The experiments covered 115 plots sampled at four wheat growth stages, yielding 920 aboveground-biomass observations across the 2024–2025 and 2025–2026 growing seasons. The research was conducted at the Xinxiang Comprehensive Base of the Chinese Academy of Agricultural Sciences in China’s Huang–Huai–Hai Plain. Three experiments were designed to create a broad range of canopy conditions.

The first crossed 13 wheat cultivars with four nitrogen levels, producing 52 treatments; it was intended to generate variation in crop growth and biomass rather than statistically compare the effects of nitrogen. A second experiment combined four irrigation regimes with three nitrogen levels for 36 plots of the cultivar Bainong5819. A third tested three irrigation regimes and three nitrogen levels across 27 plots of Bainong4199.

Field campaigns occurred at jointing, booting, anthesis and grain filling. At each sampling, three replicates of 20 wheat plants were collected per plot, oven-killed and dried to constant weight to determine biomass. Each ground campaign was synchronized with flights by two unmanned aerial platforms.

A DJI Mavic 3 M carried an RGB sensor and four multispectral bands—green, red, red edge and near infrared—while a DJI Matrice 300 RTK equipped with a Zenmuse L2 sensor collected LiDAR point clouds. The multispectral imagery had a ground sampling distance of approximately 0.70 centimetres per pixel. Following noise removal, voxel downsampling, ground classification and registration, the retained LiDAR clouds averaged about 1,200 points per square metre.

The researchers excluded the outer two rows of plants from each plot region of interest to reduce edge effects, mixed pixels and positional uncertainty. Independent checkpoints gave a horizontal root mean square error of approximately 0.02 metres for the multispectral orthomosaics, while registration error between growth stages remained below 0.04 metres. Spectral, structural and temporal information was extracted to represent different aspects of the canopy.

Eight gray-level co-occurrence matrix measures—mean, variance, homogeneity, contrast, dissimilarity, entropy, energy and correlation—were calculated in each of the four spectral bands, giving 32 texture variables. LiDAR measurements supplied canopy-height percentiles, coverage, vertical point distribution and canopy volume. Derived features combined information across growth stages: stage-to-stage differences, relative changes, canopy-volume-multiplied vegetation indices and an area-under-the-trajectory measure linking canopy volume with vegetation-index values.

Variables requiring a preceding observation could not be generated at jointing but were available at subsequent stages. Four regression algorithms—k-nearest neighbor, partial least squares regression, support vector regression and random forest—were compared within a nested cross-validation framework grouped by plot identity. Measurements from the same plot across years and stages were kept in the same group, preventing repeated observations from appearing in both training and testing sets.

Feature construction, feature selection, normalization and hyperparameter optimization were performed only within each outer training fold. Competitive adaptive reweighted sampling selected candidate variables, variance inflation factor analysis removed remaining multicollinearity, and SHAP analysis quantified the contributions of retained features to out-of-fold predictions. This design evaluated the added information from combining data sources while reducing the chance that model performance would be inflated by leakage of repeated field measurements.

Multi-source fusion produced the strongest pooled results. Models combining vegetation indices, texture features and LiDAR-derived structure reached R² values of 0.901 for random forest and 0.912 for k-nearest neighbor, with root mean square errors of 1.951 and 1.836 megagrams per hectare, respectively. Adding derived features increased the random forest R² by 2.44% relative to that basic fusion, while a model using derived features alone performed comparably to basic multi-source fusion.

The complete feature set produced R² values of 0.915 for k-nearest neighbor, 0.875 for partial least squares regression, 0.911 for support vector regression and 0.923 for random forest. Random forest had the narrowest residual distribution around zero and was used for subsequent stage-wise interpretation and plot-scale mapping. The strongest raw correlations did not necessarily become the most influential model inputs.

LiDAR structural measurements had the highest overall correlations with biomass, with coefficients from 0.85 to 0.87 for leading height variables, while several texture features reached absolute correlations of 0.72 to 0.76. In the complete random forest model, however, the derived trajectory measure combining canopy volume and near-infrared information ranked first by mean absolute SHAP value. Canopy coverage ranked second, followed by the relative change in the modified chlorophyll absorption in reflectance index.

Other derived variables—including a canopy-volume and red-band trajectory measure, a change in red-band homogeneity and a change in a canopy-volume multiplied near-infrared feature—had moderate importance. Across the four algorithms, derived features generally contributed more than the basic vegetation-index, texture and LiDAR variables, and canopy coverage remained prominent in each model. Performance changed when the pooled model’s out-of-fold predictions were separated by growth stage.

Its R² values were 0.416 at jointing, 0.555 at booting, 0.378 at anthesis and 0.485 at grain filling. The pooled model tended to overestimate biomass at jointing and booting and underestimate it at anthesis and grain filling. Growth stage by itself was a strong predictor, explaining 85.2% of biomass variation, but remote-sensing features supplied additional information about differences among plots within stages.

Incremental R² values reached 0.479 for random forest, 0.422 for k-nearest neighbor, 0.397 for support vector regression and 0.150 for partial least squares regression. The imagery therefore explained within-stage variation that was not captured by predictions based solely on the mean biomass for each growth stage. Models developed separately for each growth stage performed best at booting.

A support vector regression model using vegetation indices, textures and LiDAR achieved an R² of 0.632, a root mean square error of 1.425 megagrams per hectare and a ratio of performance to deviation of 1.659 there. The best jointing model was a k-nearest neighbor using the same three feature groups, with an R² of 0.426, a root mean square error of 0.575 megagrams per hectare and a ratio of performance to deviation of 1.335. At anthesis and grain filling, the complete-feature random forest performed better among the evaluated models, with R² values of 0.452 and 0.473, respectively.

Their ratios of performance to deviation were 1.418 and 1.413. The dominant predictors also shifted with development: the modified triangular vegetation index led at jointing, green-band contrast at booting, near-infrared reflectance and a relative canopy-volume multiplied modified triangular index at anthesis, and a red-band mean-texture change at grain filling. Applying the final random forest at plot scale mapped increasing biomass from jointing through grain filling and showed differences among cultivars, nitrogen and irrigation treatments, with generally higher estimates under greater water and nitrogen availability.

Residual maps contained both positive and negative errors, although most plots had relatively small deviations. The findings nevertheless require a bounded interpretation because the experiments came from one site and two seasons. Bidirectional validation between years produced an R² of 0.738, a root mean square error of 3.189 megagrams per hectare and a ratio of performance to deviation of 1.954, below the study’s threshold of 2.0 for reliable prediction.

Models trained on 2025 data tended to overestimate biomass when applied to 2026, whereas models trained on 2026 data tended to underestimate it when applied to 2025. The temporal design also limits how the pooled model should be used. Because its trajectory incorporates observations from later growth stages, it is better characterized as a retrospective biomass-estimation model than as an early-season forecasting system; stage-specific models used only observations available by the corresponding stage.

The experiments were also confined to one location in the Huang–Huai–Hai Plain, a particular UAV sensor configuration and a specific feature-extraction workflow. Vegetation masking depends on the separability of canopy and soil, while texture features are sensitive to window size, spatial resolution and image preprocessing. Validation across other climatic regions, wheat cultivars, management practices, sensors, illumination conditions and modeling strategies is therefore needed before the fused approach can be considered broadly applicable.

Within those limits, the study shows that combining spectral, spatial, structural and multi-stage information can improve winter wheat biomass estimation while revealing that the relative value of each predictor changes as the canopy develops.

