(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .
Observed changes in hydroclimate attributed to human forcing [1]
['Dimitris A. Herrera', 'Department Of Geography', 'Sustainability', 'University Of Tennessee-Knoxville', 'Knoxville', 'Tn', 'United States Of America', 'Instituto Geográfico Universitario', 'Universidad Autónoma De Santo Domingo', 'Santo Dominigo']
Date: 2024-02
Observational and modeling studies indicate significant changes in the global hydroclimate in the twentieth and early twenty-first centuries due to anthropogenic climate change. In this review, we analyze the recent literature on the observed changes in hydroclimate attributable to anthropogenic forcing, the physical and biological mechanisms underlying those changes, and the advantages and limitations of current detection and attribution methods. Changes in the magnitude and spatial patterns of precipitation minus evaporation (P–E) are consistent with increased water vapor content driven by higher temperatures. While thermodynamics explains most of the observed changes, the contribution of dynamics is not yet well constrained, especially at regional and local scales, due to limitations in observations and climate models. Anthropogenic climate change has also increased the severity and likelihood of contemporaneous droughts in southwestern North America, southwestern South America, the Mediterranean, and the Caribbean. An increased frequency of extreme precipitation events and shifts in phenology has also been attributed to anthropogenic climate change. While considerable uncertainties persist on the role of plant physiology in modulating hydroclimate and vice versa, emerging evidence indicates that increased canopy water demand and longer growing seasons negate the water-saving effects from increased water-use efficiency.
Funding: This work was partially funded by the Center of Global Engagement of the University of Tennessee–Knoxville through the Global Catalyst Faculty Research Grant to DH. This work was further supported by the National Science Foundation (AGS-1803995 to KA), (AGS-1751535 to TA), and (BCS-1759629 to KA). The efforts of Dr. Fasullo in this work were supported by NASA Awards 80NSSC21K1191, 80NSSC17K0565, and 80NSSC22K0046, and by the Regional and Global Model Analysis (RGMA) component of the Earth and Environmental System Modeling Program of the U.S. Department of Energy's Office of Biological & Environmental Research (BER) under Award Number DE-SC0022070. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
In this Review, we provide a survey of the literature on observed changes in hydroclimate as a result of anthropogenic emissions and land use changes during the historical period (e.g., 1850–2020). We analyze the mechanisms underlying human-forced influences on hydroclimate and the advantages and limitations of the approaches to estimating the anthropogenic contributions to change, including detection and attribution methods. Given the role of increased global temperature in exacerbating contemporaneous droughts [ 14 – 16 , 31 ], we focus on the effects of anthropogenic warming. Finally, we analyze studies on changes in extreme precipitation, wildfires, and the possible influence of plant physiology and phenology on observed hydroclimate changes and future drought risk.
Observational and modeling studies find that anthropogenic climate change has already increased drought risk [ 13 – 19 ] and extreme precipitation events [ 20 , 21 ]; reduced snowpack [ 22 ], ice sheets [ 23 ], and runoff [ 17 ]; and altered seasonal precipitation patterns [ 24 ] and phenology [ 25 ]. Many of these detection and attribution studies rely on climate models to identify and separate the anthropogenic signal from natural variability [ 12 , 26 ]. A major limitation is that considerable biases exist in models [ 27 ], especially in simulating certain hydroclimate features at regional and local scales [ 28 ], as well as in the observational datasets used for evaluating models [ 26 , 29 , 30 ]. Novel approaches for detecting hydroclimatic changes include machine learning and other statistical tools to identify outliers in precipitation, temperature, and soil moisture from instrumental and reconstructed records and climate models [ 26 ].
Climate models suggest significant changes in global hydroclimate due to the human-driven increases in greenhouse gas concentrations in the atmosphere through the twenty-first century [ 4 , 6 – 8 ]. Simulated changes in precipitation include drier conditions in the subtropics and wetter conditions in the tropics and extratropics [ 7 , 9 ]. However, this phenomenon occurs mostly over the oceans [ 10 ]. Regions that models robustly project will experience increased aridity arising from declines in precipitation, increased evapotranspiration, or both include the Mediterranean, southwestern South America, southern Australia, and southwestern North America [ 6 , 8 , 11 , 12 ].
