Llanos de Moxos climates there are typically 4–5 months where evaporation outstrips precipitation (Fig 6). This increases by the end of the simulation (Fig 11), indicating increasing water limitation. This seasonal water limitation is in line with observations where there are longer dry seasons over the southern part of the Amazon basin, which tends to be more water limited [58].
Mean differences between 1990s and 2089–2098 for each month. Top to bottom: Precipitation, evaporation, runoff, inundation fraction differences. Individual ensemble members are shown with coloured lines defined by the driving climate models as in Fig 6, and ensemble means are shown as black dotted lines. The first column shows areal means for the whole Amazon basin region. The remaining regions are defined by the boxes in Fig 1.
If, in the HadGEM3 atmosphere simulations, these precipitation reductions simulated at the end of the 21st century also resulted in mainly decreased local runoff rather than evaporation reduction, this would infer that the simulated drop in precipitation is driven by a reduction of non-local water convergence rather than local evaporation reduction. This is consistent with the analysis of [25], which proposed that recent trends in Amazon basin precipitation have been driven by atmospheric water vapour imported from the warming tropical Atlantic. As the wetter contemporary simulations have a higher proportion of precipitation from non-local moisture sources than the drier simulations, this is consistent with the former predicting overall greater drops in future rainfall.
The spatial changes in mean inundation broadly follow those of precipitation, with a general dipole split with decreases in the East and a mix of increases and decreases in the West (Figs 8–10), which is broadly consistent with [31]. Given the large upstream area of the Manaus region, the inundation extent here is heavily influenced not just by local rainfall and runoff but also by upstream hydrology impacting the river flow down the Amazon main-stem.
Following the simulated precipitation responses, the largest decreases in projected inundation generally occur in simulations with the highest present-day rainfall (Fig 10). Most simulations predict a reduction in inundation over the wetland regions studied here. Only the HadGEM3-IPSL-MR climate leads to JULES simulating increases over the northern and western wetland (Roraima and Iquitos) regions and over the Amazon basin as a whole (Figs 8 and 9). The HadGEM3-IPSL-LR climate results in a small increase in inundation over the western Iquitos region only. Over the western wetlands the overall projected change in inundation extent ranges from -29% to 27% over the different runs. The northern wetlands have the largest percentage fall in regional wetland area of up to approximately 53%. The southern and central/eastern wetlands have mean inundation reductions of 17 and 11% respectively.
The Amazon basin mean change in decadal mean precipitation, evaporation, runoff and inundation extent by 2100 are -11% (-29 to 0%), -6% (-11 to -3%), -14% (-47 to +8%) and -11% (-36 to +9%) respectively (Fig 9). As transpiration is also dependent on atmospheric CO2 concentration, with plant WUE increasing as elevated CO2 reduces stomatal opening, this tends to reduce transpiration and increase soil moisture, runoff and wetland area. As a consequence, without the inclusion of stomatal opening CO2 dependence JULES simulates a larger reduction in inundation for a given drop in precipitation.
3.3.2 Seasonal hydrological change.
The projections generally show the largest reduction in rainfall over the latter part of the year (typically September to November, Fig 11; Section F in S1 Text) which tends to correspond to the end of the dry/start of the rainy season (although the Roraima region rainy season onset is early in the calendar year) (Fig 6). There is little change in the driest two simulations over Iquitos and Llanos de Moxos over this time period however. There is less consensus between the simulations earlier in the calendar year, with the wettest two runs typically predicting the largest rainfall reduction, the intermediate simulations an increase and the driest models little change.
The inundation changes tend to follow those of local precipitation but with a short lag. This is due to a combination of delay in drainage through the soil and river flow travelling from upstream. In the northern and western wetlands the greatest loss of wetland occurs during/after the peak flooding season respectively, with these reductions being greatest in the wetter models. In the western region there is an increase in flooding early in the year which peaks early in the rainy season in all but the wettest two simulated climates. In the southern wetlands there is on average a larger drop during and directly after the low flood season, though there is less projected change in the drier, more realistic models, and a large drop in the high flood season in the wettest simulation driven by the HadGEM3-ACCESS climate. In the central/eastern region the largest inundation reductions occur after the period of minimum flooding.
