Integrating physical processes with machine learning has advanced evapotranspiration (ET) simulation, yet most hybrid models fail to partition total ET into its components: soil evaporation (E) and vegetation transpiration (T). This study introduces Residual Neural Network-Penman-Monteith (RNN-PM), a novel hybrid dual-source ET model designed to overcome this limitation. The model synergizes the physically-based Penman-Monteith framework with three specialized residual neural networks trained to estimate key conductance parameters (canopy conductance, soil surface conductance, and aerodynamic conductance). This explicit parameterization allows for the direct partitioning of total ET. Validation at National Ecological Observatory Network (NEON) flux sites using high-frequency partitioned E and T shows that RNN-PM reliably reproduces ET and the transpiration fraction (T/ET). For ET, the model achieves an average Kling-Gupta efficiency (KGE) of 0.89 and a root-meansquare error (RMSE) of 0.55 mm/day; for T/ET, the KGE is 0.87 with an RMSE of 0.06. Furthermore, RNN-PM demonstrates robust generalization, accurately simulating ET and its components well beyond the initial training dataset, even under extreme climatic conditions. This study extended the analysis by comparing the RNN-PM model with seven established dual-source ET models. The results indicate that RNN-PM outperforms both conventional machine learning models and purely physical process-based models in simulating ET components in most cases. Among the purely physical process-based dual-source models, those based on surface temperature decomposition showed improved performance as the leaf area index (LAI) decreased when evaluated against high-frequency ET component datasets. In contrast, the performance of conductance-based dual-source models declined with decreasing LAI. Although purely machine learning-based models can produce relatively accurate simulations of ET components, they often exhibit limited generalization capability, an issue that the RNN-PM model effectively overcomes. Ultimately, the RNN-PM model represents a significant advance in simulating ET components, offering a novel and scalable approach for improving the representation of land-atmosphere interactions in Earth system models.

Abstract Agricultural expansion continues to reshape landscapes across Africa, yet how patterns of cropland change vary across farming systems of different scales remains poorly understood. Here, we combined annual high-resolution field boundary mapping from Planet imagery with an empirically derived field-to-farm size relationship to examine cropland dynamics, farm-scale patterns, and their environmental and socioeconomic correlates across Zambia from 2018 to 2024. Cropland expansion was highly spatially concentrated, with local expansion rates reaching up to 170 ha yr -1 within 0.05◦ (≈5.5 km) grid cells, and occurred primarily along the margins of established agricultural regions. Areas dominated by medium-scale farms exhibited substantially faster cropland growth than areas dominated by smallholder farms. Similarly, when field size was considered as a continuous variable, cells characterized by larger median field sizes showed markedly higher expansion rates, exceeding 35 ha yr -1 in some areas, whereas expansion rates in areas dominated by smaller fields were generally below 5 ha yr -1 . Meanwhile, stable croplands showed little evidence of widespread field consolidation, with approximately 84% of grid cells exhibiting field-size trends within ±0.05 ha yr -1 , suggesting limited evidence for farm-scale expansion through widespread field consolidation within existing croplands. Environ mental and socioeconomic analyses further indicated that land tenure, represented by the proportion of land under state tenure, was the factor most consistently associated with the prevalence of medium scale farms, whereas climate stability was more strongly associated with total cropland area than with farm-size composition. Together, these findings suggest that recent agricultural growth in Zambia has been closely associated with the increasing prevalence of medium-scale farming, while widespread field consolidation within existing croplands appears limited. More broadly, the study demonstrates how high-resolution Earth observation can provide spatially explicit evidence of agricultural transformation and emerging farm structures across rapidly changing African farming systems.