Water is essential for supporting life on Earth [ 1 ]. Water moves in the Earth system through the hydrologic (water) cycle maintaining the Earth’s energy homeostasis [ 2 ]. Global estimates of annual total water exchanges between the atmosphere and surface (primarily precipitation and evapotranspiration) are on the order of 50 x 10 4 km 3 [ 2 ]. About 40 x 10 3 km 3 is transported from the ocean to land as precipitation and the same amount returns to the ocean as runoff [ 2 ]. This water exchange is critical for human well-being and development, as food security, electrical power generation, industry, and municipal water supply, amongst other socioenvironmental systems, depend on the availability and accessibility of water [ 3 , 4 ]. Changes in hydroclimate (e.g., climate-driven changes in the water cycle) are one of the most impactful consequences of climate change on human society, given the importance of precipitation (seasonality and magnitude) on food security and economic development [ 3 ]. The atmospheric water balance can be expressed as the difference between precipitation (P) and evapotranspiration (E). That, in turn, equals the sum of vertically-integrated atmospheric humidity (“precipitable water,” W) change (dW/dt) and horizontal moisture convergence (∇ ∙ Q ). Since dW/dt is negligible on monthly and longer time scales [ 5 ]: (1)
Land-use changes from human activities and wildfires further contribute to shifts in P–E [ 34 ] and hydroclimate [ 63 – 65 ]. Irrigation increases evaporation and water vapor, which contribute to higher precipitation rates locally or in nearby areas, as dictated by moisture transport [ 63 ]. Deforestation and urbanization can also modify P–E by changing surface albedo, energy, and water budgets [ 34 ], while aerosols from wildfires increase cloud albedo and cool surface temperatures, changing global P–E patterns [ 64 , 65 ]. More importantly, the direct impact of human water use on the water cycle is often misrepresented in many estimates and diagrams [ 66 ]. Current human freshwater use through agriculture, livestock, water withdrawals, and industry is 50% of global river discharge [ 66 ].
The projected shifts in P–E over land are associated with changes in other aspects of hydroclimate, including a decline in snowpack, runoff, and surface soil moisture [ 7 ]. Seasonal changes in runoff are expected, especially in snow-dependent regions [ 7 , 58 ]. In these regions, simulations indicate that snowpacks would not only melt earlier and faster [ 58 ] but also an increase in the rain/snow ratio [ 57 ] would further inhibit snow accumulation. Consequently, an increased runoff in winter and spring due to a higher rain/snow ratio and faster snow melting is expected, and the opposite in the summer and autumn [ 7 ]. Soil moisture declines are more robust than precipitation across models in many subtropical regions [ 7 , 57 ]. Significant declines in modeled soil moisture occur in regions even with small increases in precipitation [ 7 , 59 ], which highlights the role of increased evaporative demand of the atmosphere over land, land-surface feedbacks, and changes in plant water use [ 7 , 60 , 61 ]. While some studies indicate that vegetation response to increased atmospheric CO 2 concentration could ameliorate drought risk by improving the water use efficiency of plants [ 11 , 60 ], others, in contrast, suggest an amplification by increased plant growth and water use [ 61 , 62 ].
Regional and local scale hydroclimate changes in a warming climate might be enhanced by local feedback processes, including changes in moisture transport and mass convergence ( Fig 1 ) and other land-atmosphere coupling controls, such as soil moisture [ 51 , 52 ], land-use changes, and water management [ 34 ]. Modeling studies indicate that regional to local scale increases in precipitation are closer to Clausius-Clapeyron [ 9 ] and might be higher (>10% K -1 ) in certain regions, especially considering extreme events alone [ 53 ]. A higher convective available potential energy (CAPE) has been associated with increased extreme precipitation in some regions over land [ 54 ]. CAPE is expected to increase further in a warmer climate [ 55 ] due to changes in lapse rates and increased low-level humidity, but mainly over the ocean [ 56 ]. More intense precipitation could increase the risk of flooding locally and alter runoff and recharge patterns [ 7 , 57 ].