The new wetlands classification scheme is applied (see Materials and Methods) to further analyse the significance of the projected inundation changes. The climate simulations with the highest annual contemporary precipitation and largest projected rainfall reductions by the end of the 21st Century (Figs 5, 8 and 10) result in the largest drops in simulated wetland season lengths (Fig 7). For the wettest contemporary simulation (HadGEM3-ACCESS), the season is reduced by between 2–8 months for most grid boxes. Overall the drier the simulation the smaller the projected reduction (or the larger the increase) in season length. The region where there is least qualitative agreement between simulations is again Iquitos where four of the models show an overall reduction in season length and the other two (the driest) an increase.
To gain more insight into the importance of the change in hydro-period this wetland classification approach is applied to each wetland region area as a whole (see Materials and Methods). By the 2090s the simulation-average, regional wetland seasons shorten by between 1 and 5 months (Fig 12 using the 75% threshold). Over the Manaus and Iquitos regions there are simulated drops in wetland season of up to 10 and 8 months at 75% the threshold, resulting in highly seasonal wetland seasons by the end of this Century. For these regions there is little change in season length at the lower 50% and 25% thresholds. In both the northern and southern regions there are more comparable drops in the season lengths for all the thresholds considered (75%, 50% and 25%).
Fractional number of simulations defined as a wetland month for each wetland region (see Fig 1). 1990s (blue, solid line) and 2089–2098 (orange, dashed line). Thresholds used to define a wetland month (top to bottom): a) 75%, b) 50% and c) 25%. Average wetland season lengths for the 1990s and 2089–98 and the projected changes between them are shown in text in each panel.
The reduction of the rainy season in the South is predominantly driven by its later onset—Fig 12A and 12B. This is in line with the observed historical lengthening of the southern part of the basin dry season due to its delayed ending [59,60]. The simulated change in the North appears to be more driven by the earlier ending of the wetland season. Over the western region, although the rainy season length is reduced considerably on average, in the drier models the rainy season length increases and starts earlier (not shown).
4. Discussion
A fluvial inundation model suitable for the global coarse resolution JULES land surface model has been developed and validated over the Amazon basin. The updated JULES model, where fluvial and inter-fluvial inundation are combined, has enabled us to simulate both the spatial and temporal spread of inundation within the uncertainty in the observational products. This allows us to assess the potential future hydrological changes over the main Amazon basin wetland regions within the framework of a biogeochemical model. The impact of important ecological feedbacks, including the CO2 physiological effect of reduced stomatal opening under elevated atmospheric CO2, are incorporated in the hydrological assessments here. Without this physically realistic representation of transpiration, future evaporation is likely to be over-estimated [61], resulting in a likely under-estimate of projected future inundation. Furthermore, implementing dynamical flooding within the JULES land surface model enables future investigation of the interactions between hydrology, ecosystems and climate within a fully coupled Earth System model (UKESM).
Using the multiple modelled climates from the HELIX project to drive JULES off-line, JULES predicts an overall reduction in Amazon basin wetlands over most regions and simulations. It simulates a similar average drop of over 11% for both basin-mean precipitation and inundation by 2100. This comparable percentage sensitivity between precipitation change and wetland area response is in agreement with previous studies [62,63]. JULES simulates future inundation increases in the West of the Amazon basin for the simulations with the drier and locally more realistic rainfall, but decreases for the simulations with an over-estimate of contemporary rainfall. Over the North and Central/Eastern wetland regions the more realistic precipitation simulations predict a drop of inundation. This is in partial agreement with other studies: [31] show a statistically significant mean increase in future inundated area in the West of the basin and potential decreases in the central region; [35] predict a predominant increase in the West of the basin, with less consensus over the East.
Developing a simple wetland classification scheme designed for coarser resolutions typical of ESMs provides a new tool to aid the understanding of the impact of projected hydrological changes from climate models. Significant drops in wetland season are predicted in most simulations and regions, with the exception of the Iquitos region. Over this region the wettest models simulate a significant drop in wetland season length, but a small increase of up to 2 months is predicted for the driest and more accurate contemporary simulations, with the latter being consistent with [35]. Consistent with [35], our findings also simulate a shortening of the inundation season in the eastern part of the bason. The projected shortening of the future wetland season and delay in the onset of the southern wetland season found in our study are consistent with the observed historical shortening of the South American monsoon season [64] and delayed ending of the southern Amazon bason dry season [59,60].
