Methane (CH_4) is the second most powerful greenhouse gas after carbon dioxide and plays a crucial role in climate change due to its high global warming potential. Accurately modeling CH_4 fluxes across the globe and at fine temporal scales is essential for understanding its spatial and temporal variability and developing effective mitigation strategies. In this work, we introduce the first-of-its-kind cross-scale global wetland methane benchmark dataset (X-MethaneWet), which synthesizes physics-based model simulation data from TEM-MDM and the real-world observation data from FLUXNET-CH_4. This dataset can offer opportunities for improving global wetland CH_4 modeling and science discovery with new AI algorithms. To set up AI model baselines for methane flux prediction, we evaluate the performance of various sequential deep learning models on X-MethaneWet. Furthermore, we explore four different transfer learning techniques to leverage simulated data from TEM-MDM to improve the generalization of deep learning models on real-world FLUXNET-CH_4 observations. Our extensive experiments demonstrate the effectiveness of these approaches, highlighting their potential for advancing methane emission modeling and identifying new opportunities for developing more accurate and scalable AI-driven climate models.

Methane is a potent greenhouse gas that significantly contributes to global warming. However, accurately estimating global methane emissions and consumption remains challenging due to the complex interactions among environmental drivers that may vary across spatial and temporal scales. Prior data-driven methods often overlook the inherent spatiotemporal heterogeneity of ecosystems, failing to explicitly capture site-specific characteristics and cross-year evolutionary dynamics. To address these issues, we propose the Contrastive Hierarchical Adaptive Meta-network (CHAM-net), a novel framework that explicitly learns from historical context to capture site-specific dynamics. CHAM-net employs a hierarchical encoder–decoder architecture, in which the encoder captures site-specific characteristics from historical data and then dynamically conditions the decoder to generate the final prediction. Experimental results demonstrate that CHAM-net consistently outperforms all baseline methods on both simulation and observational datasets for methane emission and consumption, achieving nRMSE values as low as 0.43 and 0.88 with corresponding R^2 scores up to 0.97 and 0.68 for emission prediction.

In recent decades, human activities have gradually influenced climate variability, consequently changing vegetation dynamics. We examined China's Loess Plateau (LP), a semi-arid, erosion-prone region undergoing large-scale ecological restoration and rapid socioeconomic change, providing a natural setting to contrast climatic and anthropogenic influences. However, annual, temporally continuous quantification of multidimensional human activity-needed to disentangle its effects from climate-remains limited in the LP. We built and validated an annual Integrated Human Activity Index (IHAI) (2000-2020) and quantified EVI responses to precipitation, temperature, and IHAI using Theil-Sen trend estimation, the Mann-Kendall test, Hurst-based tendency types, and partial and multiple correlation analyses. The IHAI experienced a significant increase in LP, whereas the regional mean EVI rose from 0.446 in 2000 to 0.544 in 2020, indicating a trend of 0.002 year-1 (p = 0.002), consistent with greening. Climate signals exhibited geographical heterogeneity; future tendency classifications were mostly influenced by erratic precipitation and reversals in temperature trends, underscoring climatic non-stationarity. The relationships between climate and vegetation were generally favorable. In contrast, those between IHAI and EVI showed regional divergence and often negative correlations, with distinct differences observed across degradation, stability, and restoration zones. These patterns support an LP-wide, province-aware "Target-Match-Maintain" strategy. It prioritizes targeted controls in high-risk degradation zones, matches interventions to local hydroclimatic constraints and human-vegetation coupling, and maintains recovery gains through sustained, performance-based restoration and climate-resilient management to reduce reversal risk.