Changes in P–E patterns due to global and regional atmospheric warming patterns and dynamics are also expected in a warmer climate [ 45 ] ( Fig 1 ). However, the mechanisms underlying these changes are not well-constrained from observations compared to the thermodynamic controls [ 28 ]. A slowdown in the global atmospheric circulation, especially in the tropics [ 41 , 46 ] counteracts the thermodynamic intensification of the water cycle and, consequently, partially reduces the P–E gradients in the ocean [ 34 ]. A weakening of the tropical atmospheric overturning circulation (Hadley and Walker circulations) with warming is more robust across climate models, especially for the Walker Circulation [ 34 , 41 , 42 ]. This is coherent with a decrease in the East-West Pacific sea level pressure gradient in some observational studies [ 41 , 46 ]. Nevertheless, more recent observational and modeling studies suggest a strengthening in the Walker Circulation in the historical period [ 47 ]. A plausible reason for this discrepancy is that climate models do not correctly simulate the response of the Walker Circulation to warming [ 28 , 47 ]. For example, many state-of-the-art climate models have a cold bias in the tropical Pacific compared to observations [ 48 ]. The strengthening of the Walker Circulation might also be an initial response to the warming, followed by its subsequent weakening [ 47 ]. To date, what the response of the Walker Circulation to climate change might be is still unclear [ 4 ]. Observations have also shown a narrowing and strengthening of the intertropical convergence zone (ITCZ) and an insignificant change in its mean location [ 49 , 50 ]. This is consistent with climate model projections for the ITCZ in the twenty-first century, which suggest a further narrowing of the ITCZ [ 45 ]. The current ITCZ width varies from 250 to 1500 km [ 49 ] and has narrowed from 20% K -1 in the Atlantic Ocean to 29% K -1 in the Pacific [ 50 ].
Increased water vapor in the tropics is consistent with higher mass convergence and precipitation minus evaporation (P–E) rates in observations and simulations [ 34 ], along with higher divergence and lower P–E rates in the subtropics [ 9 , 34 ]. These changes in global P–E patterns are often referred to as the “wet-get-wetter, dry-get-drier” or “wet-events-wetter, dry-events-drier” mechanism [ 9 , 10 ], in allusion to increased precipitation in areas already humid (e.g., the tropics and high-latitudes), while the opposite occurs in arid and semi-arid regions (e.g., the subtropics) as the climate warms. However, the “wet-get-wetter, dry-get-drier” mechanism is only valid over the ocean and certain areas over land, where moisture is unlimited [ 10 ]. Over land, especially in arid and semi-arid regions where moisture is limited, this mechanism is inaccurate because, in many cases, evapotranspiration cannot exceed precipitation [ 10 ]. Another factor inhibiting precipitation over land in a warming climate is an increased land-ocean temperature gradient [ 10 ] since the ocean warms slower than the land surface. Though the increased gradient may support the development of a sea breeze [ 42 ] or monsoon [ 43 ] circulation, the cooler ocean temperatures can not provide the moisture supply needed to satisfy the moisture demand over land [ 36 ].
Because the rate at which precipitation and evaporation increase as a function of temperature is smaller than for water vapor, a higher atmospheric water vapor residence time is expected [ 9 , 39 ]. The estimated residence time of water vapor is ~8.5 days [ 40 ], which might increase by 3–6% K -1 due to the warming in the twenty-first century [ 39 ]. Higher water vapor concentration in the lower troposphere might lead to a slowdown in atmospheric circulation and a reduced mass exchange between the boundary layer and the free atmosphere [ 9 , 41 – 43 ]. The weakened atmospheric circulation balances with the increased water vapor in the tropics, which is necessary to compensate for the smaller increase in global mean precipitation as compared to water vapor [ 41 , 44 ].
This diagram depicts changes with an El Niño-like pattern. As the climate warms, the atmosphere’s water-holding capacity increases further (reddish colors). That causes higher mass convergence and lower mass divergence (dashed gray lines), increasing the contrast between areas dominated by a convergent flow (e.g., equatorial Pacific Ocean) and the opposite in areas dominated by a divergent flow (subtropics). That, in turn, may increase the intensity of hydroclimate extremes: first, higher mass convergence is associated with higher precipitable water (e.g., [ 13 ]), and second, lower relative humidity in a warmer climate could increase the risk of drought, especially flash droughts [ 112 ]. The implications of such changes are critical to understanding the effects and teleconnection patterns of climate modes of variability, such as El Niño-Southern Oscillation (ENSO) in a changing climate.