Subject of Research: Multi-source UAV feature fusion for estimating winter wheat aboveground biomass

Article Title: Integration of multi-source UAV remote sensing features and derived features for winter wheat aboveground biomass estimation

Article References: Wang, G., Yao, S., Xie, Y., Qin, A., Chen, X., Zhang, Y., & Jiang, L. (2026). Integration of multi-source UAV remote sensing features and derived features for winter wheat aboveground biomass estimation. Smart Agricultural Technology, 15, Article 102656. https://doi.org/10.1016/j.atech.2026.102656

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102656

Keywords: winter wheat, aboveground biomass, unmanned aerial vehicles, multispectral imaging, LiDAR, machine learning, random forest, feature fusion, crop monitoring, precision agriculture, Integration, multi-source

The research, published in BMC Plant Biology, was conducted by Guanjie Wang of Henan Agricultural University and colleagues, working in permanent forest monitoring plots established in the Baotianman National Nature Reserve. The team analyzed a total of six hectares of forest, a substantial area for this kind of work, in a region where the cold-loving coniferous and broadleaved forests of the temperate zone give way to the species-rich evergreen and deciduous forests of the subtropics. This transition zone is a natural laboratory for studying how ecological rules change with climate, because both ectomycorrhizal and arbuscular mycorrhizal trees coexist there in varying proportions, allowing researchers to ask what happens to a forest as one fungal guild replaces the other.

The central question was deceptively simple: as the relative abundance of ectomycorrhizal trees rises, what happens to the forest’s aboveground biomass, the total mass of living wood, bark, and branches above the soil? Aboveground biomass matters enormously because it is the most direct measure of the carbon a forest pulls from the atmosphere and locks away in its trunks. It also reflects the forest’s productivity, its successional maturity, and its vulnerability to disturbances such as storms and droughts. But biomass does not exist in isolation. It is shaped by how many species share the canopy, by the range of tree sizes present, and by the competitive and facilitative interactions among neighbors. Disentangling these threads requires both careful measurement and sophisticated statistics.

To do so, the researchers deployed a three-tiered analytical strategy. First, they used bivariate regressions to examine simple pairwise relationships between variables. Second, they built Gaussian linear mixed-effects models, which allow scientists to estimate the influence of fixed predictors, such as mycorrhizal abundance, while accounting for random variation among plots. Third, and most revealing, they constructed a mixed-effects piecewise structural equation model, a framework that maps networks of hypothesized cause-and-effect pathways simultaneously rather than testing each link in isolation. This approach matters because ecological variables are deeply intertwined: a fungus-driven shift in species composition could influence biomass directly, or indirectly by changing how many species coexist or how variable tree sizes are, and only a path analysis can separate these routes.

The headline finding is striking. In the mixed-effects model, the relative abundance of large ectomycorrhizal trees accounted for 42.66 percent of the explained fixed-effect variation in aboveground biomass, while species richness accounted for 31.46 percent. In other words, the identity of the fungal partnership a tree forms predicted more of the variation in forest carbon storage than the sheer number of species did. The piecewise structural equation model sharpened the picture further: large ectomycorrhizal tree relative abundance showed the strongest direct positive association with biomass of any variable tested, with a standardized path coefficient of 0.570 and a p-value below 0.001, indicating an extremely unlikely chance occurrence.

Yet the story is not one of fungal dominance producing unambiguously richer forests. Species richness, although its direct link to biomass was weaker, accumulated influence through indirect routes, and its total effect on biomass, 0.466, slightly exceeded the total effect of large ectomycorrhizal tree abundance, which was 0.425. This means that while ectomycorrhizal trees are the heavyweight champions of direct carbon accumulation, diversity contributes to biomass through a web of secondary pathways, perhaps by improving resource partitioning, filling canopy gaps, or buffering the stand against environmental fluctuations. Both dimensions of forest structure matter, but they matter in different ways and through different mechanisms.

The negative side of the ledger is equally important. As large ectomycorrhizal tree abundance rose, the relative abundance of arbuscular mycorrhizal trees fell, with a path coefficient of −0.509, and tree-diameter variation collapsed even more dramatically, with a coefficient of −0.803. Diameter variation is a measure of how unevenly sized the trees in a stand are; high variation signals a forest with towering old giants alongside young saplings, a hallmark of complex, multi-aged structure. Its sharp decline under ectomycorrhizal dominance suggests that these forests become more uniformly sized, perhaps because ectomycorrhizal trees, with their ability to tap nitrogen directly through fungal networks, outcompete neighbors and drive the stand toward a narrower range of growth trajectories.

Crucially, when the researchers summed up all the indirect effects of ectomycorrhizal abundance on biomass, the total was negative, at −0.145. The suppression of species richness and the flattening of tree-size variation together worked against carbon storage, even as the direct pathway pushed it upward. The net result is that the positive relationship between ectomycorrhizal dominance and biomass is driven almost entirely by the direct route, the sheer capacity of these trees and their fungi to build wood, rather than by any indirect benefit flowing through structural complexity. This is a subtle but important correction to simpler narratives that equate fungal type with forest quality.

Why would ectomycorrhizal trees be such efficient carbon accumulators? The answer likely lies in fungal physiology. Ectomycorrhizal fungi can access organic forms of nitrogen, breaking down complex soil compounds that arbuscular mycorrhizal fungi largely cannot exploit. In nitrogen-limited temperate forests, this capability gives ectomycorrhizal trees a decisive edge, allowing them to sustain high growth rates on nutrient-poor soils. Many ectomycorrhizal species, such as pines and oaks, also tend toward denser, longer-lived wood, which stores more carbon per unit volume. The trade-off is that the same traits that make these trees competitive, including dense canopies and litter that alters soil chemistry, can exclude arbuscular mycorrhizal species and reduce the recruitment opportunities that maintain a diverse, uneven-aged forest.

The implications reach well beyond the mountains of central China. As climate change shifts the ranges of tree species, and as nitrogen deposition alters soil nutrient availability, the global balance between ectomycorrhizal and arbuscular mycorrhizal forests is expected to move. Models that forecast future carbon storage often treat forests as interchangeable pools of biomass, but this study suggests that the fungal identity of the dominant trees is a first-order control on both carbon and biodiversity outcomes. A future forest dominated by ectomycorrhizal species may store more carbon per hectare while supporting fewer species and a simpler size structure, a trade-off that conservation planners and climate policymakers will need to weigh together. The hidden marketplace beneath our feet, it turns out, helps set the terms for everything above it.

Subject of Research: Associations between ectomycorrhizal tree abundance, species diversity, tree-size variation, and aboveground biomass in temperate–subtropical transition forests

Article Title: Ectomycorrhizal tree relative abundance is associated with higher aboveground biomass but reduced species richness and tree-size variation in temperate–subtropical transition forests

Article References: Wang, G., Shi, K., Zhang, T., Huang, Q., Zhao, B., Chen, Z., Qian, J., Kong, D., Yang, L., Yu, C., Jin, X., & Wang, T. (2026). Ectomycorrhizal tree relative abundance is associated with higher aboveground biomass but reduced species richness and tree-size variation in temperate–subtropical transition forests. BMC Plant Biology. https://doi.org/10.1186/s12870-026-09910-y

Image Credits: AI Generated

DOI: 10.1186/s12870-026-09910-y

Keywords: ectomycorrhiza, arbuscular mycorrhiza, aboveground biomass, species richness, tree-size variation, forest ecology, mycorrhizal fungi, temperate–subtropical transition, carbon storage, structural equation modeling, Baotianman, plant ecology

The central concept at work is the efficiency-safety trade-off, a long-standing principle in plant physiology. Plants that move water quickly through their xylem, the vascular tissue that carries water from roots to leaves, gain an advantage: high hydraulic conductivity supports high stomatal conductance, which in turn allows photosynthesis to run at full throttle. But fast, wide, or highly conductive xylem comes at a cost. Under drought, the negative pressures that pull water upward can exceed critical thresholds, causing air bubbles to nucleate and block the conduits in a process called cavitation or embolism. Plants with safer hydraulic architecture, by contrast, feature narrower conduits, denser wood, and stronger embolism resistance, allowing them to keep their vascular systems intact as soils dry, but at the price of slower water delivery and reduced photosynthetic capacity when water is plentiful.