Utilising the multiple HELIX simulations here has enabled us to investigate uncertainty in future hydrological impacts associated with SST patterns. These are found to be significant, with the projected change in total Amazon basin inundated area ranging from a long-term mean reduction of 36% to an expansion of 9% by the 2090s. In the northern Roraima, southern Llanos de Moxos and central/eastern Manaus wetland regions there are both substantial decreases or smaller increases in simulated inundation extent with means (and ranges) of -17% (-53% to +12%), -17% (-42% to 3%) and -11% (-26 to 1%). For the western Iquitos region there is considerable qualitative uncertainty with different simulations showing large contractions and expansions in wetland area, with a mean change of -4% and range from -29% to 27%. These are long-term mean projected changes, but considerable inter-annual variability is also predicted, with the continued possibility of flooding.
This spread in projected inundation is primarily caused by those in precipitation. The magnitude of the simulated contemporary HELIX precipitation is found to be a good indicator of the future projection with the wetter contemporary basin climates predicting greater future drying, and resulting in considerable predicted long-term mean drying of up to 2.0mmday-1 over the Amazon basin as a whole and 2.1, 2.5, 1.5 and 2.2mmday-1 for the Roraima, Manaus, Llanos de Moxos and Iquitos respectively. There is no simulation which consistently simulates the best contemporary precipitation over the whole of the basin, with the wetter simulations being too wet in the southern and western regions and the drier models being too dry in the northern and central/eastern regions.
Baker et al. [65] show that CMIP5 models with realistic historical hydrological (energy-limited) behaviour generally simulate drying in the east and wetting in the west of the Amazon basin. This is partially consistent with our results in that the wetter (and more realistic) present day simulations for the East produce a greater projected precipitation reduction here. These same models are also energy limited in the East, but their contemporary western climates are too wet and they simulate little projected western drying. The models which have a drier, more realistic 1990s climate in the west simulate a projected increase in precipitation there.
Applying bias correction techniques here will not resolve this uncertainty however, as these assume that the same bias correction is applicable under climate change [66] because the projected magnitude and sign of the precipitation changes in the HELIX simulations are strongly dependent on the modelled contemporary climate. Bias corrections can also introduce inconsistencies both spatially and between variables [66]. As a consequence, applying such techniques are highly likely to introduce inconsistencies in the JULES driving data which are physically inconsistent spatially and not applicable over the long term under climate change. Rather a better approach may be to weight the results based on the accuracy of the contemporary simulations.
As well as the projected spread in inundation caused by that simulated in the local and regional climate, there is uncertainty relating to the inundation modelling. The inundation scheme developed here is designed to be able to run globally within an ESM, but as such is relatively coarse and relies on sub-grid statistical information and does not include some processes such as detailed floodplain flow. Regional scale validation of such models is also hampered by limitations in observations, such as in spatial and temporal inundation extent and detailed topographic information, especially over densely forested areas.
In spite of these uncertainties, such large overall projected reductions in rainfall and inundation would also have an impact on the availability of water resources, with dry-season stress likely to increase over the eastern part of the Amazon basin [67]. Carbon stocks would also be affected. It is unclear how wetland carbon stocks would respond to hydrological changes however, with evidence of inundation variability affecting tree growth, vegetation mortality and productivity differently [16]. Fire probability would likely increase in the southern part of the Amazon basin under future climate change [68].
Hydrological changes of this magnitude could have a significant impact on ecosystems which are specifically adapted to inundated conditions. Many organisms in the extremely biodiverse Amazon basin wetlands are critically dependent on food and habitat. Indeed these wetlands support large numbers of species of fish and flood-tolerant trees [16]. Given the relationship between flooding and fish abundance [16,18], the projected future reduction in flooding over much of the region would reduce the availability of fish stocks and diversity.
5. Conclusions
Our analysis predicts significant changes in future inundation extent and seasonality over the major wetlands of the Amazon basin under a very high greenhouse gas emissions scenario. A majority of the simulations predict considerable reductions in inundation extent and seasonality changes for the northern, southern and central/eastern basin wetlands. There is less consensus over the western basin wetlands however.