Accurate prediction of terrestrial ecosystem carbon fluxes (e.g., CO_2, GPP, and CH_4) is essential for understanding the global carbon cycle and managing its impacts. However, prediction remains challenging due to strong spatiotemporal heterogeneity: ecosystem flux responses are constrained by slowly varying regime conditions, while short-term fluctuations are driven by high-frequency dynamic forcings. Most existing learning-based approaches treat environmental covariates as a homogeneous input space, implicitly assuming a global response function, which leads to brittle generalization across heterogeneous ecosystems. In this work, we propose Role-Aware Conditional Inference (RACI), a process-informed learning framework that formulates ecosystem flux prediction as a conditional inference problem. RACI employs hierarchical temporal encoding to disentangle slow regime conditioners from fast dynamic drivers, and incorporates role-aware spatial retrieval that supplies functionally similar and geographically local context for each role. By explicitly modeling these distinct functional roles, RACI enables a model to adapt its predictions across diverse environmental regimes without training separate local models or relying on fixed spatial structures. We evaluate RACI across multiple ecosystem types (wetlands and agricultural systems), carbon fluxes (CO_2, GPP, CH_4), and data sources, including both process-based simulations and observational measurements. Across all settings, RACI consistently outperforms competitive spatiotemporal baselines, demonstrating improved accuracy and spatial generalization under pronounced environmental heterogeneity.

Crop yield mapping is essential for food security and policy making. Recent machine learning (ML) and deep learning (DL) methods have achieved impressive accuracy in crop yield estimation. However, these models require numerous training samples that are scarce in regions with underdeveloped infrastructure. Furthermore, domain shifts between different spatial regions prevent DL models trained in one region from being directly applied to another without domain adaptation. This effect is particularly pronounced between regions with significant climate and environmental variations such as the U.S. and Kenya. To address this issue, we propose using fine-tuning-based transfer learning, which learns general associations between predictors and response variables from the data-abundant source domain and then fine-tunes the model on the data-scarce target domain. We assess the model’s performance on estimating corn yields using Kenya (target domain) and the U.S. (source domain). Feature variables, including time-series vegetation indices (VIs) and sequential meteorological variables from both domains, are used to pre-train and fine-tune the deep neural network model. The model is fine-tuned using data from 5 years (2019–2023) and tested using leave-one-year-out cross validation. The fine-tuned DNN achieves an overall R2 of 0.632—higher than both the U.S.-only and Kenya-only baselines—but paired significance tests show no aggregate difference, though a statistically significant gain does occur in 2023 under anomalous heat conditions. These results demonstrate that fine-tuning can reliably transfer learned representations across continents and, under certain climatic scenarios, yield meaningful improvements.

Increasing evidence highlights the disruptive effects of compound climate extremes on global crop yields under climate change. Existing studies predominantly rely on the whole growing-season scale and relative thresholds, and limit the ability to capture crop physiological sensitivities and yield responses that vary critically across growth stages. Here, we analyzed the spatiotemporal variations, dominant drivers, and potential impacts on the yields of concurrent heat-drought and chilling-rain events for single- and late-rice in southern China from 1981 to 2018. Specifically, we carefully distinguished three sensitive growth stages of rice and stage-specific climate stress types and thresholds based on rice physiology. Temporally, single-rice experienced a significant increase in concurrent heat-drought events, while late-rice experienced a modest rise in chilling-rain events. Spatially, the hotspots of concurrent heat-drought events varied greatly across the three growth stages. These spatial patterns are driven primarily by differences in crop phenology across locations, rather than by the occurrence of extreme climate conditions. The concurrent chilling-rain events of late-rice were widespread within the planting regions, with a higher incidence in certain areas. Path analysis identified heat stress as the primary driver of heat-drought impacts (particularly in jointing-booting and heading-flowering stages), whereas chilling and rain stress exerted comparable effects for late-rice. Our assessment of compound event impacts and sensitivity on rice yield revealed significant growth-stage differences, with comparable yield losses from both concurrent heat-drought and chilling-rain events. Single-rice showed the highest sensitivity to heat-drought events during the grain filling stage, whereas the late-rice exhibited greater sensitivity during the heading-flowering stage. The historical impact on yield diverged markedly across growth stages, with the largest having occurred in the grain filling stage, particularly for heat-drought events. Our study provided important information on compound agroclimatic extremes, in the context of southern China's rice production system, and the results provide important information for risk management and adaptation strategies under climate change.