What makes the new study remarkable is its scale and its focus on herbaceous plants, the dominant life form of dryland ecosystems that has often been overshadowed in hydraulic research by trees and shrubs. By sampling herbaceous communities systematically across the Mongolian Plateau, the researchers were able to aggregate individual species traits up to the community level and ask whether the efficiency-safety trade-off operates as an organizing axis for whole ecosystems. The answer, they report, is a clear yes: a community-wide trade-off between hydraulic efficiency and safety fundamentally structures ecosystem function across the region.

Crucially, the team found that aridity acts as the master variable steering this trade-off. As conditions become drier along the precipitation gradient, the composition of herbaceous communities shifts strategically from efficiency-oriented traits toward safety-oriented traits. In wetter parts of the plateau, communities are dominated by species whose leaves can move water rapidly, fueling fast growth and high carbon uptake. In the driest reaches, the flora tilts toward species engineered to survive hydraulic failure, sacrificing peak productivity for the assurance that their water transport pathways will not collapse during the frequent and severe droughts that characterize the region.

This strategic shift does not merely describe which plants live where; it propagates directly into ecosystem-level properties that matter for pastoral livelihoods, carbon cycling, and land management. The researchers quantified three core functions: aboveground biomass, water-use efficiency, and ecosystem stability. Hydraulic efficiency, they found, enhances aboveground biomass by supporting greater photosynthetic carbon assimilation, but it simultaneously reduces water-use efficiency and undermines ecosystem stability. Safety traits produce exactly the opposite pattern. Communities built for hydraulic security fix less carbon per unit of leaf area and accumulate less biomass, but they use the water they do take up more sparingly and their productivity fluctuates less through time, a hallmark of stability in variable environments.

The quantitative strength of the framework is one of its most striking features. By combining community-aggregated hydraulic traits into a predictive model, the authors explain 43 percent of the variance in aboveground biomass, 40 percent of the variance in water-use efficiency, and fully 60 percent of the variance in ecosystem stability across the 306 study sites. In ecology, where field data are notoriously noisy and single variables rarely explain more than a modest fraction of observed variation, these are substantial effect sizes. They suggest that a relatively simple axis, the position of a community along the efficiency-safety spectrum, captures much of the functional character of these grasslands.

The implications extend well beyond the Mongolian Plateau. Drylands cover roughly 40 percent of the terrestrial land surface and support hundreds of millions of people, many of whom depend on herbaceous vegetation for grazing livestock. As climate change intensifies aridity across much of Central Asia and other dryland regions, the trait composition of these communities is expected to shift further toward the safety end of the spectrum. The study’s findings imply that such a shift would likely come with declining biomass production and altered water-use patterns, even as stability characteristics change in complex ways. Understanding this coupling between plant hydraulics and ecosystem function provides a mechanistic basis for anticipating how grassland productivity and resilience will respond to ongoing drying.

The work also carries direct lessons for resilience-focused ecosystem management. Grassland restoration and grazing management decisions are often guided by aboveground indicators such as biomass or species cover, which can be misleading proxies for underlying functional capacity. The trait-based framework developed here suggests that monitoring or targeting hydraulic trait composition could offer a more predictive handle on how a given grassland will behave under drought. Communities dominated by efficiency-oriented species may deliver high forage production in favorable years but risk instability when drought strikes, whereas safety-oriented communities trade some productivity for steadier performance. Managers seeking to balance livestock production against long-term ecosystem resilience could, in principle, use this trade-off as a diagnostic lens.

Methodologically, the study exemplifies a growing trend in ecosystem ecology: moving from single-species measurements to community-wide, trait-based approaches that link organismal physiology to landscape-scale function. Rather than treating drought response as an emergent mystery, the researchers traced it through an explicit causal chain, from the hydraulic architecture of leaves, through stomatal behavior and carbon assimilation, to biomass accumulation, water-use efficiency, and temporal stability of the ecosystem. Each link in that chain was quantified across hundreds of field sites, giving the framework empirical grounding that laboratory studies alone cannot provide.

The research, published open access in Communications Earth & Environment with contributions from ecologists at Inner Mongolia University, Lanzhou University, the Inner Mongolia Institute of Meteorological Sciences, Inner Mongolia Normal University, and the Institute of Water Resources for Pastoral Areas, arrives at a moment when the world’s drylands face mounting pressure from warming temperatures, shifting precipitation regimes, and intensifying land use. By demonstrating that a single, measurable trade-off in leaf hydraulics can explain large fractions of variation in the functions that sustain grassland ecosystems, the study offers both a warning and a tool: a warning that increasing aridity will push these systems toward safer but less productive configurations, and a tool in the form of a quantified, predictive framework that can help scientists and managers anticipate and perhaps buffer those changes before they unfold.

Subject of Research: Leaf hydraulic trait trade-offs and their effects on ecosystem function in Mongolian Plateau grasslands

Article Title: Herbaceous leaf hydraulic trade-offs shape ecosystem function across the Mongolian Plateau

Article References: Guo, X., Wang, X., Jia, C., Liu, T., Li, Z., Liu, H., Wang, L., Wang, Y., Dong, L., Liang, C., & Zhang, J. (2026). Herbaceous leaf hydraulic trade-offs shape ecosystem function across the Mongolian Plateau. Communications Earth & Environment. https://doi.org/10.1038/s43247-026-04121-7

Image Credits: AI Generated

DOI: 10.1038/s43247-026-04121-7

Keywords: plant hydraulics, leaf traits, efficiency-safety trade-off, Mongolian Plateau, grassland ecology, drought, aridity gradient, ecosystem stability, water-use efficiency, aboveground biomass, functional traits, drylands

Aboveground biomass density, often abbreviated AGBD, is the dry mass of living vegetation per unit area held in trunks, branches and foliage. It is the single most important number for converting a forest map into a carbon budget, because roughly half of dry woody biomass is carbon by mass. Yet measuring it directly is brutal work: field crews must identify trees in sample plots, measure trunk diameters at breast height, estimate or measure heights, apply allometric equations that convert those dimensions into mass, and then extrapolate from a handful of plots to entire landscapes. In the rugged, fragmented terrain of the Upper Blue Nile basin, where forest patches interleave with farmland on steep slopes, that extrapolation has historically been the weakest link. The new study attacks exactly that gap by using field plots as ground truth and letting machine learning do the scaling.

The technological heart of the study is a deliberate marriage of sensors that see the forest in fundamentally different ways. The researchers drew on GEDI L2A, a product from NASA’s Global Ecosystem Dynamics Investigation, a full-waveform light detection and ranging instrument mounted on the exterior of the International Space Station. GEDI fires laser pulses at the canopy and records the returning energy, yielding precise vertical profiles of the forest from which metrics such as RH98, the height below which 98 percent of returned energy lies, can be extracted. Because the instrument samples the ground in discrete footprints along orbital tracks rather than imaging every square meter, its observations must be interpolated to produce continuous maps, and that is where the other sensors come in.