Such large uncertainties in Amazon basin inundation projections and therefore their ecological impact underline the importance of reducing the uncertainty in regional precipitation under climate change. Improvements in ESM simulations through better parameterisations and higher resolution models [32] will reduce these rainfall and hydrological impact uncertainties.
This fluvial inundation model developed here can be implemented globally and, incorporating it into global JULES will enable us to add riverine inundation into the fully coupled climate model UKESM in the future. This will enable the more complete modelling of interactions in both fluvial and inter-fluvial wetland hydrology, water resources, carbon stocks and methane emissions.
Supporting information
S1 Text. JULES overbank inundation parameterization: Description and validation.
A detailed description and validation of the JULES overbank inundation paramerisation.
https://doi.org/10.1371/journal.pwat.0000225.s001
(DOCX)
Acknowledgments
We thank John Caesar for performing the HadGEM3 simulations as part of the HELIX project funded by the European Commission 7th Framework Programme (Grant Agreement 603864). We thank all data providers especially Zhen Zhang, Ben Poulter, Catherine Prigent and Taikan Oki for providing advice and use of the WAD2M, SWAMPS-GLWD, GIEMS and TRIP datasets.
References
- 1. Fluet-Chouinard E, Stocker BD, Zhang Z, Malhotra A, Melton JR, Poulter B et al. Extensive global wetland loss over the past three centuries. Nature 2023 pmid:36755174
- 2.
Ramsar Convention on Wetlands. 2018. Global Wetland Outlook: State of the World’s Wetlands and their Services to People. Gland, Switzerland: Ramsar Convention Secretariat.
- 3. Salimi S, Suhad A A.A.N. Almuktar SAAAN, Scholz M. Impact of climate change on wetland ecosystems: A critical review of experimental wetlands. J. Environmental Management. 2021 pmid:33611067
- 4. Nahlik A and Fennessy M. Carbon storage in US wetlands. Nat Commun. 2016 pmid:27958272
- 5. Saunois M, Stavert AR, Poulter B, Bousquet P, Canadell JG, Jackson RB et al. The global methane budget 2000–2017. Earth System Science Data. 2020 15;12(3):1561–623.
- 6. Gedney N, Huntingford C, Comyn-Platt E, Wiltshire A. Significant feedbacks of wetland methane release on climate change and the causes of their uncertainty. Environmental Research Letters. 2019
- 7. Comyn-Platt E, Hayman G, Huntingford C, Chadburn SE, Burke EJ, Harper AB et al. Carbon budgets for 1.5 and 2°C targets lowered by natural wetland and permafrost feedbacks. Nat. Geo. 2018. 11, pages568–573.
- 8. Davidson NC, Fluet-Chouinard E, Finlayson CM. Global extent and distribution of wetlands: trends and issues. Marine and Freshwater Research 2018
- 9. Gedney N, Cox PM. The Sensitivity of Global Climate Model Simulations to the Representation of Soil Moisture Heterogeneity. J. Hydrometeorology. 2003
- 10. Papa F, Prigent C, Aires F, Jimenez C, Rossow W, Matthews E. Interannual variability of surface water extent at the global scale, 1993–2004. Journal of Geophysical Research, 2010
- 11. d’Orgeval T, Polcher J, de Rosnay P. (2008), Sensitivity of the West African hydrological cycle in ORCHIDEE to infiltration processes, Hydrol. Earth Syst. Sci., 2008 12, 1387–1401,
- 12. Dadson SJ, Ashpole I, Harris P, Davies HN, Clark DB, Blyth E, et al. Wetland inundation dynamics in a model of land surface climate: Evaluation in the Niger inland delta region. J. Geophys. Res. Atmos. 2010 115: 1–7.
- 13. Trigg MA, Wilson MD, Bates PD, Horritt MS, Alsdorf DE, Forsberg BR, et al. Amazon flood wave hydraulics, J. Hydrol. 2009. 374, 92–105.
- 14. Yamazaki D, Kanae S, Kim H and Oki T. A physically based description of floodplain inundation dynamics in a global river routing model. Water Res. Res. 2011. 47, W04501,
- 15. Reis V, Hermoso V, Hamilton SK, Bunn SE, Fluet‐Chouinard E, Venables B, et al. Characterizing seasonal dynamics of Amazonian wetlands for conservation and decision making. Aquatic Conserv: Mar Freshw Ecosyst. 2019; 29:1073–1082.