Soil moisture drydown patterns encode signatures of vegetation water‐use. Previous characterizations of the drydown patterns assume a static linear relationship between water‐limited transpiration and available moisture. However, ecohydrological studies show that vegetation exhibits a spectrum of responses to water availability, suggesting that soil moisture loss functions may be nonlinear. To represent these dynamics, we introduce a nonlinearity parameter to the loss function. Our analysis shows that the nonlinear loss model improves the characterization of the satellite‐observed soil moisture drydowns. Globally, functional responses of drydowns are dominated by convex nonlinearity, showing less ecosystem water loss in dry soils than the linear loss function predicts. We find distinct degrees of nonlinearity among different vegetation types; areas with non‐woody vegetation more frequently exhibit a concave nonlinearity, the signature of aggressive water‐use strategies. We propose the nonlinear loss function as a continuous and dynamic framework to represent vegetation water‐use under changing water availability.

Accurate estimation of evapotranspiration (ET) in drylands is critically dependent on capturing fine-scale spatial variability, yet current thermal remote sensing approaches face significant scaling limitations. While satellite-based thermal imagery provides broad coverage for ET estimation, its coarse resolution fails to capture the heterogeneous vegetation patterns characteristic of dryland ecosystems, leading to systematic biases in ET estimates. The non-linear relationship between land surface temperature (LST) and ET means that coarse-resolution LST measurements cannot simply be averaged to estimate ecosystem-scale ET. Instead, the underlying spatial variance in LST must be properly accounted for when scaling between observations at different resolutions. Here, we demonstrate an approach using very high resolution (VHR) UAV-derived thermal imagery (0.3-m resolution) combined with multi-scale satellite observations (up to 90-m resolution) to develop scaling relationships between LST variance and spatial resolution. We show how these relationships vary with vegetation composition and seasonal dynamics in a dryland ecosystem over one year. By modeling how LST variance changes across scales, we can better estimate ET from coarser thermal imagery while preserving the influence of fine-scale heterogeneity. Our results indicate that vegetation pattern and phenological stage significantly influence scaling behavior, allowing us to identify optimal measurement resolutions for different ecosystem conditions. This approach reduces uncertainty in ET estimates from satellite thermal imagery by incorporating the effects of sub-pixel spatial variability revealed by VHR observations. The scaling relationships we develop provide a framework for improving regional ET estimates in drylands while accounting for their characteristic fine-scale vegetation patterns.

1.Soil CO₂ flux is a critical component of ecosystem carbon cycling, but due to high cost and mechanistic constraints, existing measurement systems are often limited by trade-offs between resolution (temporal and spatial), and spatial coverage. These constraints hinder efforts to monitor soil fluxes across diverse, heterogeneous landscapes and environmental gradients. 2.We developed Fluxbot 2.0, a low-cost, autonomous chamber system capable of continuous, distributed soil CO₂ flux measurements without external power or infrastructure. To assess its capability to capture landscape-scale variability, we deployed two Fluxbot 2.0 arrays, one at each of two hemlock forest sites in Harvard Forest, Massachusetts, USA, and compared its estimates of flux to those from existing, well-established automated chamber arrays that rely on multiplexed chambers and high-accuracy CO2 analyzer units. 3. Fluxbots successfully captured site means, spatial variability, temporal patterns, and environmental responses, including temperature-driven flux dynamics. These measurements reflected differences in forest conditions between two sites and showed that distributed arrays of low-cost sensors can effectively capture both fine-scale variability and broader patterns across a landscape. 4. By enabling low-cost, autonomous monitoring of soil carbon flux in strategically distributed arrays, Fluxbot 2.0 addresses key gaps in existing soil CO₂ flux datasets. The system facilitates measurements across environmental gradients and heterogeneous landscapes, supporting research on soil carbon dynamics and biotic interactions that influence carbon cycling.