To fill the gaps between GEDI footprints, the team turned to Planet-NICFI imagery, a high-resolution basemap program covering the tropics, alongside the European Sentinel-2 optical satellites and the Japanese ALOS PALSAR-2 radar mission. Optical data captures spectral signatures of green vegetation, including indices sensitive to canopy density, but it saturates in dense forests and is hostage to the region’s persistent cloud cover. Radar, by contrast, penetrates clouds and responds to the water content and structure of woody biomass, but its longer wavelengths also suffer from saturation in tall, humid forests. The researchers’ central premise was that no single sensor could carry the full signal, and their results bore this out: combining LiDAR, radar and optical variables consistently outperformed any sensor alone, alleviating the saturation effects that plague single-sensor approaches.

All of this data crunching happened on Google Earth Engine, the cloud-based planetary computing platform that has quietly transformed environmental science by letting researchers process petabyte-scale archives without downloading a single image. The modeling engine was Random Forest, an ensemble machine learning method that builds hundreds of decision trees on random subsets of the training data and averages their predictions. Random Forest has become the workhorse of satellite-based biomass mapping because it handles nonlinear relationships, resists overfitting, and tolerates the mixed predictor types that multi-sensor fusion produces. In the first stage of the analysis, the team trained a Random Forest model to predict forest canopy height, using field measurements as the reference and GEDI RH98 observations fused with Planet-NICFI imagery as predictors.

The canopy height model was the study’s first headline result. By integrating GEDI L2A RH98 observations with Planet-NICFI imagery, the researchers achieved a modeling performance of R-squared equal to 0.89 with a root mean square error of 1.91 meters, and from that model they produced a forest canopy height product for the year 2023 at an unprecedented 5-meter spatial resolution. That resolution matters enormously in a landscape like the Upper Blue Nile basin, where forest fragments are often smaller than the pixel of coarser global products. A 5-meter map can distinguish a remnant church forest from the surrounding cropland, capture edges where biomass degrades fastest, and support management decisions at the scale at which Ethiopian foresters actually operate.

With canopy height established, the study moved to its main event: predicting biomass density itself. The researchers used field-derived AGBD as reference data and trained Random Forest models on GEDI L2A, ALOS PALSAR-2 and Sentinel-2 variables, first with each sensor independently and then with the multi-sensor integration. The integrated model was the clear winner, yielding an R-squared of 0.972 and a root mean square error of 18.27 megagrams per hectare. From this model the team generated an AGBD product at 10-meter spatial resolution for 2023, incorporating multisource remote sensing variables together with topographic information such as elevation and slope, which strongly influence both forest structure and species composition in mountainous terrain.

When the researchers compared their satellite-derived biomass map against the field measurements, the spatial distribution patterns matched closely, with the GEDI-derived canopy height product adding enhanced detail and accuracy in dense, high-canopy forests, precisely the environments where optical and radar data tend to falter. This validation step is what elevates the study from a technical exercise to a usable product. A biomass map is only as good as its ground truth, and the agreement between predicted and measured values across the basin suggests the method captures real ecological structure rather than statistical artifacts. For a region where forest loss and degradation drive both carbon emissions and watershed decline, a trustworthy, current, wall-to-wall map is a genuinely scarce commodity.

The implications ripple outward well beyond Ethiopia. Under international frameworks such as REDD+, developing countries can receive performance-based payments for verified emissions reductions from avoided deforestation, but verification demands exactly the kind of accurate, repeatable biomass accounting this study delivers. The same data streams feed national greenhouse gas inventories, forest management plans, and the growing voluntary carbon market, where inflated or uncertain baselines have repeatedly undermined credibility. Because the workflow runs on Google Earth Engine using freely available Sentinel data and open products like GEDI, the approach is replicable by national agencies and universities in other data-scarce tropical regions without prohibitive computing costs, lowering the barrier that has historically kept high-quality carbon mapping confined to wealthy institutions.

The study also offers a glimpse of where forest monitoring is heading. Spaceborne LiDAR from GEDI, radar backscatter from missions like ALOS PALSAR-2 and the upcoming generation of bistatic radar satellites, and ever-improving optical constellations are converging into a multi-sensor ecosystem in which the weaknesses of each instrument are patched by the strengths of the others. Machine learning provides the glue, and cloud platforms provide the scale. Kerebeh and Taddese frame their findings as evidence-based support for forest policy, climate change mitigation and sustainable ecosystem management, and the numbers back the framing: sub-2-meter height accuracy and a near-0.97 correlation for biomass in a fragmented tropical highland landscape would have been unthinkable a decade ago. As the final version of the open-access paper enters the literature, its 5-meter canopy height and 10-meter biomass maps of the Upper Blue Nile stand as proof that the world’s most important forests can now be weighed from orbit, one fused pixel at a time.

Subject of Research: Multi-source remote sensing estimation of forest aboveground biomass in the Upper Blue Nile basin, northwestern Ethiopia

Article Title: Assessment of aboveground biomass using integrated multi-source remote sensing and field data in forests of the Upper Blue Nile basin Northwestern Ethiopia

Article References: Kerebeh, H., & Taddese, H. (2026). Assessment of aboveground biomass using integrated multi-source remote sensing and field data in forests of the Upper Blue Nile basin Northwestern Ethiopia. Discover Sustainability. https://doi.org/10.1007/s43621-026-04933-9

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04933-9

Keywords: aboveground biomass, remote sensing, GEDI, LiDAR, Sentinel-2, ALOS PALSAR-2, Random Forest, Google Earth Engine, canopy height, Ethiopia, forest carbon, data fusion

The stakes of this measurement problem are enormous. Mangroves are among the most carbon-rich forests in the tropics, storing vast quantities of carbon in their trees and, even more importantly, in the waterlogged sediments beneath them. When these forests are cleared or degraded, that carbon can be released rapidly, turning a powerful sink into a source of greenhouse gases. Indonesia holds some of the world’s most extensive mangrove estates, which makes the country a linchpin in global blue carbon accounting, yet ground surveys of mangrove biomass are laborious, expensive, and often dangerous in the mud, tides, and mosquitoes of a tropical estuary. A reliable remote sensing shortcut has long been the holy grail of coastal carbon science.

The logic behind the new approach rests on a decades-old insight from vegetation remote sensing. The NDVI exploits the fact that healthy, chlorophyll-rich leaves absorb red light strongly while reflecting near-infrared light, so the ratio between the two bands serves as a proxy for the amount and vigor of green vegetation. The Sentinel-2 mission, with its twin satellites delivering multispectral imagery at ten-meter resolution every few days, offers an ideal platform for this kind of work: the data are free, frequent, and sharp enough to resolve the patchwork of mangrove stands, aquaculture ponds, and open water that characterizes an Indonesian bay. What the Pangpang Bay study adds is a direct, locally calibrated link between that spectral signal and the actual kilograms of carbon locked in tree trunks, branches, and leaves.

To forge that link, the researchers established thirteen sample plots in the field, measured the trees within them, and converted those structural measurements into aboveground biomass using established mangrove allometric equations, the standard formulas that translate trunk diameter and tree height into woody mass. They then extracted NDVI values from Sentinel-2 imagery corresponding to each plot and fitted an empirical regression between the two. The resulting model, in which estimated biomass equals negative 871.15 plus 4562.73 multiplied by NDVI, produced a coefficient of determination of 0.649 with a p-value below 0.001, meaning the relationship was statistically strong and that the vegetation index accounted for approximately two-thirds of the observed variability in biomass. For a single, freely available spectral index, that level of explanatory power is respectable, particularly in the notoriously heterogeneous environment of a tropical mangrove bay.