- 16. Melack JM, Coe MT. Amazon floodplain hydrology and implications for aquatic conservation. Aquatic Conserv.: Mar. Freshw. Ecosyst. 2021 31:1029–1040.
- 17. Junk WJ, Fernandez Piedade MT, Schongart J, Cohn-Haft M, Adeney JM, Wittman F. A Classification of Major Naturally-Occurring Amazonian Lowland Wetlands. Wetlands 2011
- 18. Tomasella J, Pinho PF, Borma LS, Marengo JA, Nobre CA, Bittencourt ORFO et al. The droughts of 1997 and 2005 in Amazonia: floodplain hydrology and its potential ecological and human impacts. Climatic Change 2013
- 19. Hess LL, Melack JM, Affonso AG, Barbosa C, Gastil-Buhl M, Novo EMLM. Wetlands of the Lowland Amazon Basin: Extent, Vegetative Cover, and Dual-season Inundated Area as Mapped with JERS-1 Synthetic Aperture Radar. Wetlands 2015
- 20. Hess LL, Melack JM, Affonso AG, Barbosa C, Gastil-Buhl M, Novo EMLM. 2015. LBA-ECO LC-07 Wetland Extent, Vegetation, and Inundation: Lowland Amazon Basin. ORNL DAAC, Oak Ridge, Tennessee, USA.
- 21. Fleischmann AS, Papa F, Fassoni-Andrade A, Melack JM, Wongchuig S, Paiva RCD et al. How much inundation occurs in the Amazon River basin? Remote Sensing of Environment 2022
- 22. Gauci V, Figueiredo V, Gedney N, Pangala SR, Stauffer T, Weedon GP et al. Non-flooded riparian Amazon trees are a regionally significant methane source. Phil. Trans. of Royal Soc. A. 2021 pmid:34865530
- 23. Pangala SR, Enrich-Prast A, Basso LS, Bittencourt Peixoto R, Bastviken D, Hornibrook ERC et al. Large emissions from floodplain trees close the Amazon methane budget. Nature 2017 552, 230–234. pmid:29211724
- 24. Heerspink BP, Kendal AD, Coe MT, Hyndman DW. Trends in streamflow, evapotranspiration, and groundwater storage across the Amazon Basin linked to changing precipitation and land cover. Journal of Hydrology: Regional Studies 2020 32 (100755).
- 25. Gloor M, Brienen RJW, Galbraith D, Feldpausch TR, Schöngart J, Guyot JL et al. Intensification of the Amazon hydrological cycle over the last two decades. Geophys. Res. Lett. 2013 40, pp1729–1733,
- 26. Coe MY Costa MH, Soares-Filho BS. The influence of historical and potential future deforestation on the stream flow of the Amazon River–Land surface processes and atmospheric feedbacks. Journal of Hydrology 2009 369 pp165–174.
- 27. da Rocha NS, Veettil BK, Grondona A and Rolim S. The influence of ENSO and PDO on tropical Andean glaciers and their impact on the hydrology of the Amazon Basin. Singapore J. Trop. Geog. 2019.
- 28. Farinosi F, Arias ME, Lee E, Longo M, Pereira FF, Livino A et al. Future Climate and Land Use Change Impacts on River Flows in the Tapajós Basin in the Brazilian Amazon. Earth’s Future. 2019
- 29. Alves LM, Chadwick R, Moise A, Brown J, Marengo JA. Assessment of rainfall variability and future change in Brazil across multiple timescales. Int J. Climat. 2020
- 30. Cook B, Zeng N, Yoon JH. Will Amazonia Dry Out? Magnitude and Causes of Change from IPCC Climate Model Projections. Earth Interactions. 2012
- 31. Sorribas MV, Paiva RCD, Melack JM, Bravo JM, Jones C, Carvalho L et al. Projections of climate change effects on discharge and inundation in the Amazon basin. Climatic Change 2016
- 32. Bambach NE, Rhoades AM, Hatchett BJ, Jones AD, Ullrich PA and Zarzycki CM. Projecting climate change in South America using variable-resolution Community Earth System Model: An application to Chile. Int. J. Climatology 2021
- 33. Marengo JA, Souza CA, Thonicke K, Burton C, Halladay K, Betts RA et al. Changes in Climate and Land Use Over the Amazon Region: Current and Future Variability and Trends. Front. Earth Sci. 2018
- 34. Parsons LA. Implications of CMIP6 Projected Drying Trends for 21st Century Amazonian Drought Risk. Earth’s Future. 2020
- 35. Langerwisch F, Rost S, Gerten D, Poulter B, Rammig A, Cramer W. Potential effects of climate change on inundation patterns in the Amazon Basin. Hydrol. Earth Syst. Sci. 2013
- 36.
