Capturing patterns of evolutionary relatedness with reflectance spectra to model and monitor biodiversity.

Biogeographic history can set initial conditions for vegetation community assemblages that determine their climate responses at broad extents that land surface models attempt to forecast. Numerous studies have indicated that evolutionarily conserved biochemical, structural, and other functional attributes of plant species are captured in visible-to-short wavelength infrared, 400 to 2,500 nm, reflectance properties of vegetation. Here, we present a remotely sensed phylogenetic clustering and an evolutionary framework to accommodate spectra, distributions, and traits. Spectral properties evolutionarily conserved in plants provide the opportunity to spatially aggregate species into lineages (interpreted as "lineage functional types" or LFT) with improved classification accuracy. In this study, we use Airborne Visible/Infrared Imaging Spectrometer data from the 2013 Hyperspectral Infrared Imager campaign over the southern Sierra Nevada, California flight box, to investigate the potential for incorporating evolutionary thinking into landcover classification. We link the airborne hyperspectral data with vegetation plot data from 1372 surveys and a phylogeny representing 1,572 species. Despite temporal and spatial differences in our training data, we classified plant lineages with moderate reliability (Kappa = 0.76) and overall classification accuracy of 80.9%. We present an assessment of classification error and detail study limitations to facilitate future LFT development. This work demonstrates that lineage-based methods may be a promising way to leverage the new-generation high-resolution and high return-interval hyperspectral data planned for the forthcoming satellite missions with sparsely sampled existing ground-based ecological data.

Changes in timing of seasonal peak photosynthetic activity in northern ecosystems.

Seasonality in photosynthetic activity is a critical component of seasonal carbon, water, and energy cycles in the Earth system. This characteristic is a consequence of plant's adaptive evolutionary processes to a given set of environmental conditions. Changing climate in northern lands (>30°N) alters the state of climatic constraints on plant growth, and therefore, changes in the seasonality and carbon accumulation are anticipated. However, how photosynthetic seasonality evolved to its current state, and what role climatic constraints and their variability played in this process and ultimately in carbon cycle is still poorly understood due to its complexity. Here, we take the "laws of minimum" as a basis and introduce a new framework where the timing (day of year) of peak photosynthetic activity (DOYPmax ) acts as a proxy for plant's adaptive state to climatic constraints on its growth. Our analyses confirm that spatial variations in DOYPmax reflect spatial gradients in climatic constraints as well as seasonal maximum and total productivity. We find a widespread warming-induced advance in DOYPmax (-1.66 ± 0.30 days/decade, p < 0.001) across northern lands, indicating a spatiotemporal dynamism of climatic constraints to plant growth. We show that the observed changes in DOYPmax are associated with an increase in total gross primary productivity through enhanced carbon assimilation early in the growing season, which leads to an earlier phase shift in land-atmosphere carbon fluxes and an increase in their amplitude. Such changes are expected to continue in the future based on our analysis of earth system model projections. Our study provides a simplified, yet realistic framework based on first principles for the complex mechanisms by which various climatic factors constrain plant growth in northern ecosystems.

Persistent effects of a severe drought on Amazonian forest canopy.

Recent Amazonian droughts have drawn attention to the vulnerability of tropical forests to climate perturbations. Satellite and in situ observations have shown an increase in fire occurrence during drought years and tree mortality following severe droughts, but to date there has been no assessment of long-term impacts of these droughts across landscapes in Amazonia. Here, we use satellite microwave observations of rainfall and canopy backscatter to show that more than 70 million hectares of forest in western Amazonia experienced a strong water deficit during the dry season of 2005 and a closely corresponding decline in canopy structure and moisture. Remarkably, and despite the gradual recovery in total rainfall in subsequent years, the decrease in canopy backscatter persisted until the next major drought, in 2010. The decline in backscatter is attributed to changes in structure and water content associated with the forest upper canopy. The persistence of low backscatter supports the slow recovery (>4 y) of forest canopy structure after the severe drought in 2005. The result suggests that the occurrence of droughts in Amazonia at 5-10 y frequency may lead to persistent alteration of the forest canopy.