With the model in hand, the team turned it loose on the archive. They applied the regression to Sentinel-2 scenes from 2019, 2022, and 2025, producing multitemporal maps of biomass and carbon across the bay and revealing a story that unfolds in space as well as time. The temporal signal was sobering. Between 2019 and 2022, mean NDVI across the mangrove area fell by 35.1 percent, and the model translated that spectral decline into a marked drop in aboveground biomass and carbon stocks. By 2025 the forests had shown only limited signs of recovery, suggesting that whatever disturbance drove the loss, whether conversion, cutting, storm damage, or hydrological change, had left a lasting imprint on the canopy and that regrowth of mature, carbon-dense trees is a slow business measured in decades rather than years.

The spatial patterns were equally telling. Areas dominated by mature Rhizophora and Bruguiera stands, the classic old-growth mangrove genera with their stilted roots and dense, heavy wood, consistently registered higher biomass and carbon stocks than stands dominated by Sonneratia, a faster-growing, softer-wooded pioneer genus that often colonizes newly opened mudflats. This distinction matters for carbon accounting because it means that two patches of mangrove that look superficially similar on a map can differ substantially in the carbon they hold, and that species composition, stand age, and successional stage all leave fingerprints in the satellite record. A bay-wide carbon estimate that ignores these differences risks being badly wrong, which is precisely why a locally calibrated NDVI-to-biomass relationship, rather than a generic global model, is such a valuable asset.

The authors are careful to frame the work as an exploratory local assessment, and the limitations are worth taking seriously. Thirteen field plots is a modest sample for calibrating a regression, and the scatter around the fitted line means that roughly a third of the biomass variability remains unexplained by NDVI alone. Canopy greenness, after all, is not a perfect stand-in for wood volume: a dense stand of young Sonneratia can be as green as a sparse stand of ancient Rhizophora while holding far less carbon. Saturation effects also loom, since NDVI tends to plateau in very dense vegetation, potentially underestimating the most carbon-rich stands. The researchers position the approach accordingly, not as a replacement for field inventories but as a practical, cost-effective screening and monitoring tool for coastal environments where the resources for repeated ground surveys simply do not exist.

That framing points directly at one of the most consequential acronyms in modern climate policy: MRV, or monitoring, reporting, and verification. Carbon markets, national greenhouse gas inventories, and international climate finance mechanisms all demand credible, repeatable, and affordable measurements of carbon stocks, and mangrove ecosystems have historically been among the hardest to deliver. A workflow that combines a modest number of field plots with freely available Sentinel-2 imagery offers exactly that: a method that a provincial forestry office or a conservation NGO in Banyuwangi could operate without a supercomputer or a six-figure budget. The study’s authors explicitly highlight applications for blue carbon assessment, ecosystem restoration planning, and MRV programs in data-limited coastal settings, and the multitemporal demonstration in Pangpang Bay shows the method doing real work, detecting a major biomass decline and a sluggish recovery that would otherwise have gone unquantified.

The broader context makes the findings resonate well beyond one bay in East Java. Indonesia has set ambitious targets for mangrove rehabilitation as part of its low-carbon development strategy, and recent research suggests that conserving and restoring mangroves and peat swamps could mitigate a substantial share of Southeast Asia’s land-use carbon emissions. Yet restoration is expensive, and global analyses of mangrove restoration costs underscore that every hectare planted must be justified and then monitored for decades. Tools like the one demonstrated here provide the accounting backbone for that effort, allowing managers to identify which existing stands are the most carbon-rich and therefore most worth protecting, and to verify whether young restoration sites are actually accumulating biomass over time.

There is also a quiet democratization at work in this study. The Sentinel-2 data are free, the NDVI calculation requires nothing more exotic than arithmetic on two spectral bands, and the calibration demands only a handful of well-executed field plots. In an era when high-end biomass mapping increasingly relies on airborne laser scanning, spaceborne radar, and machine learning pipelines that demand both money and expertise, the demonstration that a humble vegetation index can capture two-thirds of the biomass variability in a tropical mangrove bay is a reminder that sometimes the simplest tool, applied locally and honestly, is the one that gets used. As Pangpang Bay’s mangroves slowly claw back the carbon they lost between 2019 and 2022, the satellites above will be watching, and now, thanks to this study, scientists will know what they are looking at.

Subject of Research: Estimating mangrove aboveground biomass and blue carbon stocks in Pangpang Bay, Indonesia, using Sentinel-2 NDVI and field calibration

Article Title: A local exploratory assessment of mangrove aboveground carbon stocks based on Sentinel-2 NDVI in Pangpang Bay, Indonesia

Article References: Hidayat, E. F., Simarmata, N., Cansery, A., Julian, M. M., & Ansari, K. (2026). A local exploratory assessment of mangrove aboveground carbon stocks based on Sentinel-2 NDVI in Pangpang Bay, Indonesia. Environmental Monitoring and Assessment, 198(11), Article 1153. https://doi.org/10.1007/s10661-026-16008-3

Image Credits: AI Generated

DOI: 10.1007/s10661-026-16008-3

Keywords: mangroves, blue carbon, Sentinel-2, NDVI, aboveground biomass, carbon stocks, remote sensing, Indonesia, Pangpang Bay, coastal monitoring, MRV, ecosystem restoration

In a study published in Earth Science Informatics, Alexander Cotrina-Sanchez and Michele Torresani of the Free University of Bozen-Bolzano, together with Leonel Corado of the Universidade de Évora, present GEDIMetrics, an open-source plugin for the widely used geographic information system QGIS. The tool integrates the entire GEDI processing chain, from discovering and downloading raw data files to filtering, clipping, and fusing four complementary data products into a single analysis-ready dataset, all through a graphical interface that requires no programming. The plugin is freely available through the official QGIS Plugin Repository and on GitHub under the GNU General Public License.

The technical challenge the plugin addresses is deceptively simple to describe but notoriously tedious in practice. GEDI’s laser footprints, roughly 25 meters in diameter and spaced about 60 meters apart along the satellite’s track, capture the vertical distribution of returned laser energy. From these waveforms, NASA derives several distinct products: Level 2A provides canopy height and ground elevation metrics, Level 2B characterizes vertical canopy structure including canopy cover, plant area index, and foliage height diversity, Level 4A delivers aboveground biomass density estimates from globally calibrated models, and Level 4C offers the Waveform Structural Complexity Index, a machine-learning-derived measure of three-dimensional canopy complexity. Each product, however, is distributed as independent HDF5 files organized by laser beam, and the products must be aligned using a unique identifier called the shot number before any joint analysis becomes possible.

Existing tools each solved only fragments of this puzzle. NASA’s EarthData Search portal handles discovery and download but performs no subsetting, filtering, or cross-product integration. Programming libraries such as rGEDI and pyGEDI offer flexible scripted processing but demand coding expertise. Command-line frameworks like GEDI-Pipeline introduced automation, yet none provided a graphical workflow suitable for non-programmers, and none supported automated multi-product merging. GEDIMetrics closes that gap with a five-stage pipeline: granule discovery through NASA’s Common Metadata Repository API, authenticated download from EarthData, variable extraction with quality filtering, spatial subsetting to a user-defined region, and footprint-level merging of the selected products, followed by export to standard geospatial formats such as GeoPackage and GeoParquet.

The engineering behind the plugin reflects careful attention to both performance and reproducibility. Spatial filtering proceeds in two stages, beginning with a fast bounding-box pre-filter and followed by precise point-in-polygon clipping using Shapely geometric operations, an approach that substantially reduces processing time for granules covering far more area than the target region. Because data availability can differ across products due to variations in temporal coverage, algorithm convergence, or quality criteria, the merging stage uses Level 2A as the base product and offers configurable join strategies: an inner join retains only footprints present in all selected products, guaranteeing cross-product consistency, while an outer join preserves maximum spatial coverage at the cost of null values. Output files embed metadata recording the full processing configuration, including selected products, temporal range, region geometry, quality thresholds, and plugin version, ensuring that every analysis can be reproduced exactly.