Randall DA et al. Climate models and their evaluation, in: Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, edited by: Solomon S., Qin D., Manning M., Chen Z., Marquis M., Averyt K. B., Tignor M., and Miller H. L., Cambridge University Press, 2007.
- 37. Betts RA, Alfieri L, Bradshaw C, Caesar J, Feyen L, Friedlingstein P et al. Changes in climate extremes, fresh water availability and vulnerability to food insecurity projected at 1.5°C and 2°C global warming with a higher-resolution global climate model. Phil. Trans. R. Soc. A. 2018 pmid:29610383
- 38. Best MJ, Pryor M, Clark DB, Rooney GG, Essery RLH, Ménard CB, et al. The joint UK land environment simulator (JULES), model description–Part 1: energy and water fluxes. Geosci. Model Dev. 2011 4, 677–699.
- 39. Clark DB, Mercado LM, Sitch S, Jones CD, Gedney N, Best MJ et al. The joint UK land environment simulator (JULES), model description–Part 2: carbon fluxes and vegetation dynamics. Geoscientific Model Dev. 2011
- 40. Prigent C, Papa F, Aires F, Rossow W and Matthews E. Global inundation dynamics inferred from multiple satellite observations, 1993–2000. Journal of Geophysical Research, 2007
- 41. Poulter B, Bousquet P, Canadell JG, Ciais P, Peregon A, Saunois M et al. Global wetland contribution to 2000–2012 atmospheric methane growth rate dynamics Env. Res. Lett. 2017
- 42. Lehner B, Doll P. Development and validation of a global database of lakes, reservoirs and wetlands J. Hydrol. 2004 296 1–22.
- 43. Pekel J-F, Cottam A, Gorelick N, Belward AS. High-resolution mapping of global surface water and its long-term changes Nature 2016 540, 418–422, pmid:27926733
- 44. Leopold LB, Maddock TJ. The Hydraulic Geometry of Stream Channels and Some Physiographic Implications. Geol. Surv. Prof. Pap. 1953 252 57.
- 45. Yamazaki D, Ikeshima D, Sosa J, Bates PD, Allen GH, Pavelsky TM. MERIT Hydro: A High-Resolution Global Hydrography Map Based on Latest Topography Dataset. Water Resour. Res. 2019. 1–21.
- 46. Marthews TR, Dadson SJ, Lehner B, Abele S, Gedney N. High-resolution global topographic index values for use in large-scale hydrological modelling. Hydrol. Earth Syst. Sci. 2015 19:91–104.
- 47. Field CB, Jackson RB, Mooney HA. Stomatal responses to increased CO2: Implications from the plant to the global scale. Plant, Cell & Env. 1995
- 48. Weedon GP, Balsamo G, Bellouin N, Gomes S, Best MJ, Viterbo P. The WFDEI meteorological forcing data set: watch forcing data methodology applied to ERA-Interim reanalysis data. Wat. Resour. Res. 2014 50.
- 49. Hewitt HT, Copsey D, Culverwell ID, Harris CM, Hill RSR, Keen AB et al. Design and implementation of the infrastructure of HadGEM3: the next-generation Met Office climate modelling system. Geosci. Model Dev. 4, 223–253 2011.
- 50. Martin GM, Bellouin N, Collins WJ, Culverwell ID, Halloran PR, Hardiman SC et al. The HadGEM2 family of met office unified model climate configurations. Geosci. Model Dev. 4, 723–757 2011.
