Hydrological concept formation inside long short-term memory (LSTM) networks

Hydrology and earth system sciences, Journal Year: 2022, Volume and Issue: 26(12), P. 3079 - 3101

Published: June 20, 2022

Abstract. Neural networks have been shown to be extremely effective rainfall-runoff models, where the river discharge is predicted from meteorological inputs. However, question remains: what these models learned? Is it possible extract information about learned relationships that map inputs outputs, and do mappings represent known hydrological concepts? Small-scale experiments demonstrated internal states of long short-term memory (LSTMs), a particular neural network architecture predisposed modelling, can interpreted. By extracting tensors which translation (precipitation, temperature, potential evapotranspiration) outputs (discharge), this research seeks understand LSTM captures system. We assess hypothesis replicates real-world processes we LSTM. examine cell-state vector, represents LSTM, explore ways in learns reproduce stores water, such as soil moisture snow cover. use simple regression approach state vector our target (soil snow). Good correlations (R2>0.8) between probe variables interest provide evidence contains reflects comparable with concept variable-capacity stores. The implications study are threefold: (1) LSTMs processes. (2) While conceptual theoretical assumptions embedded model priori, derives data. These representations interpretable by scientists. (3) used gain an estimate intermediate water moisture. machine learning interpretability still nascent field technique for exploring has learned, results robust different initial conditions variety benchmarking experiments. therefore argue deep approaches advance scientific goals well predictive goals.

Language: Английский

---

Unprecedented fire activity above the Arctic Circle linked to rising temperatures

Science, Journal Year: 2022, Volume and Issue: 378(6619), P. 532 - 537

Published: Nov. 3, 2022

Arctic fires can release large amounts of carbon from permafrost peatlands. Satellite observations reveal that burned ~4.7 million hectares in 2019 and 2020, accounting for 44% the total area Siberian entire 1982-2020 period. The summer 2020 was warmest four decades, with burning an unprecedentedly carbon-rich soils. We show factors fire associated temperature have increased recent decades identified a near-exponential relationship between these annual area. Large are likely to recur climatic warming before mid-century, because trend is reaching threshold which small increases exponential burned.

Language: Английский

---

Shifts in vegetation activity of terrestrial ecosystems attributable to climate trends

Nature Geoscience, Journal Year: 2023, Volume and Issue: 16(2), P. 147 - 153

Published: Feb. 1, 2023

Abstract Climate change is expected to impact the functioning of entire Earth system. However, detecting changes in ecosystem dynamics and attributing such anthropogenic climate has proved difficult. Here we analyse vegetation 100 sites representative diversity terrestrial types using remote-sensing data spanning past 40 years a dynamic model plant growth, forced by reanalysis data. We detect activity for all find these can be attributed trends climate-system parameters. Ecosystems dry warm locations responded primarily soil moisture, whereas ecosystems cooler temperature. that effects CO 2 fertilization on are limited, potentially due masking other environmental drivers. Observed trend switching widespread dominated shifts from greening browning, suggesting many studied accumulating less carbon. Our study reveals clear fingerprint exhibited over recent decades.

Language: Английский

---

CO ₂ fertilization of terrestrial photosynthesis inferred from site to global scales

Proceedings of the National Academy of Sciences, Journal Year: 2022, Volume and Issue: 119(10)

Published: March 1, 2022

SignificanceThe magnitude of the CO2 fertilization effect on terrestrial photosynthesis is uncertain because it not directly observed and subject to confounding effects climatic variability. We apply three well-established eco-evolutionary optimality theories gas exchange photosynthesis, constraining main processes using measurable variables. Using this framework, we provide robust observationally inferred evidence that a strong detectable in globally distributed eddy covariance networks. Applying our method upscale globally, find comparable its situ counterpart but highlight potential for substantial underestimation tropical forests many reflectance-based satellite products.

Language: Английский

---

Excess mortality attributed to heat and cold: a health impact assessment study in 854 cities in Europe

The Lancet Planetary Health, Journal Year: 2023, Volume and Issue: 7(4), P. e271 - e281

Published: March 16, 2023

Language: Английский

---

Caravan - A global community dataset for large-sample hydrology

Scientific Data, Journal Year: 2023, Volume and Issue: 10(1)

Published: Jan. 31, 2023

Abstract High-quality datasets are essential to support hydrological science and modeling. Several CAMELS (Catchment Attributes Meteorology for Large-sample Studies) exist specific countries or regions, however these lack standardization, which makes global studies difficult. This paper introduces a dataset called Caravan (a series of CAMELS) that standardizes aggregates seven existing large-sample hydrology datasets. includes meteorological forcing data, streamflow static catchment attributes (e.g., geophysical, sociological, climatological) 6830 catchments. Most importantly, is both open-source software allows members the community extend new locations by extracting data in cloud. Our vision democratize creation use globally-standardized truly resource.