Global land cover has changed during the past decades, influencing biogeochemical cycles and the global climate system. This study aimed to improve understanding of global land cover dynamics to enable more effective future land management practices and conservation actions. This study quantified interannual changes in global land cover types from 2001 to 2020 and distinguished intermittent transitions from stable gains and losses. From the interannual perspective, we found that global barren lands, forests, shrublands, and snow-covered areas decreased by 5281, 1804, 952, and 188 kha yr −1 , respectively. In contrast, grasslands, croplands, urban areas, and water bodies increased at 6529, 1407, 237, and 51 kha yr −1 , respectively, from 2001 to 2020. According to the definitions provided in this paper, of the global forest areas, 75% was Stable (no change), 4% was Gain, 5% was Loss, and 16% was Unstable. Of the cropland areas, 56% was Stable, 9% was Gain, 9% was Loss, and 26% was Unstable. Hotspots for forest loss were Brazil, the Rest of South America, and Sub-Saharan Africa, and grassland was the most common land cover classification following forest loss. The global cropland expansion hotspots were Brazil, Canada, China, India, and the Rest of South America. The cropland gains were mainly converted from grasslands. On the other hand, barren areas in China and Middle Eastern and North Africa were changed to grasslands. A certain amount of shrublands were changed to forest in temperate regions. This paper provided land cover changes at a 500 m spatial resolution as a benchmark for future assessments. The findings showed that unstable pixels play an important role in determining the sources of uncertainty when assessing land cover changes using satellite data. Land cover assessments are sensitive to the time steps used for analysis and the definition of changes.

Crop phenology provides essential information for crop management and production. Satellite-based methods are commonly used for phenology estimation but still struggle to capture interannual variations of phenological events. The importance of climate variation in crop phenology has been well acknowledged, but the potential of incorporating climate data to improve phenology estimation remains unclear. Here, we developed a hybrid model by incorporating the growth-specific climate predictors and satellite-derived phenology using random forest approach. Results showed that our hybrid model successfully reduced errors by over 60% compared to traditional satellite-based methods. The inclusion of climate data provided additional contributions beyond what was offered by satellite data, resulting in a 13% average improvement in R ^2 . Among climate predictors, temperature-related indicators contributed the most to accuracy enhancement. Additionally, CSIF outperformed LAI in the hybrid model in terms of absolute error, due to its finer temporal resolution. Our hybrid model highlights the importance of considering the diverse climatic information to further improve crop phenology estimation, rather than relying solely on satellite data. We expect our proposed model can offer new insights into improving crop phenology estimation and understanding the effects of climate variations on crop phenology.

Understanding the changes in crop yield stability (mu/sigma) under climate change is critical for food security and farmer livelihoods. Unstable crop yields have been identified as detrimental to international food trade and markets. However, the association between climate extremes and crop yield stability has not been well documented. Here, we present the sensitivity of corn and soybean yield stability to heat, drought and excessive wetness by using statistical modeling for rainfed corn and soybean based on survey yield records in the US Midwest. The results using survey data indicate that increased heat, drought and excessive wetness are collectively associated with reduced yield stability for corn and soybean. We find that the changes in corn and soybean yield stability in the US Midwest are predominantly related to heat stress. Additionally, irrigation can mitigate the yield stability reduction associated with heat and drought. In contrast, well-irrigated yield is more sensitive to excess wet. Our results highlight the importance of examining the correlation between climate extremes and crop yield stability.

Abstract Wetland methane (CH4) emissions have a significant impact on the global climate system. However, the current estimation of wetland CH4 emissions at the global scale still has large uncertainties. Here we developed six distinct bottom‐up machine learning (ML) models using in situ CH4 fluxes from both chamber measurements and the Fluxnet‐CH4 network. To reduce uncertainties, we adopted a multi‐model ensemble (MME) approach to estimate CH4 emissions. Precipitation, air temperature, soil properties, wetland types, and climate types are considered in developing the models. The MME is then extrapolated to the global scale to estimate CH4 emissions from 1979 to 2099. We found that the annual wetland CH4 emissions are 146.6 ± 12.2 Tg CH4 yr−1 (1 Tg = 1012 g) from 1979 to 2022. Future emissions will reach 165.8 ± 11.6, 185.6 ± 15.0, and 193.6 ± 17.2 Tg CH4 yr−1 in the last two decades of the 21st century under SSP126, SSP370, and SSP585 scenarios, respectively. Northern Europe and near‐equatorial areas are the current emission hotspots. To further constrain the quantification uncertainty, research priorities should be directed to comprehensive CH4 measurements and better characterization of spatial dynamics of wetland areas. Our data‐driven ML‐based global wetland CH4 emission products for both the contemporary and the 21st century shall facilitate future global CH4 cycle studies.