Vegetation responses to soil moisture limitation play a key role in land-atmosphere interactions and are a major source of uncertainty in future projections of the global water and carbon cycles. Vegetation water-use strategies-that is, how plants regulate transpiration rates as the soil dries-are highly dynamic across space and time, presenting a major challenge to inferring ecosystem responses to water limitation. Here we show that, when aggregated globally, water-use strategies derived from point-based soil moisture observations exhibit emergent patterns across and within climates and vegetation types along a spectrum of aggressive to conservative responses to water limitation. Water use becomes more conservative, declining more rapidly as the soil dries, as mean annual precipitation increases and as woody cover increases from grasslands to savannahs to forests. We embed this empirical synthesis within an ecohydrological framework to show that key ecological (leaf area) and hydroclimatic (aridity) factors driving demand for water explain up to 77% of the variance in water-use strategies within ecosystem types. All biomes respond to ecological and hydroclimatic demand by shifting towards more aggressive water-use strategies. However, woodlands reach a threshold beyond which water use becomes increasingly conservative, probably reflecting the greater hydraulic risk and cost of tissue damage associated with sustaining high transpiration rates under water limitation for trees than grasses. These findings highlight the importance of characterizing the dynamic nature of vegetation water-use strategies to improve predictions of ecosystem responses to climate change.

Riparian corridors act as thermal and moisture refugia for a range of plant and animal species, particularly in water-limited environments. Declining water tables, increasing temperatures, and an increase in extreme hydrologic events due to climate change threaten the diversity of life these landscapes support. Successful adaptive conservation management strategies require an understanding of how species are responding to climate change and an ability to anticipate how changing patterns of water resource availability and demand will alter vegetation patterns and processes. Here, we investigate dryland riparian plant responses to fluctuating water availability and atmospheric demand using a novel drone-based approach for estimating transpiration. Integrating thermal imagery, structural data, and a suite of environmental sensors mounted on an unmanned aerial vehicle (UAV) platform, this approach was specifically designed to capture fine-scale functional data and variation in individual-level plant functional traits within riparian ecosystems and allows for efficient calculation of evapotranspiration for a site solely using data collected from the UAV. Using UAV-based measurements of transpiration across seasonal, diurnal, and spatial gradients of water stress, we quantify individual-scale hydraulic sensitivity to fluctuating water availability and atmospheric moisture demand. Finally, we highlight how these fine-scale estimates of plant water use facilitate understanding of how ecologically important plants respond to the increasingly variable hydrologic regimes that sustain them, yielding valuable insights into how such ecosystems will evolve in the face of global environmental change.

<p>Expanding access to remotely sensed Earth observations provides us with an opportunity to examine the underlying spatiotemporal coupling between vegetation, both natural and managed, and the hydroclimate. Applying approximately 20 years of satellite records, we demonstrate a method to quantify the sensitivity and stability of land-atmosphere interactions. Here we evaluate the predictability of vegetation via the Normalized Difference Vegetation Index (NDVI) across croplands, shrublands, grasslands, and woodlands of East Africa as it relates to fluctuations in precipitation, soil moisture, evapotranspiration, and land surfaced temperature. In this study, we detect the strength of state dependency among these variables at the dekadal (10-day) to monthly scale using a data-driven approach known as Empirical Dynamic Modeling (EDM). There is notable spatial variability in NDVI predictability, with equatorial areas generally expressing the poorest skill, which can be attributed to the inconsistent rainfall seasonality and high aridity. Woodlands exhibit strong predictability throughout the region while vegetation response to environmental drivers in grasslands is less reliable. Our results suggest water availability, uptake and storage are important factors influencing the NDVI cycle. For a one-month lead time, high predictive skill can be retrieved from the time series, though skill weakens by a four- to sixth-month lead, at which point the overall seasonality appears to play a dominant role. One contribution to highlight is the advancement in our understanding of the relationship between vegetation and land surface temperature, which is particularly valuable in drought-prone East Africa. In this presentation, we introduce an application of EDM for biogeosciences, assess how historical seasonal information of the hydroclimate and vegetation across various land use and land covers can inform future environmental patterns, and identify critical areas of inquiry with a changing climate and extending agricultural production.</p>

