GLEAM4: global land evaporation and soil moisture dataset at 0.1 resolution from 1980 to near present

Scientific Data, Journal Year: 2025, Volume and Issue: 12(1)

Published: March 10, 2025

Terrestrial evaporation plays a crucial role in modulating climate and water resources. Here, we present continuous, daily dataset covering 1980–2023 with 0.1°spatial resolution, produced using the fourth generation of Global Land Evaporation Amsterdam Model (GLEAM). GLEAM4 embraces developments hybrid modelling, learning evaporative stress from eddy-covariance sapflow data. It features improved representation key factors such as interception, atmospheric demand, soil moisture, plant access to groundwater. Estimates are inter-compared existing global products validated against situ measurements, including data 473 sites, showing median correlation 0.73, root-mean-square error 0.95 mm d−1, Kling–Gupta efficiency 0.49. land is estimated at 68.5 × 103 km3 yr−1, 62% attributed transpiration. Beyond actual its components (transpiration, interception loss, evaporation, etc.), also provides potential sensible heat flux, stress, facilitating wide range hydrological, climatic, ecological studies.

Language: Английский

---

A long-term reconstruction of a global photosynthesis proxy over 1982–2023

Scientific Data, Journal Year: 2025, Volume and Issue: 12(1)

Published: March 3, 2025

Abstract Satellite-observed solar-induced chlorophyll fluorescence (SIF) is a powerful proxy for the photosynthetic characteristics of terrestrial ecosystems. Direct SIF observations are primarily limited to recent decade, impeding their application in detecting long-term dynamics ecosystem function. In this study, we leverage two surface reflectance bands available both from Advanced Very High-Resolution Radiometer (AVHRR, 1982–2023) and MODerate-resolution Imaging Spectroradiometer (MODIS, 2001–2023). Importantly, calibrate orbit-correct AVHRR against MODIS counterparts during overlapping period. Using bias-corrected data MODIS, neural network trained produce Long-term Continuous SIF-informed Photosynthesis Proxy (LCSPP) by emulating Orbiting Carbon Observatory-2 SIF, mapping it globally over 1982–2023 Compared with previous photosynthesis proxies, LCSPP has similar skill but can be advantageously extended Further comparison three widely used vegetation indices (NDVI, kNDVI, NIRv) shows higher or comparable correlation satellite site-level GPP estimates across types, ensuring greater capacity representing activity.

Language: Английский

---

Upscaling Land Surface Fluxes Through Hyper Resolution Remote Sensing in Space, Time, and the Spectrum

Journal of Geophysical Research Biogeosciences, Journal Year: 2024, Volume and Issue: 129(10)

Published: Oct. 1, 2024

Abstract Numerous efforts to measure land surface fluxes, from leaf canopy scales, have significantly advanced the field of biogeoscience. However, upscaling these estimates larger spatial and temporal scales remains a challenge. Recent advancements in remote sensing provide new opportunities bridge gaps efforts. In this review, I propose that emerging satellite data can support robust fluxes terms space through constellations low Earth orbit satellites, time geostationary spectrum via optical, thermal, microwave satellites. Lastly, recommend development long‐term network integrating tower‐based hyperspectral, instruments rigorously evaluate process fluxes.

Language: Английский

---

Atmospheric Transport Modeling of CO₂ With Neural Networks

Journal of Advances in Modeling Earth Systems, Journal Year: 2025, Volume and Issue: 17(2)

Published: Feb. 1, 2025

Abstract Accurately describing the distribution of in atmosphere with atmospheric tracer transport models is essential for greenhouse gas monitoring and verification support systems to aid implementation international climate agreements. Large deep neural networks are poised revolutionize weather prediction, which requires 3D modeling atmosphere. While similar this regard, subject new challenges. Both, stable predictions longer time horizons mass conservation throughout need be achieved, while IO plays a larger role compared computational costs. In study we explore four different (UNet, GraphCast, Spherical Fourier Neural Operator SwinTransformer) have proven as state‐of‐the‐art prediction assess their usefulness modeling. For this, assemble CarbonBench data set, systematic benchmark tailored machine learning emulators Eulerian transport. Through architectural adjustments, decouple performance our from shift caused by steady rise . More specifically, center input fields zero mean then use an explicit flux scheme fixer assure balance. This design enables conserving over 6 months all network architectures. study, SwinTransformer displays particularly strong emulation skill: 90‐day physically plausible multi‐year forward runs. work paves way toward high resolution inverse inert trace gases networks.