Variations in atmospheric CO2 growth rates coupled with tropical temperature.

Previous studies have highlighted the occurrence and intensity of El Niño-Southern Oscillation as important drivers of the interannual variability of the atmospheric CO2 growth rate, but the underlying biogeophysical mechanisms governing such connections remain unclear. Here we show a strong and persistent coupling (r(2) ≈ 0.50) between interannual variations of the CO2 growth rate and tropical land-surface air temperature during 1959 to 2011, with a 1 °C tropical temperature anomaly leading to a 3.5 ± 0.6 Petagrams of carbon per year (PgC/y) CO2 growth-rate anomaly on average. Analysis of simulation results from Dynamic Global Vegetation Models suggests that this temperature-CO2 coupling is contributed mainly by the additive responses of heterotrophic respiration (Rh) and net primary production (NPP) to temperature variations in tropical ecosystems. However, we find a weaker and less consistent (r(2) ≈ 0.25) interannual coupling between CO2 growth rate and tropical land precipitation than diagnosed from the Dynamic Global Vegetation Models, likely resulting from the subtractive responses of tropical Rh and NPP to precipitation anomalies that partly offset each other in the net ecosystem exchange (i.e., net ecosystem exchange ≈ Rh - NPP). Variations in other climate variables (e.g., large-scale cloudiness) and natural disturbances (e.g., volcanic eruptions) may induce transient reductions in the temperature-CO2 coupling, but the relationship is robust during the past 50 y and shows full recovery within a few years after any such major variability event. Therefore, it provides an important diagnostic tool for improved understanding of the contemporary and future global carbon cycle.

A high-resolution approach to estimating ecosystem respiration at continental scales using operational satellite data.

A better understanding of the local variability in land-atmosphere carbon fluxes is crucial to improving the accuracy of global carbon budgets. Operational satellite data backed by ground measurements at Fluxnet sites proved valuable in monitoring local variability of gross primary production at highly resolved spatio-temporal resolutions. Yet, we lack similar operational estimates of ecosystem respiration (Re) to calculate net carbon fluxes. If successful, carbon fluxes from such a remote sensing approach would form an independent and sought after measure to complement widely used dynamic global vegetation models (DGVMs). Here, we establish an operational semi-empirical Re model, based only on data from the Moderate Resolution Imaging Spectroradiometer (MODIS) with a resolution of 1 km and 8 days. Fluxnet measurements between 2000 and 2009 from 100 sites across North America and Europe are used for parameterization and validation. Our analysis shows that Re is closely tied to temperature and plant productivity. By separating temporal and intersite variation, we find that MODIS land surface temperature (LST) and enhanced vegetation index (EVI) are sufficient to explain observed Re across most major biomes with a negligible bias [R² = 0.62, RMSE = 1.32 (g C m(-2) d(-1)), MBE = 0.05 (g C m(-2) d(-1))]. A comparison of such satellite-derived Re with those simulated by the DGVM LPJmL reveals similar spatial patterns. However, LPJmL shows higher temperature sensitivities and consistently simulates higher Re values, in high-latitude and subtropical regions. These differences remain difficult to explain and they are likely associated either with LPJmL parameterization or with systematic errors in the Fluxnet sampling technique. While uncertainties remain with Re estimates, the model formulated in this study provides an operational, cross-validated and unbiased approach to scale Fluxnet Re to the continental scale and advances knowledge of spatio-temporal Re variability.

A large proportion of North American net ecosystem production is offset by emissions from harvested products, river/stream evasion, and biomass burning.