Large-scale changes in precipitation (intensity, duration, and frequency), soil moisture, runoff, and evapotranspiration are expected due to thermodynamic and dynamic drivers as the climate warms [ 7 , 8 ] ( Fig 1 ). Global thermodynamic changes are explained through the Clausius-Clapeyron relationship, which predicts an exponential increase in the water-holding capacity of the atmosphere and thus, a higher water vapor with temperature of ~7% K -1 warming, assuming a constant relative humidity [ 9 ]. Observations and simulations suggest an increase in water vapor of nearly 6% K -1 , slightly lower than predicted by Clausius-Clapeyron [ 32 ]. However, the distribution of the change is spatially complex, with increases of ∼4%–5% K -1 near the surface and 10%–15% K -1 at the Upper Troposphere and predominantly over the tropical Pacific Ocean [ 32 – 34 ]. In contrast, the observed increases in global mean precipitation and evaporation are roughly 1–3% K -1 [ 8 , 9 ]. For precipitation alone, climate models indicate increases of 2.1–3.1% K -1 in global mean precipitation (e.g., [ 35 , 36 ]) and 5.5% K -1 in the intensity of extreme precipitation (e.g., a high percentile of daily precipitation) [ 36 ]. That means that the observed and expected increases in global mean precipitation and evaporation, especially over land, falls below that predicted by Clausius-Clapeyron [ 9 , 10 ]. An explanation for this discrepancy is that, at global scales, precipitation and evaporation are constrained by the atmospheric energy balance and dynamics [ 37 , 38 ].
3. Observed changes in hydroclimate
Separating and estimating the anthropogenic contributions to observed hydroclimate changes from other external forcings and natural variability can be challenging [67]. Detection and attribution methods attempt to do so by combining statistical tools with instrumental and simulated climate data to detect changes in the climate system and quantify the causes of these observed changes [68]. Detection and attribution are also necessary to evaluate the skill of climate models in simulating crucial processes occurring in the climate system, both globally and regionally, which is inferred from their ability to replicate observations [68].
The traditional detection and attribution approach involves using climate models to assess the changes in the probability of occurrence of an event and if those changes are driven by anthropogenic-forced change [68]. This approach is well-suited to quantify the contribution of anthropogenic warming to hydroclimate extremes controlled by thermodynamics [67], including drought intensification and duration due to temperature-driven increase in evapotranspiration [13–16, 69]. However, it is limited to quantifying the anthropogenic contribution to extreme hydroclimate events further enhanced by global and local atmospheric dynamics [67]. Newer approaches like the optimal fingerprint method [12, 70] have been used recently in detection and attribution studies of human-forced hydroclimate change, including to quantify the contribution of climate change to drought risk [12], the occurrence of flash droughts [69], and changes in the mean and extreme precipitation [71]. Despite the advances in detection and attribution methods, limitations persist for several reasons [67]. Some of those limitations arise from systematic biases in climate models, deficiencies in observational data (including the quality, length, and spatial coverage of instrumental records), and shortcomings in the detection and attribution approaches themselves [67]. Nevertheless, a growing body of evidence suggests that anthropogenic climate change is already impacting global hydroclimate and is likely to do so as the climate warms further in the twenty-first century [11, 12, 57, 67, 71].
3.1 Observed changes in drought Observations and simulations indicate an increase in the frequency, duration, and severity of drought in southwestern North America [15], the Mediterranean and southern Europe [72], Australia [73, 74], southwestern and central South America [18, 75, 76], Central America and the Caribbean [16, 17, 77], Africa [78, 79], and eastern Asia [18, 80] (Fig 2). These indications are drawn from a range of drought metrics such as the Palmer Drought Severity Index (PDSI) [12, 18, 70] (Fig 2A), the standardized precipitation index (SPI), standardized precipitation minus evaporation index (SPEI) [19, 81], and soil moisture [15]. PPT PowerPoint slide
PNG larger image
TIFF original image Download: Fig 2. Trends in observed precipitation and Palmer Drought Severity Index (PDSI). (A) trends from version 2020 of the Global Precipitation Climatology Centre (GPCCv2020), (B) trends from the “self-calibrated” PDSI dataset from Dai [18], (C) trends from the Climatic Research Unit version TSv 4.01 (CRU TSv.4.01), and (D) “self-calibrated” PDSI dataset from van der Schrier et al. [147]. The trends were calculated over a common period from 1950–2018. While GPCC and CRU precipitation datasets indicate similar patterns in their trends, the PDSI datasets differ in their trends, but most importantly in their magnitudes. In addition to using different input climate data, those PDSI data sets use different calibration periods. For example, the CRU PDSI product uses the whole period (i.e., 1901–2021), and Dai PDSI uses 1950–2000. This Figure was made with Natural Earth. Free vector and raster map data
https://www.naturalearthdata.com.