Accurate and timely monitoring of biochemical and biophysical traits associated with crop growth is essential for indicating crop growth status and yield prediction for precise field management. This study evaluated the application of three combinations of feature selection and machine learning regression techniques based on unmanned aerial vehicle (UAV) multispectral images for estimating the bio-parameters, including leaf area index (LAI), leaf chlorophyll content (LCC), and canopy chlorophyll content (CCC), at key growth stages of winter wheat. The performance of Support Vector Regression (SVR) in combination with Sequential Forward Selection (SFS) for the bio-parameters estimation was compared with that of Least Absolute Shrinkage and Selection Operator (LASSO) regression and Random Forest (RF) regression with internal feature selectors. A consumer-grade multispectral UAV was used to conduct four flight campaigns over a split-plot experimental field with various nitrogen fertilizer treatments during a growing season of winter wheat. Eighteen spectral variables were used as the input candidates for analyses against the three bio-parameters at four growth stages. Compared to LASSO and RF internal feature selectors, the SFS algorithm selects the least input variables for each crop bio-parameter model, which can reduce data redundancy while improving model efficiency. The results of the SFS-SVR method show better accuracy and robustness in predicting winter wheat bio-parameter traits during the four growth stages. The regression model developed based on SFS-SVR for LAI, LCC, and CCC, had the best predictive accuracy in terms of coefficients of determination (R2), root mean square error (RMSE) and relative predictive deviation (RPD) of 0.967, 0.225 and 4.905 at the early filling stage, 0.912, 2.711 μg/cm2 and 2.872 at the heading stage, and 0.968, 0.147 g/m2 and 5.279 at the booting stage, respectively. Furthermore, the spatial distributions in the retrieved winter wheat bio-parameter maps accurately depicted the application of the fertilization treatments across the experimental field, and further statistical analysis revealed the variations in the bio-parameters and yield under different nitrogen fertilization treatments. This study provides a reference for monitoring and estimating winter wheat bio-parameters based on UAV multispectral imagery during specific crop phenology periods.

Accurate and efficient estimation of biochemical traits, including leaf index area (LAI), leaf chlorophyll content (LCC) and canopy chlorophyll content (CCC), is crucial for crop growth monitoring in agricultural management. Recent advancements in unmanned aerial vehicle (UAV) multispectral remote sensing have enabled fast and costeffective measurements of these traits. However, traditional statistical regression models trained on specific datasets lack scalability and transferability across practical field conditions without retraining. This study proposed an efficient physics-informed transfer learning model (PITL) for winter wheat biochemical traits estimation from UAV multispectral data. The PITL integrates the strengths of physical radiative transfer simulations and deep neural network architectures through transfer learning to improve the estimation of biochemical traits from UAV multispectral data. The PITL was tested with convolutional neural network (CNN), deep neural network (DNN), and long short-term memory (LSTM) architectures. Results indicated that PITLDNN had better accuracy than PITLCNN and PITLLSTM models in predicting LAI (R2=0.94, RMSE = 0.32 m(2)/m(2)), LCC (R-2=0.81, RMSE = 5.20 mu g/cm(2)) and CCC (R-2=0.928, RMSE = 0.2 g/m(2)). Moreover, PITLDNN demonstrated higher capability in computational efficiency, making it suitable for processing large volumes of UAV multispectral data in crop growth monitoring applications. Furthermore, PITL's integration of radiative transfer knowledge with labeled field data yielded higher predictive accuracy compared to physically-based inversion model, pure data-driven deep neural network approaches, and hybrid models. This study highlighted the performance of PITLDNN in accurately and efficiently quantifing biochemical traits from UAV multispectral data, thereby providing timely and accurate information for guiding crop growth monitoring applications.