Diagnostic carbon cycle models produce estimates of net ecosystem production (NEP, the balance of net primary production and heterotrophic respiration) by integrating information from (i) satellite-based observations of land surface vegetation characteristics; (ii) distributed meteorological data; and (iii) eddy covariance flux tower observations of net ecosystem exchange (NEE) (used in model parameterization). However, a full bottom-up accounting of NEE (the vertical carbon flux) that is suitable for integration with atmosphere-based inversion modeling also includes emissions from decomposition/respiration of harvested forest and agricultural products, CO2 evasion from streams and rivers, and biomass burning. Here, we produce a daily time step NEE for North America for the year 2004 that includes NEP as well as the additional emissions. This NEE product was run in the forward mode through the CarbonTracker inversion setup to evaluate its consistency with CO2 concentration observations. The year 2004 was climatologically favorable for NEP over North America and the continental total was estimated at 1730 ± 370 TgC yr(-1) (a carbon sink). Harvested product emissions (316 ± 80 TgC yr(-1) ), river/stream evasion (158 ± 50 TgC yr(-1) ), and fire emissions (142 ± 45 TgC yr(-1) ) counteracted a large proportion (35%) of the NEP sink. Geographic areas with strong carbon sinks included Midwest US croplands, and forested regions of the Northeast, Southeast, and Pacific Northwest. The forward mode run with CarbonTracker produced good agreement between observed and simulated wintertime CO2 concentrations aggregated over eight measurement sites around North America, but overestimates of summertime concentrations that suggested an underestimation of summertime carbon uptake. As terrestrial NEP is the dominant offset to fossil fuel emission over North America, a good understanding of its spatial and temporal variation - as well as the fate of the carbon it sequesters ─ is needed for a comprehensive view of the carbon cycle.

Importance of carbon dioxide physiological forcing to future climate change.

An increase in atmospheric carbon dioxide (CO(2)) concentration influences climate both directly through its radiative effect (i.e., trapping longwave radiation) and indirectly through its physiological effect (i.e., reducing transpiration of land plants). Here we compare the climate response to radiative and physiological effects of increased CO(2) using the National Center for Atmospheric Research (NCAR) coupled Community Land and Community Atmosphere Model. In response to a doubling of CO(2), the radiative effect of CO(2) causes mean surface air temperature over land to increase by 2.86 +/- 0.02 K (+/- 1 standard error), whereas the physiological effects of CO(2) on land plants alone causes air temperature over land to increase by 0.42 +/- 0.02 K. Combined, these two effects cause a land surface warming of 3.33 +/- 0.03 K. The radiative effect of doubling CO(2) increases global runoff by 5.2 +/- 0.6%, primarily by increasing precipitation over the continents. The physiological effect increases runoff by 8.4 +/- 0.6%, primarily by diminishing evapotranspiration from the continents. Combined, these two effects cause a 14.9 +/- 0.7% increase in runoff. Relative humidity remains roughly constant in response to CO(2)-radiative forcing, whereas relative humidity over land decreases in response to CO(2)-physiological forcing as a result of reduced plant transpiration. Our study points to an emerging consensus that the physiological effects of increasing atmospheric CO(2) on land plants will increase global warming beyond that caused by the radiative effects of CO(2).

Temporal Interpolation of Geostationary Satellite Imagery With Optical Flow.

Applications of satellite data in areas such as weather tracking and modeling, ecosystem monitoring, wildfire detection, and land-cover change are heavily dependent on the tradeoffs to spatial, spectral, and temporal resolutions of observations. In weather tracking, high-frequency temporal observations are critical and used to improve forecasts, study severe events, and extract atmospheric motion, among others. However, while the current generation of geostationary (GEO) satellites has hemispheric coverage at 10-15-min intervals, higher temporal frequency observations are ideal for studying mesoscale severe weather events. In this work, we present a novel application of deep learning-based optical flow to temporal upsampling of GEO satellite imagery. We apply this technique to 16 bands of the GOES-R/Advanced Baseline Imager mesoscale dataset to temporally enhance full-disk hemispheric snapshots of different spatial resolutions from 10 to 1 min. Experiments show the effectiveness of task-specific optical flow and multiscale blocks for interpolating high-frequency severe weather events relative to bilinear and global optical flow baselines. Finally, we demonstrate strong performance in capturing variability during convective precipitation events.

A Framework for Deep Learning Emulation of Numerical Models With a Case Study in Satellite Remote Sensing.