https://doi.org/10.1371/journal.pclm.0000303.g002 In southwestern North America, the anthropogenic fingerprint on hydroclimate has been shown to be robust, as indicated by an array of detection and attribution studies using model simulations, proxy and observational records to quantify the contribution of human-forced change [11–15, 26, 31, 70]. A notable event in this region is the megadrought from 2000–2018 [15], ranked as one of the most severe droughts since at least 800 CE [15]. While natural variability has also influenced this multidecadal event, anthropogenic climate change accounted for 46% [15], largely by increasing atmospheric vapor pressure deficit [14, 15, 31]. Similarly, earlier studies indicate that anthropogenic warming increased the severity of the 2011–2016 California drought by 8–27% [14], making it the most severe drought since 1200 CE [31]. As in other parts of Western North America, precipitation deficits associated with the 2011–2016 California drought have been linked with natural variability but worsened by higher temperatures [14, 15, 31]. Decreased streamflow of the Colorado River in 2000–2014 is consistent with the droughts observed in southwestern North America in the early twenty-first century [82]. Reduced streamflow from 2.7 to 9% has been estimated as a result of a 0.9°C increase in the mean temperature of the Upper Colorado River Basin [82]. Observed hydroclimate changes in southern Europe and the Mediterranean are also associated with increased aridity since at least 1960 [72, 84], which is consistent with climate model projections as a result of the warmth [72, 83]. Across the region, a trend toward aridity is observed in precipitation and soil moisture [72], including droughts like the severe Syrian drought in 2007–2010 [84]. It is estimated that anthropogenic climate change doubled or tripled the likelihood of the 2007–2010 Syrian drought [84]. Although previous studies have found an anthropogenic signal in precipitation reductions and warming temperatures [12], this is heterogeneous, and some parts of the Mediterranean are more strongly influenced by natural variability than others [81, 85]. For example, tree ring-reconstructed soil moisture indicates that droughts of similar duration and severity as the 2007–2010 Syrian drought have occurred across Mediterranean regions in the last 900 years [85]. However, the 1998–2012 Levant drought is likely the most severe since 1100 CE [85]. A significant increase in the duration, frequency, and severity of seasonal drought has been observed in portions of southwestern and southeastern Australia since the twentieth century, particularly in the autumn and winter [73, 74]. This change has been attributed to anthropogenic forcing, arising from greenhouse gas emissions and changes in atmospheric ozone levels [74]. The duration of seasonal droughts in parts of southeastern Australia has also increased between 10 and 69% since the mid-twentieth century [73]. The worst drought recorded in southeastern Australia, the 2001–2009 “Millenium Drought”, severely affected the agricultural sector of the country [86]. Earlier studies argue that the cause of the Millennium Drought was linked to fewer La Niña events and a negative Indian Ocean Dipole [87], suggesting a strong contribution from natural variability [87, 88]. Also, while higher temperatures are associated with drought in Australia, it remains unclear whether warmer temperatures contributed to or resulted from the drought [88]. Reconstructed soil moisture data, for example, indicate that droughts like the Millennium Drought in eastern Australia are in the range of natural variability for at least the last 500 years [88, 89]. The anthropogenic contribution to the recent 2019–2020 drought and associated bushfires is complex, as the signal is present in some drivers (e.g., temperature) [90]. Trends toward increased aridity are observed in the central (Brazil) and southwestern (Chile and Argentina) areas of South America [18]. In central Chile, the drying trend since late 1970s [68, 69] is characterized by multiple droughts, including an ongoing multidecadal megadrought that began in 2010 [76, 91]. The megadrought has reduced snowpack in the Andes, streamflow of major rivers, vegetation productivity [75], and lake levels [92]. While previous studies have found an anthropogenic contribution of ~25% to the Chilean megadrought, natural variability has also contributed [76, 93]. Furthermore, significant uncertainties persist in modeled precipitation in South America [94], precluding more accurate estimates of the anthropogenic contribution to the recent drought. However, reconstructed soil moisture indicates