- 51. Walters DN, Best MJ, Bushell AC, Copsey D, Edwards JM, Falloon PD et al. The Met Office Unified Model Global Atmosphere 3.0/3.1 and JULES global land 3.0/3.1 configurations. Geosci. Model Dev. 4, 919–941 2011.
- 52. Fluet-Chouinard E, Lehner B, Rebelo LM, Papa F, Hamilton SK. Development of a global inundation map at high spatial resolution from topographic downscaling of coarse-scale remote sensing data, Remote Sensing of Environment. 2015 158: 348–361.
- 53. Quesada CA, Lloyd J, Anderson LO, Fyllas NM, Schwarz M and Czimczik CI. Soils of Amazonia with particular reference to the RAINFOR sites. Biogeosciences. 2011.
- 54. Jiménez-Muño JC, Mattar C, Barichivich J, Santamaría-Artigas A, Takahashi K, Malhi Y et al. Record-breaking warming and extreme drought in the Amazon rainforest during the course of El Niño 2015–2016. Scientific Reports 6. 2016.
- 55. Lehner F, Wood AW, Vano JA, Lawrence DM, Clark MP, Mankin JS. The potential to reduce uncertainty in regional runoff projections from climate models. Nat. Clim. Chang. 2019
- 56. Williams KD, Copsey D, Blockley EW, Bodas-Salcedo A, Calvert D, Comer R et al. The Met Office Global Coupled Model 3.0 and 3.1 (GC3.0 and GC3.1) Configurations. J Adv in Modeling Earth Systems. 2017
- 57. Almazroui M, Ashfaq M.M. Islam MN, Rashid IU, Kamil S, Abid et al. Assessment of CMIP6 Performance and Projected Temperature and Precipitation Changes Over South America. Earth Syst. Env. 2021
- 58. Costa MH, Biajoli MC, Sanches L, Malhado ACM, Hutyra LR, da Rocha HR et al. Atmospheric versus vegetation controls of Amazonian tropical rain forest evapotranspiration: Are the wet and seasonally dry rain forests any different? J. Geophys. Res. 2010
- 59. Xu H, Lian X, Slette IJ, Yang H, Zhang Y, Chen A et al. Rising ecosystem water demand exacerbates the lengthening of tropical dry seasons Nat. Com. 2022 pmid:35835788
- 60. Fu R, Yin L, Mynemi RB. Increased dry-season length over southern Amazonia in recent decades and its implication for future climate projection. PNAS. 2013 pmid:24145443
- 61. Caoa L, Balab G, Caldeiraa K, Nemanid R, Ban-Weiss G. Importance of carbon dioxide physiological forcing to future climate change. PNAS. 2010 pmid:20445083
- 62.
Melack JM, Coe MT. Climate change and the Floodplain Lakes of the Amazon Basin. In: Goldman CR, Kumagai M, Robarts RD (eds) Climatic change and global warming of inland waters: impacts and mitigation for ecosystems and societies. Wiley, Chichester. https://doi.org/10.1002/9781118470596.ch17 2013.
- 63. de Paiva RCD, Costa Buarque D, Collischonn W, Bonnet M-P, Frappart F, Calmant S et al. Large-scale hydrologic and hydrodynamic modeling of the Amazon River basin. Water Resources Res. 2013
- 64. Arias PA, Fu R, Vera C, Rojas M. A correlated shortening of the North and South American monsoon seasons in the past few decades. Clim. Dyn. 2015 45:3183–3203
- 65. Baker JCA, Garcia-Carreras L, Buermann W, Castilho de Souza D, Marsham JH, Kubota PY et al. Robust Amazon precipitation projections in climate models that capture realistic land–atmosphere interactions. E.R.L. 2021
- 66. Maraun D. Bias Correcting Climate Change Simulations—a Critical Review. Curr. Clim. Change Rep. 2016 2:211–220
- 67. Mahli Y, Roberts JT, Betts RA, Killeen TJ, Li W, Nobre CA. Climate Change, Deforestation, and the Fate of the Amazon. Science 2008 pmid:18048654
- 68. Fonseca MG, Alves LM, Aguiar APD, Arai E, Anderson LO, Rosan TM et al. Effects of climate and land-use change scenarios on fire probability during the 21st century in the Brazilian Amazon. Glob. Change. Biol. 2019 pmid:31304669