Language: Английский

---

Enhanced habitat loss of the Himalayan endemic flora driven by warming-forced upslope tree expansion

Nature Ecology & Evolution, Journal Year: 2022, Volume and Issue: 6(7), P. 890 - 899

Published: June 2, 2022

Language: Английский

---

City-level impact of extreme temperatures and mortality in Latin America

Nature Medicine, Journal Year: 2022, Volume and Issue: 28(8), P. 1700 - 1705

Published: June 27, 2022

Climate change and urbanization are rapidly increasing human exposure to extreme ambient temperatures, yet few studies have examined temperature mortality in Latin America. We conducted a nonlinear, distributed-lag, longitudinal analysis of daily temperatures among 326 American cities between 2002 2015. observed 15,431,532 deaths ≈2.9 billion person-years risk. The excess death fraction total was 0.67% (95% confidence interval (CI) 0.58-0.74%) for heat-related 5.09% CI 4.64-5.47%) cold-related deaths. relative risk 1.057 1.046-1.067%) per 1 °C higher during heat 1.034 1.028-1.040%) lower cold. In cities, substantial proportion is attributable nonoptimal temperatures. Marginal increases hot associated with steep These risks were strongest older adults cardiovascular respiratory

Language: Английский

---

LamaH-CE: LArge-SaMple DAta for Hydrology and Environmental Sciences for Central Europe

Earth system science data, Journal Year: 2021, Volume and Issue: 13(9), P. 4529 - 4565

Published: Sept. 16, 2021

Abstract. Very large and comprehensive datasets are increasingly used in the field of hydrology. Large-sample studies provide insights into hydrological cycle that might not be available with small-scale studies. LamaH-CE (LArge-SaMple DAta for Hydrology Environmental Sciences Central Europe, LamaH short; geographical extension “-CE” is omitted text dataset) a new dataset large-sample comparative hydrology Europe. It covers entire upper Danube to state border Austria–Slovakia, as well all other Austrian catchments including their foreign upstream areas. an area about 170 000 km2 nine countries, ranging from lowland regions characterized by continental climate high alpine zones dominated snow ice. Consequently, wide diversity properties present individual catchments. We represent this variability 859 gauged over 60 catchment attributes, covering topography, climatology, hydrology, land cover, vegetation, soil geological properties. further contains collection runoff time series meteorological series. These provided daily hourly resolution. All majority cover span 35 years, which enables long-term analyses temporal The classified 20 attributes information human impacts indicators data quality completeness. structure based on well-known CAMELS (Catchment Attributes MEteorology Studies) datasets. In contrast, however, does only consider independent basins, full area. Intermediate covered well, allows together novel considering network river topology applications. describe basic methodology preparation but also focus possible limitations uncertainties. additionally results conceptual baseline model checking plausibility inputs benchmarking. Potential applications outlined since it intended serve uniform basis research. at https://doi.org/10.5281/zenodo.4525244 (Klingler et al., 2021).

Language: Английский

---

A 1 km daily soil moisture dataset over China using in situ measurement and machine learning

Earth system science data, Journal Year: 2022, Volume and Issue: 14(12), P. 5267 - 5286

Published: Nov. 30, 2022

Abstract. High-quality gridded soil moisture products are essential for many Earth system science applications, while the recent reanalysis and remote sensing data often available at coarse resolution only surface soil. Here, we present a 1 km long-term dataset of derived through machine learning trained by in situ measurements 1789 stations over China, named SMCI1.0 (Soil Moisture China data, version 1.0). Random forest is used as robust approach to predict using ERA5-Land time series, leaf area index, land cover type, topography properties predictors. provides 10-layer with 10 cm intervals up 100 deep daily period 2000–2020. Using benchmark, two independent experiments were conducted evaluate estimation accuracy SMCI1.0: year-to-year (ubRMSE ranges from 0.041 0.052 R 0.883 0.919) station-to-station 0.045 0.051 0.866 0.893). generally has advantages other products, including ERA5-Land, SMAP-L4, SoMo.ml. However, high errors located North Monsoon Region. Overall, highly accurate estimations both ensure applicability study spatial–temporal patterns. As based on it can be useful complement existing model-based satellite-based datasets various hydrological, meteorological, ecological analyses models. The DOI link http://dx.doi.org/10.11888/Terre.tpdc.272415 (Shangguan et al., 2022).

Language: Английский

---