Evapotranspiration (ET) is the largest loss term in the terrestrial water balance and plays a key role in the energy and carbon cycles. Accurate and timely measurements of ET are critical for understanding the ecosystem responses to climate change and managing water resources. Traditional methods for measuring ET are either highly individualistic leaf- or stem-scale approaches or large-scale tools that aggregate across entire landscapes. Unmanned aerial vehicles (UAVs) constitute a new frontier in measurement of ET that bridges the gap between in situ measurements and remotely sensed observations of water and energy fluxes. With advances in sensor technology and data processing algorithms, UAV-based remote sensing of ET provides both an avenue to refine satellite-based algorithms for retrieving water use and an improved understanding of the fundamental exchange processes between vegetation and the atmosphere.We present an approach for estimating ET at leaf to landscape scales using thermal imagery, structural data, and a suite of environmental sensors mounted on a UAV platform. Our approach derives ET solely from UAV-acquired data using a combined atmospheric profiling and surface energy balance algorithm. Centimeter-scale leaf position and orientation information derived from Structure-from-Motion (SfM) are integrated with the functional data to constrain available energy, allowing for multi-scale estimation of plant water use within and across canopies.Using thermal imagery and a suite of environmental sensors mounted on a UAV platform, we calculated ET of a Mediterranean grassland in Southern California at <1-m spatial resolution for 16 flights across the 2021 and 2022 growing seasons. We compare UAV-derived fluxes using four different formulations of aerodynamic resistance to measurements from an eddy covariance tower at the site. We then discuss the relative importance of surface temperature, aerodynamic terms, and meteorological variables for calculating ET from surface energy balance, highlighting the limitations of current approaches and the potential opportunities for future studies.

CONTEXT: Smallholder farmers in Africa are among those most impacted by climate change. Employing stra-tegies such as planting early maturing or drought tolerant hybrid seeds is one common climate adaptive strategy for these households. However, seed choice has become increasingly complex for farmers. One way farmers look for clarity about seeds is to consult with other farmers. OBJECTIVE: We investigate smallholders' advice seeking within the context of a community water project in rural Kenya, a type of community-based common pool resource management organization. We examine a maize seed advice seeking network and compare it with a more general advice seeking network to better understand the social networks of maize seed advice seeking, and to characterize how peer-to-peer advice networks might factor into farmer decision-making about seeds. METHODS: We use exponential random graph modeling for the maize seed advice and general advice networks to test what factors predict advice-seeking among farmers in 104, or 92% of households in the community water project.

Key Points Earth's Future thanks its reviewers who contributed in 2022

Accurate and operational indicators of the start of growing season (SOS) are critical for crop modeling, famine early warning, and agricultural management in the developing world. Erroneous SOS estimates–late, or early, relative to actual planting dates–can lead to inaccurate crop production and food-availability forecasts. Adapting rainfed agriculture to climate change requires improved harmonization of planting with the onset of rains, and the rising ubiquity of mobile phones in east Africa enables real-time monitoring of this important agricultural decision. We investigate whether antecedent agro-meteorological variables and household-level attributes can be used to predict planting dates of small-scale maize producers in central Kenya. Using random forest models, we compare remote estimates of SOS with field-level survey data of actual planting dates. We compare three years of planting dates (2016–2018) for two rainy seasons (the October-to-December short rains, and the March-to-May long rains) gathered from weekly Short Message Service (SMS) mobile phone surveys. In situ data are compared to SOS from the Water Requirement Satisfaction Index (SOSWRSI) and other agro-meteorological variables from Earth observation (EO) datasets (rainfall, NDVI, and evaporative demand). The majority of farmers planted within 20 days of the SOSWRSI from 2016 to 2018. In the 2016 long rains season, many farmers reported planting late, which corresponds to drought conditions. We find that models relying solely on EO variables perform as well as models using both socio-economic and EO variables. The predictive accuracy of EO variables appears to be insensitive to differences in reference periods that were tested for deriving EO anomalies (1, 3, 5, or 10 years). As such, it would appear that farmers are either responding to short-term weather conditions (e.g., intra-seasonal variability), or longer trends than were included in this study (e.g., 25–30 years), when planting. The methodologies used in this study, weekly SMS surveys, provide an operational means for estimating farmer behaviors–information which is traditionally difficult and costly to collect.