Numerical models based on physics represent the state of the art in Earth system modeling and comprise our best tools for generating insights and predictions. Despite rapid growth in computational power, the perceived need for higher model resolutions overwhelms the latest generation computers, reducing the ability of modelers to generate simulations for understanding parameter sensitivities and characterizing variability and uncertainty. Thus, surrogate models are often developed to capture the essential attributes of the full-blown numerical models. Recent successes of machine learning methods, especially deep learning (DL), across many disciplines offer the possibility that complex nonlinear connectionist representations may be able to capture the underlying complex structures and nonlinear processes in Earth systems. A difficult test for DL-based emulation, which refers to function approximation of numerical models, is to understand whether they can be comparable to traditional forms of surrogate models in terms of computational efficiency while simultaneously reproducing model results in a credible manner. A DL emulation that passes this test may be expected to perform even better than simple models with respect to capturing complex processes and spatiotemporal dependencies. Here, we examine, with a case study in satellite-based remote sensing, the hypothesis that DL approaches can credibly represent the simulations from a surrogate model with comparable computational efficiency. Our results are encouraging in that the DL emulation reproduces the results with acceptable accuracy and often even faster performance. We discuss the broader implications of our results in light of the pace of improvements in high-performance implementations of DL and the growing desire for higher resolution simulations in the Earth sciences.

Greening of human-dominated ecosystems in India.

Satellite data show the Earth has been greening and identify croplands in India as one of the most prominent greening hotspots. Though India's agriculture has been dependent on irrigation enhancement to reduce crop water stress and increase production, the spatiotemporal dynamics of how irrigation influenced the satellite observed greenness remains unclear. Here, we use satellite-derived leaf area data and survey-based agricultural statistics together with results from state-of-the-art Land Surface Models (LSM) to investigate the role of irrigation in the greening of India's croplands. We find that satellite observations provide multiple lines of evidence showing strong contributions of irrigation to significant greening during dry season and in drier environments. The national statistics support irrigation-driven yield enhancement and increased dry season cropping intensity. These suggest a continuous shift in India's agriculture toward an irrigation-driven dry season cropping system and confirm the importance of land management in the greening phenomenon. However, the LSMs identify CO2 fertilization as a primary driver of greening whereas land use and management have marginal impacts on the simulated leaf area changes. This finding urges a closer collaboration of the modeling, Earth observation, and land system science communities to improve representation of land management in the Earth system modeling.

Structural complexity biases vegetation greenness measures.

Vegetation 'greenness' characterized by spectral vegetation indices (VIs) is an integrative measure of vegetation leaf abundance, biochemical properties and pigment composition. Surprisingly, satellite observations reveal that several major VIs over the US Corn Belt are higher than those over the Amazon rainforest, despite the forests having a greater leaf area. This contradicting pattern underscores the pressing need to understand the underlying drivers and their impacts to prevent misinterpretations. Here we show that macroscale shadows cast by complex forest structures result in lower greenness measures compared with those cast by structurally simple and homogeneous crops. The shadow-induced contradictory pattern of VIs is inevitable because most Earth-observing satellites do not view the Earth in the solar direction and thus view shadows due to the sun-sensor geometry. The shadow impacts have important implications for the interpretation of VIs and solar-induced chlorophyll fluorescence as measures of global vegetation changes. For instance, a land-conversion process from forests to crops over the Amazon shows notable increases in VIs despite a decrease in leaf area. Our findings highlight the importance of considering shadow impacts to accurately interpret remotely sensed VIs and solar-induced chlorophyll fluorescence for assessing global vegetation and its changes.

NASA Global Daily Downscaled Projections, CMIP6.

We describe the latest version of the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6). The archive contains downscaled historical and future projections for 1950-2100 based on output from Phase 6 of the Climate Model Intercomparison Project (CMIP6). The downscaled products were produced using a daily variant of the monthly bias correction/spatial disaggregation (BCSD) method and are at 1/4-degree horizontal resolution. Currently, eight variables from five CMIP6 experiments (historical, SSP126, SSP245, SSP370, and SSP585) are provided as procurable from thirty-five global climate models.

New generation geostationary satellite observations support seasonality in greenness of the Amazon evergreen forests.