Quality filtering, a critical step for GEDI data, is handled through an intuitive interface. Users can set minimum thresholds for product-specific quality flags, exclude footprints degraded by cloud or atmospheric effects, restrict analysis to land surfaces, and apply a global minimum beam sensitivity threshold, defaulting to 0.90 following criteria established for broad-scale applications. The sensitivity threshold can be raised for particular ecosystems, for example to 0.96 for temperate forests or higher for dense tropical canopies where the laser signal struggles to penetrate to the ground. A logging panel reports the number of granules identified and the estimated data volume before download begins, allowing users to adjust their parameters before committing to lengthy retrievals.

To demonstrate the plugin’s capabilities, the team tested GEDIMetrics across two contrasting forest ecosystems of roughly 100 square kilometers each: a temperate mixed forest in northeastern France and a humid tropical rainforest in the Madre de Dios region of southeastern Peru, an area partially overlapping the Tambopata National Reserve where intact forest faces mounting pressure from illegal gold mining. Using data from March 2019 to April 2023 with default quality settings, the temperate site yielded 96 granules and 5,939 forested footprints, while the tropical site produced only 23 granules and 1,712 footprints, a disparity that reflects the International Space Station’s orbital geometry, which delivers more frequent coverage at mid-latitudes than near the equator.

The case study also revealed striking asymmetries in product-level data availability that cannot be inferred from footprint counts alone. Retention of the Level 2B canopy structure product was moderate at the temperate site, at 60.8 percent of forested footprints, but nearly absent at the tropical site, at just 0.9 percent. The biomass product showed the inverse pattern, with 7.2 percent retention in France versus 87.2 percent in Peru. Despite these differences, the merged datasets painted a coherent ecological picture: the tropical forest exhibited taller canopies, with a 98th-percentile height of 24.05 meters compared to 18.86 meters at the temperate site, higher aboveground biomass density of 157.49 versus 139.51 megagrams per hectare, and greater structural complexity. Most strikingly, footprints inside the Tambopata National Reserve showed consistently taller canopies, 38 percent higher biomass density, and higher structural complexity than those in the surrounding buffer zone, a quantifiable signature of the degradation pressure from gold mining at the reserve’s edges.

At the temperate site, the plugin’s integrated dataset captured seasonal phenology with unexpected clarity. Canopy cover and plant area index rose sharply during the April-to-October vegetative period, consistent with the flush of deciduous foliage, while canopy height remained essentially stable across seasons, confirming that height metrics are largely insensitive to phenological state. The most dramatic seasonal signal appeared in the median height percentile, which jumped from 1.27 meters in the leafless months to 6.36 meters during the growing season, reflecting the increased interception of laser energy by mid-canopy leaves. This ability to distinguish genuine structural change from seasonal foliage dynamics has direct implications for disturbance monitoring and carbon accounting.

The researchers are candid about current limitations. Processing is single-machine and sequential, which can make large regions slow to analyze, and the plugin does not yet correct GEDI’s footprint geolocation uncertainty of roughly 10 meters, though its modular design allows integration with external correction tools. Future development priorities include polygon-level summary statistics, diagnostics quantifying footprint loss at each filtering stage, in-plugin visualization of vertical foliage profiles, and a hybrid local-cloud workflow leveraging NASA’s evolving cloud-native archives. The team also envisions extending the plugin to ICESat-2 canopy height data, which would push coverage beyond GEDI’s latitudinal limits into boreal and high-latitude forests. For now, GEDIMetrics stands as a quietly transformative piece of software: by collapsing a multi-step, script-heavy workflow into a transparent graphical process inside the world’s most widely used open-source GIS, it puts the vertical architecture of the world’s forests within reach of anyone with a computer, a QGIS installation, and a free NASA EarthData account.

Subject of Research: An open-source QGIS plugin for accessing, filtering, and fusing NASA GEDI spaceborne LiDAR forest structure data products

Article Title: GEDIMetrics: a QGIS plugin for accessing and integrating multi-product GEDI spaceborne LiDAR data

Article References: Cotrina-Sanchez, A., Torresani, M., & Corado, L. (2026). GEDIMetrics: a QGIS plugin for accessing and integrating multi-product GEDI spaceborne LiDAR data. Earth Science Informatics, 19(11), Article 205. https://doi.org/10.1007/s12145-026-02254-z

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02254-z

Keywords: GEDI, spaceborne LiDAR, QGIS, forest structure, aboveground biomass, canopy height, remote sensing, open-source software, geospatial data, NASA EarthData, tropical forest, forest carbon

The research, led by Gao-Lin Wu of Shanxi Agricultural University and Northwest A&F University, together with Jesús Rodrigo-Comino of the University of Granada, Zhen Cheng of Lanzhou University, and Xiaogang Wu, was conducted as an in-situ field trial in arid sandy terrain. Rather than testing amendments in pots or laboratory columns, the team worked directly in the degraded landscape, measuring how incorporated residues altered soil water infiltration, evaporation, moisture retention, vegetation coverage, and aboveground biomass under real field conditions.

The headline results are striking. Compared with untouched bare sandy land, plots amended with crushed ryegrass (Lolium perenne L.) residues showed maximum increases of 130.8 percent in vegetation coverage and 207.4 percent in aboveground biomass. Alfalfa (Medicago sativa L.) residues produced maximum gains of 76.9 percent in coverage and 216.4 percent in biomass. In practical terms, the treated sand went from a surface where plants barely survive to one where natural vegetation could establish and accumulate more than three times the standing plant material of controls in the best cases.

The mechanism behind these gains lies in the physics of water movement through coarse soil. Sandy soils are dominated by large macropores, so rainwater races downward past the root zone before plants can capture it—a process the researchers describe through reduced infiltration rates. Incorporating crushed residues into the topsoil partially clogged and restructured these pore networks. Stable infiltration rates, the steady speed at which water eventually percolates, dropped moderately, by 1.6 to 4.8 percent in alfalfa-treated plots and 3.0 to 12.3 percent in ryegrass-treated plots. That modest slowing is enough to keep precious moisture within reach of germinating seeds and shallow roots for longer.

Equally important is what happened to the water that stayed. Field capacity in the top 15 centimeters of soil—the amount of water the soil can hold against gravity—increased slightly but consistently, by 1.8 to 8.6 percent across residue treatments. In a landscape where every millimeter of plant-available water counts, even a few percentage points of extra storage in the rooting zone can determine whether a seedling crosses the threshold from establishment to survival. The residues effectively act as a sponge woven through the sand, buying time between rainfall pulses and plant uptake.

The study also identified evaporation as a critical lever. Bare sandy surfaces lose vast quantities of water directly to the atmosphere, particularly after small rain events that wet only the top few centimeters. The crushed residue layer reduced soil evaporation, keeping moisture locked in the shallow subsurface where it is biologically useful. According to the authors’ analysis, the reductions in initial infiltration rates and the decreased soil evaporation under crushed plant residues were the primary drivers of the increased aboveground biomass—meaning the water saved at the surface translated directly into plant growth.

What makes this approach notable is its circularity. Alfalfa and ryegrass residues are agricultural byproducts, often treated as waste or low-value material. Here they become a soil amendment that requires no manufacturing, no mining of clays, and no petrochemical hydrogels—alternatives that have been explored for sandy soil improvement but carry cost and environmental trade-offs. The researchers frame the technique as a contribution to a circular and green economy, in which biomass grown in one part of a landscape is recycled to repair another, closing nutrient and carbon loops while restoring degraded land.