Language: Английский

---

Using Geostationary Satellite Observations and Machine Learning Models to Estimate Ecosystem Carbon Uptake and Respiration at Half Hourly Time Steps at Eddy Covariance Sites

Journal of Advances in Modeling Earth Systems, Journal Year: 2024, Volume and Issue: 16(10)

Published: Oct. 1, 2024

Abstract Polar‐orbiting satellites have significantly improved our understanding of the terrestrial carbon cycle, yet they are not designed to observe sub‐daily dynamics that can provide unique insight into cycle processes. Geostationary offer remote sensing capabilities at temporal resolutions 5‐min, or even less. This study explores use geostationary satellite data acquired by Operational Environmental Satellite—R Series (GOES‐R) estimate gross primary productivity (GPP) and ecosystem respiration (RECO) using machine learning. We collected processed from 126 AmeriFlux eddy covariance towers in Contiguous United States synchronized with imagery GOES‐R Advanced Baseline Imager (ABI) 2017 2022 develop ML models assess their performance. Tree‐based ensemble regressions showed promising performance for predicting GPP (R 2 0.70 ± 0.11 RMSE 4.04 1.65 μmol m −2 s −1 ) RECO 0.77 0.10 0.90 0.49 on a half‐hourly time step surface products top‐of‐atmosphere observations. Our findings align global efforts utilize improve flux estimation how dioxide fluxes near‐real time.

Language: Английский

---

Learning extreme vegetation response to climate drivers with recurrent neural networks

Nonlinear processes in geophysics, Journal Year: 2024, Volume and Issue: 31(4), P. 535 - 557

Published: Nov. 13, 2024

Abstract. The spectral signatures of vegetation are indicative ecosystem states and health. Spectral indices used to monitor characterized by long-term trends, seasonal fluctuations, responses weather anomalies. This study investigates the potential neural networks in learning predicting response, including extreme behavior from meteorological data. While machine methods, particularly networks, have significantly advanced modeling nonlinear dynamics, it has become standard practice approach problem using recurrent architectures capable capturing effects accommodating both long- short-term memory. We compare four recurrent-based models, which differ their training architecture for at different forest sites Europe: (1) (RNNs), (2) long memory (LSTMs), (3) gated unit (GRUs), (4) echo state (ESNs). our results show minimal quantitative differences performances, ESNs exhibit slightly superior across various metrics. Overall, we that network prove generally suitable prediction yet limitations under conditions. highlights prediction, emphasizing need further research address conditions within dynamics.

Language: Английский

---

An increasing Arctic-boreal CO₂sink offset by wildfires and source regions

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 12, 2024

Abstract The Arctic-Boreal Zone (ABZ) is rapidly warming, impacting its large soil carbon stocks. We use a new compilation of terrestrial ecosystem CO 2 fluxes, geospatial datasets and random forest models to show that although the ABZ was an increasing sink from 2001 2020 (mean ± standard deviation in net exchange: −548 140 Tg C yr -1 ; trend: −14 , p<0.001), more than 30% region source. Tundra regions may have already started function on average as sources, demonstrating critical shift dynamics. After factoring fire emissions, no longer statistically significant (budget: −319 −9 ), with permafrost becoming neutral −24 123 −3 underscoring importance this region.

Language: Английский

---

Predicting multi-annual green roof net ecosystem exchange using machine learning

Building and Environment, Journal Year: 2024, Volume and Issue: 263, P. 111878 - 111878

Published: July 26, 2024

Green roofs are an urban mitigation strategy to increase CO2 uptake by green infrastructure. Reliable quantification of multi-year net ecosystem exchange (NEE) is essential for evaluating carbon balance, but proves challenging due lack long-term data and transferability models. Machine learning can effectively predict NEE identify non-linear patterns spatially temporally upscaling. For the first time, we developed Random Forest (RF) models using eddy covariance (EC) from 2015 2020 extensive roof at Berlin-Brandenburg Airport (BER) understand influence meteorological predictors on prediction transferability. A simple (M1), extended (M2) optimized (M3) model based different combinations were built. M3 performed best, deviating −13 % observed −132.4 gC m−2 a−1 (R2 0.74), with M2 = 0.73) showing similar results. Key volumetric water content (VWC) solar radiation flux densities. The showed robust performance under pronounced environmental conditions. Drought in 2018 introduced significant uncertainties, whereas higher VWC 2019 led enhanced performance. All tended overestimate assimilation underestimate respiration roof. M1 deviated −58 over entire period, indicating other locations not feasible, better potential broader application minimal calibration. This study highlights importance water-related variables available energy demonstrates that RF, incorporating process understanding, NEE.

Language: Английский

---

Escape ratio contributes more than fluorescence yield to SIF-GPP relationship over crops and rainforest

IEEE Geoscience and Remote Sensing Letters, Journal Year: 2024, Volume and Issue: 21, P. 1 - 5

Published: Jan. 1, 2024

Language: Английский

---

Observed increasing light-use efficiency of terrestrial gross primary productivity

Agricultural and Forest Meteorology, Journal Year: 2024, Volume and Issue: 359, P. 110269 - 110269

Published: Oct. 22, 2024

Language: Английский

---