Abstract. There is increasing concern regarding the impact of compound agroclimatic extreme events on crop yield, particularly in the context of projected increases in their frequency and intensity due to climate change. While previous studies have generally focused on compound hot and dry events in maize and wheat using growing-season relative thresholds, the time-variant physiological sensitivity of crops to climate extremes has not been sufficiently considered. We determined the spatiotemporal variations of compound climate extremes (CEs) for single- and late-rice in southern China during 1980–2014 and their underlying drivers using growth-stage specific physiological thresholds. Specifically, we carefully distinguished between concurrent compound events (CCEs) and consecutive compound events (CSEs). Our results indicated an increasing trend of compound hot-dry events for single-rice, but a decreasing trend of compound chilling-rainy events for late-rice. Spatially, the hotspots of compound hot-dry events for single-rice shifted from the lower Yangtze River Basin to its upper stream, and were dominated by the spatial differences in phenology rather than the occurrence of extreme events. The hotspots of compound chilling-rainy events for late-rice remained concentrated near the northwest edges of late-rice growing areas, indicating the limitation of thermal conditions. The occurrence and duration of CCEs was closely related to local temperature-moisture coupling (negative correlation). A path analysis suggested that temperature was the dominant factor influencing the changes in compound hot-dry events for single-rice. For the changes in compound chilling-rainy events for late-rice, the effect of temperature was only slightly larger than that of moisture. Our study has improved the understanding of compound climate extremes in China’s rice production system, and the results provide important information for risk management and adaptation strategies under climate change.

Machine learning (ML) has proven to be a powerful tool for utilizing the rapidly increasing amounts of remote sensing data for environmental monitoring. Yet ML models often require a substantial amount of ground truth labels for training, and models trained using labeled data from one domain often demonstrate poor performance when directly applied to other domains. Transfer learning (TL) has emerged as a promising strategy to address domain shift and alleviate the need for labeled data. Here we provide the first systematic review of TL studies in environmental remote sensing. We start by defining the different forms of domain shift and then describe five commonly used TL techniques. We then present the results of a systematic search for peer-reviewed articles published between 2017 and 2022, which identified 1676 papers. Applications of TL in remote sensing have increased rapidly, with nearly 10 times more publications in 2022 than in 2017. Across seven categories of applications (land cover mapping, vegetation monitoring, soil property estimation, crop yield prediction, biodiversity monitoring, water resources management, and natural disaster management) we identify several recent successes of TL as well as some remaining research gaps. Finally, we highlight the need to organize benchmark datasets explicitly for TL in remote sensing for model evaluation. We also discuss potential research directions for TL studies in environmental remote sensing, such as realizing scale transfer, improving model interpretability, and leveraging foundation models for remote sensing tasks.

Climate warming affects global livestock productivity. The meat yield from cattle farming (cattle meat per animal) represents livestock productivity at the individual level. However, the impact of warming on cattle meat yield at a global scale is not well understood. In this study, we combine country-level data on the annual meat yield from cattle farming and socioeconomic data from 1961 to 2020 with climate projections from General Circulation Models. The findings show that cattle meat yield increases as temperatures rise from low to medium and then decreases when annual average temperatures exceed 7 °C; this repose is pronounced in the grassland-based livestock system. Further, we show that warming creates unequal impacts between high- and low-income countries due to the divergent baseline temperature conditions. Future warming aggravates these unequal burdens between countries, with the most pronounced effects observed under the upper-middle emissions scenario.

Livestock snow disaster is one of the most serious natural disasters in pastoral areas during cold seasons, leading to substantial livestock losses over the past few decades. Accurate assessment of livestock exposure is crucial for reducing the risk of livestock snow disaster, while much less is known about the livestock exposure due to the lack of available gridded livestock density data with long time series. In this study, the gridded density datasets of large livestock, sheep and total livestock at a spatial resolution of 1 km across the Qinghai Plateau (QP) during 1983-2018 were generated by developing livestock density models and disaggregating livestock census data. Then, the spatiotemporal patterns of livestock exposure to snow hazards in the QP were explored and the relative contributions of influencing factors (snow hazard, livestock density and prevention capacity) were quantified. There was a significant negative trend (p < 0.10) of -1.16 x 10(6) standard sheep units (SSUs) decade(-1) in annual total livestock number and -4.67 x 10(7) SSUs-day decade(-1) (p < 0.05) in annual total livestock exposure across the QP from 1983 to 2018. Spatially, snow hazard, livestock density and prevention capacity dominated the total livestock exposure changes in 38.11 %, 22.70 % and 34.95 % of pastoral areas, respectively. However, the reduction in total livestock exposure across the QP was primarily contributed by prevention capacity (56.13 %). This study provides a comprehensive understanding and scientific decision-making basis for livestock snow disaster mitigation and risk management, contributing to the high-quality development of pastoral areas.