The broader context gives the findings urgency. Drylands cover roughly 40 percent of the global land surface, and anthropogenic climate change has already pushed more than five million square kilometers of them toward desertification, according to prior research cited by the team. Desertification control efforts worldwide have struggled with the same core problem this study addresses: you cannot replant a landscape whose soil cannot hold water long enough for plants to take hold. Conventional fixes—clay deposition, mulching, irrigation infrastructure—are often expensive or unsustainable at scale. A low-tech residue incorporation method that measurably shifts infiltration, storage, and evaporation could be deployed by local land managers with little more than a crusher and spreading equipment.

The study also connects to a growing body of work on hydrological processes in sandy ecosystems. Earlier research by some of the same authors showed that litter crusts and root decay processes influence soil infiltration capacity and water replenishment in arid and semi-arid regions, and that nature-based solutions can reduce vertical water leakage in sandy land. The new trial extends that logic from understanding natural water-conservation mechanisms to actively engineering them, using plant material itself as the structural agent. It complements parallel findings that green manures and residue retention reduce evaporation and improve water productivity in arid cropping systems, suggesting a consistent principle: organic matter in the right place at the right particle size reshapes the soil water economy.

Caveats remain, as with any single field trial. The reported field capacity gains are modest, and the long-term durability of residue effects will depend on decomposition rates, repeated applications, and local climate. The authors note that data supporting the findings are available from the corresponding author upon reasonable request, and the work was funded by the Key Research and Development Plan of Ningxia Hui Autonomous Region and the National Natural Science Foundation of China. Still, the magnitude of the vegetation response—coverage more than doubling and biomass tripling in the best treatments—marks residue incorporation as one of the most promising low-cost tools yet documented for coaxing life back into arid sandy land. If the results hold across other degraded drylands, the humble act of crushing leftover plants and folding them into the sand could become a cornerstone of global desertification reversal.

Subject of Research: Effects of crushed plant residue incorporation on soil water retention and natural vegetation establishment in arid sandy land

Article Title: Incorporation of plant residues enhances water retention capacity and promotes natural vegetation establishment in arid sandy land

Article References: Wu, G.-L., Rodrigo-Comino, J., Cheng, Z., & Wu, X. (2026). Incorporation of plant residues enhances water retention capacity and promotes natural vegetation establishment in arid sandy land. Plant and Soil. https://doi.org/10.1007/s11104-026-09161-8

Image Credits: AI Generated

DOI: 10.1007/s11104-026-09161-8

Keywords: arid sandy land, soil restoration, plant residues, water retention, soil infiltration, soil evaporation, field capacity, vegetation coverage, aboveground biomass, desertification, circular economy, nature-based solutions

The research, led by Dhia Zalfa Zahira, Anjar Dimara Sakti, and Ketut Wikantika of Institut Teknologi Bandung, goes beyond simply counting lost trees. It tracks how much carbon those forests actually stored — both above ground in trunks, branches, and leaves, and below ground in root systems — and how that storage changed year by year. Most previous studies of the region offered static snapshots for a single year or focused only on aboveground biomass. By modelling both aboveground and belowground biomass carbon across five time points, the team has produced one of the most dynamic pictures yet of a tropical carbon reservoir in decline.

Defining what counts as an intact forest was itself a technical feat. The researchers started with the Hansen tree cover dataset derived from Landsat imagery, flagging any pixel with at least 30 percent canopy cover in 2000. They then filtered out oil-palm plantations using a global planting-year dataset and removed cropland using the MODIS land cover product. From that baseline, they iteratively subtracted three drivers of loss — stand-replacing deforestation, new cropland, and new oil-palm planting — at five-year intervals, generating intact forest maps for 2000, 2005, 2010, 2015, and 2020. This multi-layer filtering matters because oil-palm plantations, though tree-covered, are not ecologically intact forests; they typically replace natural forest and store far less carbon.

With the forest maps in hand, the team turned to a Random Forest regression model, an ensemble machine learning method that aggregates predictions from hundreds of decision trees. Thirteen environmental predictors fed the model, spanning vegetation indices such as NDVI and NIRv, productivity metrics including gross and net primary productivity, leaf area index, and the fraction of absorbed photosynthetically active radiation, along with climate variables like land surface temperature and precipitation, and topographic features including elevation, slope, and a topographic wetness index. All layers were processed at 300-metre resolution and aggregated to 1.5-by-1.5-kilometre grids to reduce noise and pixel misalignment.

A key design choice set this model apart: rather than training only on forest pixels, the researchers trained across the full gradient of Southeast Asian vegetation — forests, shrublands, grasslands, and cropland alike — using more than two million grid cells from 2010. This ensures the model learned the entire spectrum of carbon density, reducing bias at forest edges and recently disturbed areas. The data was split 80-20 into training and testing sets, with a Kolmogorov-Smirnov test confirming that both subsets covered the region geographically without spatial bias. Against the NASA reference biomass dataset, the model achieved a testing R² of 0.899 for aboveground carbon and 0.874 for belowground carbon, indicating strong predictive skill with limited overfitting.

Independent validation against the European Space Agency’s Climate Change Initiative Biomass product — built from radar observations by Sentinel-1, Envisat, and the ALOS satellites — showed moderate but consistent agreement, with R² values of 0.645 and 0.632 for the two carbon components. Among the thirteen predictors, the fraction of absorbed photosynthetically active radiation emerged as the single most influential variable, which makes physical sense: how much solar energy a canopy captures directly governs how much carbon it can fix into biomass. Land surface temperature and precipitation followed, while the classic vegetation indices contributed surprisingly little, apparently limited in their sensitivity to structural biomass in dense tropical canopies.

The carbon accounting that followed is sobering. In 2000, intact forests in the region held an estimated 26.18 megatonnes of aboveground biomass carbon and 6.61 megatonnes belowground, a combined total of roughly 32.8 megatonnes. By 2020, those figures had fallen to 15.96 and 3.96 megatonnes respectively — a drop of more than 39 percent in twenty years. The steepest losses occurred between 2010 and 2015, when forest area shrank by nearly 384,000 square kilometres and biomass carbon declined at a rate of about 1,496 megatonnes per year in total. The hardest-hit landscapes were Kalimantan and Sumatra in Indonesia, Sarawak and Sabah in Malaysia, and cropland-driven loss zones in northern Thailand, Myanmar, and parts of Papua.

To translate those ecological losses into terms policymakers cannot ignore, the team ran an indicative economic valuation under two carbon pricing scenarios. At the current average price across ASEAN+3 countries — a modest 6.45 US dollars per tonne of CO₂ equivalent — the region’s intact forest carbon was worth about 776 billion dollars in 2000, falling to roughly 471 billion by 2020. Under the globally recommended price of 75 dollars per tonne, aligned with social cost of carbon estimates, the value dropped from about 9.02 trillion dollars to 5.48 trillion. The authors are careful to stress these are illustrative, non-market figures, not credit-eligible carbon trading values, but they make the scale of the liability unmistakable: a loss of nearly 4 trillion dollars in indicative carbon assets in two decades.

The findings land at a delicate moment for climate policy. Southeast Asia harbours nearly 15 percent of the world’s tropical forests, and their continued erosion threatens not only regional biodiversity but global mitigation commitments under the Paris Agreement, REDD+ frameworks, and Sustainable Development Goal targets on sustainable forest management. There is also a real risk of a tipping dynamic: as intact forests degrade, they can flip from carbon sinks to net carbon sources, emitting more than they absorb and accelerating the very warming that stresses them. The spatially explicit maps produced by this study could help governments prioritise conservation zones where avoided deforestation yields the greatest climate benefit, and design community-based forest management schemes that recognise indigenous peoples as the primary stewards of these carbon reservoirs.