that the ongoing megadrought is the most severe of the last millennium and that its occurrence is unexpected from natural variability alone [95]. In this context, the drying trend in central Chile and western Argentina is associated with changes in the Southern Annular Mode and the expansion of the Hadley cell, both of which are expected from anthropogenic climate change [76, 93, 95]. Land-use feedback in the Amazon is also critical in recent aridity in portions of South America [96]. Deforestation has contributed to 4% of recent drying in parts of the Amazon, while longer dry seasons are associated with higher deforestation rates [96]. In the Caribbean Islands and Central America, a drying trend observed since at least 1950 includes several short-term (up to a year-long) and multiyear droughts [16, 17, 77, 97, 98]. Although previous studies found a modest but statistically significant decline in precipitation in some parts of these regions [97], more recent studies indicate that rainfall has not markedly changed and has indeed slightly increased in certain areas [18, 99, 100]. Further, there is a strong influence of natural variability in precipitation in the Caribbean and Central America, modulated by the El Niño–Southern Oscillation, the North Atlantic Oscillation, and the Atlantic Meridional Mode [77, 99, 101, 102]. Some of the worst droughts in these regions, for example, occurred during El Niño events [98, 103]. Paleoclimate and modeling studies have further suggested an El Niño-like pattern associated with drought in Central America [104]. However, drought conditions in Caribbean Islands and Central America are likely exacerbated by an increased atmospheric evaporative demand [16], driven by higher temperatures [99, 100, 105]. Droughts have also intensified in portions of Africa, including the Sahel, northern, and southwestern South Africa in recent decades [11, 18, 57, 78, 79, 106–108]. Some of the worst African droughts resulted from a failure of the rainy season, associated with the Indo–Pacific internal variability [78], resulting in droughts in the Sahel [106], eastern [107, 108], and southern Africa [79]. Although some studies have attributed the drying over southwestern South Africa to anthropogenic climate change [79], others suggest that the recent droughts are possible due to natural variability [100]. For example, Pascale et al. [79] estimated anthropogenic climate change increased the 2015–2017 “Day Zero Drought” five to six-fold. The rationale of the human-driven hydroclimate change in Africa is an increased sea surface temperature over the Indian Ocean and the opposite in the North Atlantic, as a result of a higher atmospheric greenhouse gas concentration and aerosols, respectively, and the southward shift of the ITCZ [11, 57, 109]. Further, paleoclimate records indicate that the drying experienced in eastern Africa is uncommon from natural variability alone [110]. Given the limitations and contradictory results regarding the anthropogenic fingerprint in African drought, it is still difficult to quantify the contribution. Other regions that have experienced a drying trend are southeast and eastern Asia, as observed in precipitation, streamflow of major river basins, and PDSI between 1950 and 2018 [18, 80]. Studies suggest that the drying in eastern Asia is due to increased temperatures and a significant decline in humidity and precipitation [18, 80]. From a long-term perspective, paleoclimate records indicate that the region has seen multidecadal or megadroughts in the Common Era [80, 91, 111], suggesting that major droughts are possible as a result of natural variability. An increasingly frequent type of drought is "flash drought" (i.e., a drought that unfolds relatively fast [112]). Over 74% of the land surface has seen an increased frequency of flash droughts attributed to anthropogenic warmth [69]. However, Qing et al. [113] found that flash droughts did not increase in frequency in 2000–2020, but their onset speed was faster. The anthropogenic contribution to the increased flash/slow drought ratio is 48%, while the warming has increased the onset of flash droughts by 39% [69]. A flash drought results from a marked precipitation deficit and strong evapotranspiration associated with low humidities, high temperatures, reduced cloud cover, and strong winds [112]. In this context, an increased evaporative demand of the atmosphere driven by anthropogenic climate change might increase flash drought risk.
[END]
---
[1] Url:
https://journals.plos.org/climate/article?id=10.1371/journal.pclm.0000303
Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.
via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/