Compound climate events are major threats to crop production under climate change. However, the heterogeneity in the impact of compound events on crop yield and its drivers remain poorly understood. Herein, we used empirical approach to evaluate the impact of compound hot–dry (HD) and cold–wet (CW) events on maize yield in China at the county level from 1990 to 2016, with a special focus on the spatial heterogeneity. Our findings indicate comparable impact of extremely compound CW events (−12.8 ± 3.6%) on maize yield loss to extremely compound HD events (−11.3 ± 2.1%). The spatial pattern of compound HD and CW events impacts on maize yield was dominantly associated with moisture regime, followed by management practices and soil properties. Specifically, drier counties and counties with less fraction of clay soil and organic carbon tend to experience greater yield loss due to compound HD events, and wet condition, excessive fertilizer, clay soil and rich organic carbon aggravate the maize yield loss due to compound CW events. Moreover, the land–atmosphere coupling exacerbated the heterogeneous yield impact through divergent heat transfer. In drier regions, the greater proportion of sensible heat creates a positive feedback between drier land and hotter atmosphere. In contrast, the greater proportion of latent heat in wetter regions results in a positive feedback between wetter land and colder atmosphere. Our results highlighted a critical element to explore in further studies focused on the land–atmosphere coupling in agricultural risk under climate change.

Understanding the impact of climate change on year-to-year variation of crop yield is critical to global food stability and security. While crop model emulators are believed to be lightweight tools to replace the models, few emulators have been developed to capture such interannual variation of crop yield in response to climate variability. In this study, we developed a statistical emulator with a machine learning algorithm to reproduce the response of year-to-year variation of four crop yields to CO2 (C), temperature (T), water (W), and nitrogen (N) perturbations defined in the Global Gridded Crop Model Intercomparison Project (GGCMI) phase 2. The emulators were able to explain more than 52 % of the variance of simulated yield and performed well in capturing the year-to-year variation of global average and gridded crop yield over current croplands in the baseline. With the changes in CO2–temperature–water–nitrogen (CTWN) perturbations, the emulators could reproduce the year-to-year variation of crop yield well over most current cropland. The variation of R and the mean absolute error was small under the single CTWN perturbations and dual-factor perturbations. These emulators thus provide statistical response surfaces of yield, including both its mean and interannual variability, to climate factors. They could facilitate spatiotemporal downscaling of crop model simulation, projecting the changes in crop yield variability in the future and serving as a lightweight tool for multi-model ensemble simulation. The emulators enhanced the flexibility of crop yield estimates and expanded the application of large-ensemble simulations of crop yield under climate change.

Abstract Climate warming is one of the major threats to global livestock production. However, the impact of climate warming on livestock meat yield at global scale is rarely investigated. In this study, we present a statistical evidence from country-level cattle meat yield that the response of global cattle meat yield to climate warming is invert-U quadratic nonlinear, which is more obvious in the grassland-based livestock system. Moreover, the nonlinear response determines that climate warming creates unequal burden between high- and low- income countries due to the divergent baseline temperature condition. Future climate warming aggravates these unequal burdens, with the most pronounced effects observed under SSP585. Our results highlight the need for focusing on the global food security under climate change from livestock meat production.

Incorporating seasonality into livestock spatial distribution is of great significance for studying the complex system interaction between climate, vegetation, water, and herder activities, associated with livestock. The Qinghai-Tibet Plateau (QTP) has the world's most elevated pastoral area and is a hot spot for global environmental change. This study provides the spatial distribution of cattle, sheep, and livestock grazing on the warm-season and cold-season pastures at a 15 arc-second spatial resolution on the QTP. Warm/cold-season pastures were delineated by identifying the key elements that affect the seasonal distribution of grazing and combining the random forest classification model, and the average area under the receiver operating characteristic curve of the model is 0.98. Spatial disaggregation weights were derived using the prediction from a random forest model that linked county-level census livestock numbers to topography, climate, vegetation, and socioeconomic predictors. The coefficients of determination of external cross-scale validations between dasymetric mapping results and township census data range from 0.52 to 0.70. The data could provide important information for further modeling of human-environment interaction under climate change for this region.