The study is candid about its limitations. The 2000 baseline may include some old plantations that the Hansen dataset cannot distinguish from natural forest; the model assumes predictor relationships stable across two decades; and the absence of explicit spatial block cross-validation means the reported accuracy figures may be somewhat optimistic. The economic valuation, by design, omits additionality, leakage, and monitoring requirements. Still, the core signal — a 59 percent collapse in intact forest area and a 39 percent loss of biomass carbon, tracked with low reported uncertainty — is difficult to dispute. Future work integrating radar-based biomass data and country-specific carbon accounting could sharpen the picture further. What this study makes clear is that Southeast Asia’s intact forests are not merely an ecological treasure; they are a quantifiable, and rapidly depreciating, planetary asset.

Subject of Research: Remote sensing and random forest modelling of intact forest biomass carbon dynamics and carbon economics in Southeast Asia from 2000 to 2020

Article Title: Intact forest carbon dynamics and indicative carbon economy in Southeast Asia: A remote sensing and random forest approach

Article References: Zahira, D. Z., Sakti, A. D., & Wikantika, K. (2026). Intact forest carbon dynamics and indicative carbon economy in Southeast Asia: A remote sensing and random forest approach. Environmental Advances, 26, Article 100758. https://doi.org/10.1016/j.envadv.2026.100758

Image Credits: AI Generated

DOI: 10.1016/j.envadv.2026.100758

Keywords: intact forests, Southeast Asia, biomass carbon, remote sensing, random forest, deforestation, oil palm, carbon pricing, REDD+, aboveground biomass, belowground biomass, machine learning

The research, authored by Lin Shang of Jiuquan Vocational Technical University, tackles the core weakness of previous approaches: most remote sensing estimates rely on a single data source and simple data stitching, which breaks down when weather, terrain and vegetation structure vary. Instead of merely concatenating datasets, the new framework performs what the author calls deep integration, fusing multi-scale spatial features with time-frequency dynamics extracted from high-resolution optical imagery, synthetic aperture radar, MODIS time series, meteorological records and ground measurements. The result is a feature representation that correlates with measured biomass at a Pearson coefficient of 0.95, a striking figure for such complex natural scenes.

The technical pipeline is built from three sequential signal-processing stages. First, a Gaussian pyramid applies Gaussian kernel convolution and layer-by-layer downsampling, with a 5 by 5 kernel and four downsampling levels, to generate a coarse-to-fine stack of images. This preserves both fine pixel-level texture and broad regional patterns, addressing the hierarchical spatial heterogeneity that grasslands display from individual plants to whole patches. Because pyramid downsampling inevitably discards high-frequency detail, the framework then turns to the wavelet transform, using the Mexican Hat wavelet as a basis function across three decomposition layers with scale parameters of 1, 2 and 4.

The wavelet stage is where the method gets genuinely clever about time. Meteorological drivers such as temperature and precipitation produce non-stationary, fluctuating biomass dynamics, and single-domain analysis tends to smear this information away. By computing the energy spectrum, the sum of squared wavelet coefficients, and the root mean square error of coefficients for each frequency band, the algorithm compresses each pixel’s entire time series into a single feature vector that retains the distinct contribution of every frequency band. This avoids the dimensionality explosion of naively stacking multi-temporal data while keeping the time-frequency fingerprint of the growing season intact.

With a high-dimensional, heavily collinear feature set in hand, principal component analysis takes over. The data are standardized to zero mean and unit variance, the covariance matrix is decomposed into eigenvalues and eigenvectors, and the top principal components are retained as a compact, non-redundant projection. In benchmarks against three simplified feature extraction schemes, the full PCA-GP-WT pipeline achieved a variance inflation factor of just 2.25 at a data size of 2500, far below the 4.29 of a seasonal decomposition baseline and the 7.91 of a sliding-window filter approach, while also posting the lowest mean squared error, 142.43, and a response time of 12.86 seconds at 5000 samples.

The estimation model itself, named PGW-UX, pairs a U-shaped network with extreme gradient boosting in a deliberately decoupled design. UNet, with its four downsampling and four upsampling stages and skip connections, acts purely as a spatial semantic encoder, preserving shallow pixel detail while capturing deep structure. Its feature maps are then fed, alongside the low-dimensional PCA-GP-WT features and ground measurements, into XGBoost, which performs the final nonlinear regression with regularization to suppress overfitting. This division of labor sidesteps a classic pitfall: end-to-end deep regression tends to overfit when ground truth samples are scarce, which is almost always the case in remote grasslands.

That scarcity was real here. The study drew on 48 field plots across 16 core observation areas spanning temperate grasslands in Inner Mongolia, alpine meadows on the Qinghai-Tibet Plateau and desert grasslands in Xinjiang, each harvested, dried at 65 degrees Celsius and weighed during the 2022 growing season. To give UNet enough training signal, the author used ordinary kriging interpolation to generate a continuous biomass surface as an auxiliary training target, while reserving the actual measured values for XGBoost calibration and validation. Spatial block cross-validation, which forced samples closer than 500 meters into the same fold, shifted error by less than 0.5 grams per square meter, suggesting spatial autocorrelation did not inflate the results.

The performance numbers are the study’s most viral-worthy element. Across the three grassland types, PGW-UX achieved an R-squared of 0.88 and a minimum root mean square error of 10.69 grams per square meter, with accuracies of 93.04 percent in Inner Mongolia and 92.13 percent on the Qinghai-Tibet Plateau. Comparison models, including random forest and support vector machine pipelines built on simpler features, trailed by measurable margins, and ablation experiments confirmed that every component earns its place: adding the multi-source integration module improved R-squared by 0.06, adding UNet spatial features cut mean absolute error by 1.96, and adding XGBoost calibration lifted R-squared by another 0.09.

Generalization was tested on an independent dataset from the Qinghai-Xizang Plateau covering 80 plots and six grassland types collected between 2020 and 2022. There, PGW-UX posted an R-squared of 0.81 and an RMSE of 118.25 grams per square meter, edging out a Transformer-based estimator and clearly beating random forest and standalone XGBoost. Bootstrap resampling with 1000 draws produced 90 percent confidence intervals roughly plus or minus 3 to 4 grams per square meter wide, with an overall coverage of 88.2 percent, giving ecologists a quantified sense of how much to trust each pixel-level prediction.

The implications reach well beyond grassland science. Reliable, large-scale biomass maps underpin carbon accounting, grazing carrying capacity assessments and degradation early-warning systems, and the author notes the framework could transfer to forest carbon stock and crop yield estimation. Limitations remain: the model has not been stress-tested in extreme climate years, resampling loses some spatial information, and cross-regional deployment still requires local calibration. Future work points toward active and transfer learning to better represent rare samples, spatiotemporal fusion to sharpen meteorological resolution, and lightweight versions of the model for operational, real-time monitoring. For now, the study stands as a compelling demonstration that when satellites, radar, weather records and field science are fused intelligently rather than merely stacked, the invisible weight of the world’s grasslands comes into focus.

Subject of Research: AI-driven multi-source data integration for estimating aboveground biomass in grassland ecosystems

Article Title: Estimation techniques of aboveground biomass in grassland ecosystems based on multi-source data integration

Article References: Shang, L. (2026). Estimation techniques of aboveground biomass in grassland ecosystems based on multi-source data integration. Discover Artificial Intelligence, 6(1), Article 1351. https://doi.org/10.1007/s44163-026-02376-9

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02376-9

Keywords: aboveground biomass, grassland ecosystems, remote sensing, deep learning, UNet, XGBoost, data fusion, wavelet transform, Gaussian pyramid, PCA, Sentinel-2, carbon cycling