Extreme hot-humid heat impacts both urban and rural livelihoods, reducing labor output and damaging health. As such, increasing exposure to hot-humid heat may be reducing food security for both rural and urban household in Africa. Yet, due to a lack of fine-resolution meteorological data, we have a poor understanding of where urban and rural exposure to hot-humid heat is impacting food security across the continent’s diverse geographies. To fill this gap, using more than 20,000 geo-located surveys from the Demographic and Health Survey Program, we map how the spatial relationship between household-level food security and heat exposure has varied among rural and urban populations since the 1980s. We document spatial and temporal heterogeneity, identifying areas of concern where dangerously hot-humid heat is increasingly co-impacting both urban and rural food security outcomes. Given that hot-humid heat waves will worsen across much of Africa as we warm our climate, our results add to growing calls for effective extreme heat warning systems, including seasonal forecasts, tailored to reduce the impacts of hot humid-heat for all people, regardless of where they live.

The partitioning of evapotranspiration (ET) into surface evaporation (E) and stomatal-based transpiration (T) is essential for analyzing the water cycle and earth surface energy budget. Similarly, the partitioning of net ecosystem exchange (NEE) of carbon dioxide into respiration (R) and photosynthesis (P) is needed to quantify the controls on its sources and sinks. Promising approaches to obtain these components from field measurements include partitioning models based on analysis of conventional high frequency eddy-covariance data. Here, two such existing approaches, based on similarity between non-stomatal (R and E) and stomatal (P and T) components, are considered: the Modified Relaxed Eddy Accumulation (MREA) and Flux-Variance Similarity (FVS) models. Moreover, a simpler technique is proposed based on a Conditional Eddy-Covariance (CEC) scheme. All approaches were evaluated against independent estimates of transpiration and respiration. The CEC method agreed better with measurements of transpiration over a grass field, with a smaller root mean square error (5.9 W m(-2)) and higher correlation (0.96). At a forest site, better agreement with soil respiration was found for FVS above the canopy, while CEC and MREA performed better below the canopy. Further application of these methods over a vineyard and a pine forest across different seasons provided insight into the main strengths and weaknesses of each approach. FVS and MREA converge less often when ground flux components dominate, while CEC might result in noisy P and R for small NEE. Finally, in the CEC and MREA framework, the ratio T/ET is shown to be related to the correlation coefficient for carbon dioxide and water vapor concentrations, which can thus be used as a qualitative measure of the importance of stomatal and non-stomatal components. Overall, these results advance the understanding of the skill and agreement of all three methods, and inform future studies where the various approaches can be applied simultaneously and intercompared.

Mapping the characteristics of Africa's smallholder-dominated croplands, including the sizes and numbers of fields, can provide critical insights into food security and a range of other socioeconomic and environmental concerns. However, accurately mapping these systems is difficult because there is 1) a spatial and temporal mismatch between satellite sensors and smallholder fields, and 2) a lack of high-quality labels needed to train and assess machine learning classifiers. We developed an approach designed to address these two problems, and used it to map Ghana’s croplands. To overcome the spatio-temporal mismatch, we converted daily, high resolution imagery into two cloud-free composites (the primary growing season and subsequent dry season) covering the 2018 agricultural year, providing a seasonal contrast that helps to improve classification accuracy. To address the problem of label availability, we created a platform that rigorously assesses and minimizes label error, and used it to iteratively train a Random Forests classifier with active learning, which identifies the most informative training sample based on prediction uncertainty. Minimizing label errors improved model F1 scores by up to 25%. Active learning increased F1 scores by an average of 9.1% between first and last training iterations, and 2.3% more than models trained with randomly selected labels. We used the resulting 3.7 m map of cropland probabilities within a segmentation algorithm to delineate crop field boundaries. Using an independent map reference sample (n=1,207), we found that the cropland probability and field boundary maps had respective overall accuracies of 88% and 86.7%, user’s accuracies for the cropland class of 61.2% and 78.9%, and producer’s accuracies of 67.3% and 58.2%. An unbiased area estimate calculated from the map reference sample indicates that cropland covers 17.1% (15.4-18.9%) of Ghana. Using the most accurate validation labels to correct for biases in the segmented field boundaries map, we estimated that the average size and total number of field in Ghana are 1.73 ha and 1,662,281, respectively. Our results demonstrate an adaptable and transferable approach for developing annual, country-scale maps of crop field boundaries, with several features that effectively mitigate the errors inherent in remote sensing of smallholder-dominated agriculture.

Earth and Space Science Open Archive Presented WorkOpen AccessYou are viewing the latest version by default [v1]Phenological Classification and Atmospheric Drought Response of Riparian Vegetation in Drylands of the Southwestern United StatesAuthorsConorMcMahonDarRobertsMichaelSingeriDKellyCaylorJohnStellaSee all authors Conor McMahonCorresponding AuthorUniversity of California Santa Barbaraview email addressThe email was not providedcopy email addressDar RobertsUniversity of California Santa Barbaraview email addressThe email was not providedcopy email addressMichael SingeriDCardiff UniversityiDhttps://orcid.org/0000-0002-6899-2224view email addressThe email was not providedcopy email addressKelly CaylorUniversity of Californiaview email addressThe email was not providedcopy email addressJohn StellaSUNY College of Environmental Science and Forestryview email addressThe email was not providedcopy email address

We propose a sensing system comprising a large network of tiny, battery-less, Radio Frequency (RF)-powered sensors that use backscatter communication. The sensors use an entirely passive technique to 'sense' the parameters of the wireless channel between themselves. Since the material properties influence RF channels, this fine-grain sensing can uncover multiple material properties both at a large scale and fine spatial resolution. In this paper, we study the feasibility of the proposed passive technique for monitoring parameters of material in which the sensors are embedded. We performed a set of experiments where the sensor-to-sensor wireless channel parameters are well-defined using physics-based modeling, and we compared the theoretical and experimentally obtained values. For some material parameters of interest, like humidity or strain, the relationship with the observed wireless channel parameters have to be modeled relying on data-driven approaches. The initial experiments show an observable difference in the sensor-to-sensor channel phase with variation in the applied weights.

Unmanned aerial vehicles (UAVs) constitute a new frontier in remote sensing of ET that bridges the gap between in situmeasurements and remotely sensed observations of plant water use. While a single satellite pixel often comprises a mixture of plant types and bare soil, UAV imagery can resolve fine- (m- to cm-) scale differences in surface temperature without thermal unmixing. Furthermore, they can be used to observe diurnal patterns of plant water use and photosynthesis, providing critical insights into the timing and severity of plant water stress. We highlight a novel approach for estimating ET at leaf- to canopy-scales using thermal infrared (TIR) imagery, structural data, and a suite of environmental sensors mounted on a UAV platform. ET is calculated solely from these UAV-acquired data using a combined atmospheric profile and surface energy balance algorithm. Centimeter-scale leaf position and orientation information derived from Structure-from-Motion (SfM) are integrated with the functional data to constrain available energy, allowing for multi-scale estimation of plant water use within and across canopies. We present UAV-derived ET across diurnal and seasonal time scales for two landscapes, a native California grassland and a riparian oak woodland. Grassland flights were conducted at 90-minute intervals spanning early morning to late afternoon during the 2021 and 2022 growing seasons. Results show good agreement (<20%) with measured ET fluxes from a collocated eddy covariance tower throughout the growing season. Riparian oak canopies were observed monthly and diurnally over the summer of 2021. Ground measurements of surface temperature, stomatal conductance, and soil moisture were collected during each flight. Water-stressed tress at the driest site showed peak ET at midday, decreasing into afternoon, reflecting down-regulation of photosynthesis to preserve hydraulic function. Relative canopy water use and stress across a range of tree sizes will also be discussed using measurements of stem and canopy area and ET for individual tree crowns extracted from the UAV imagery. By collecting comprehensive meteorological data from sensors on the UAV itself, our approach eliminates the need for extensive field data collection and enables characterization of highly spatially and temporally resolved fluxes within and across complex landscapes. This work opens up new avenues to investigate how ecologically important species—and even individual trees—respond to drought and the impacts of these responses on water use, water stress, and the ecological health of critical habitats like riparian forests.