Assessing the seasonal patterns of the Amazon rainforests has been difficult because of the paucity of ground observations and persistent cloud cover over these forests obscuring optical remote sensing observations. Here, we use data from a new generation of geostationary satellites that carry the Advanced Baseline Imager (ABI) to study the Amazon canopy. ABI is similar to the widely used polar orbiting sensor, the Moderate Resolution Imaging Spectroradiometer (MODIS), but provides observations every 10-15 min. Our analysis of NDVI data collected over the Amazon during 2018-19 shows that ABI provides 21-35 times more cloud-free observations in a month than MODIS. The analyses show statistically significant changes in seasonality over 85% of Amazon forest pixels, an area about three times greater than previously reported using MODIS data. Though additional work is needed in converting the observed changes in seasonality into meaningful changes in canopy dynamics, our results highlight the potential of the new generation geostationary satellites to help us better understand tropical ecosystems, which has been a challenge with only polar orbiting satellites.

Biophysical impacts of Earth greening largely controlled by aerodynamic resistance.

Satellite observations show widespread increasing trends of leaf area index (LAI), known as the Earth greening. However, the biophysical impacts of this greening on land surface temperature (LST) remain unclear. Here, we quantify the biophysical impacts of Earth greening on LST from 2000 to 2014 and disentangle the contributions of different factors using a physically based attribution model. We find that 93% of the global vegetated area shows negative sensitivity of LST to LAI increase at the annual scale, especially for semiarid woody vegetation. Further considering the LAI trends (P ≤ 0.1), 30% of the global vegetated area is cooled by these trends and 5% is warmed. Aerodynamic resistance is the dominant factor in controlling Earth greening's biophysical impacts: The increase in LAI produces a decrease in aerodynamic resistance, thereby favoring increased turbulent heat transfer between the land and the atmosphere, especially latent heat flux.

DeepEmSat: Deep Emulation for Satellite Data Mining.

The growing volume of Earth science data available from climate simulations and satellite remote sensing offers unprecedented opportunity for scientific insight, while also presenting computational challenges. One potential area of impact is atmospheric correction, where physics-based numerical models retrieve surface reflectance information from top of atmosphere observations, and are too computationally intensive to be run in real time. Machine learning methods have demonstrated potential as fast statistical models for expensive simulations and for extracting credible insights from complex datasets. Here, we develop DeepEmSat: a deep learning emulator approach for atmospheric correction, and offer comparison against physics-based models to support the hypothesis that deep learning can make a contribution to the efficient processing of satellite images.

China and India lead in greening of the world through land-use management.

Satellite data show increasing leaf area of vegetation due to direct (human land-use management) and indirect factors (climate change, CO2 fertilization, nitrogen deposition, recovery from natural disturbances, etc.). Among these, climate change and CO2 fertilization effect seem to be the dominant drivers. However, recent satellite data (2000-2017) reveal a greening pattern that is strikingly prominent in China and India, and overlapping with croplands world-wide. China alone accounts for 25% of the global net increase in leaf area with only 6.6% of global vegetated area. The greening in China is from forests (42%) and croplands (32%), but in India is mostly from croplands (82%) with minor contribution from forests (4.4%). China is engineering ambitious programs to conserve and expand forests with the goal of mitigating land degradation, air pollution and climate change. Food production in China and India has increased by over 35% since 2000 mostly due to increasing harvested area through multiple cropping facilitated by fertilizer use and surface/ground-water irrigation. Our results indicate that the direct factor is a key driver of the "Greening Earth", accounting for over a third, and likely more, of the observed net increase in green leaf area. They highlight the need for realistic representation of human land-use practices in Earth system models.

Deep neural networks for texture classification-A theoretical analysis.

We investigate the use of Deep Neural Networks for the classification of image datasets where texture features are important for generating class-conditional discriminative representations. To this end, we first derive the size of the feature space for some standard textural features extracted from the input dataset and then use the theory of Vapnik-Chervonenkis dimension to show that hand-crafted feature extraction creates low-dimensional representations which help in reducing the overall excess error rate. As a corollary to this analysis, we derive for the first time upper bounds on the VC dimension of Convolutional Neural Network as well as Dropout and Dropconnect networks and the relation between excess error rate of Dropout and Dropconnect networks. The concept of intrinsic dimension is used to validate the intuition that texture-based datasets are inherently higher dimensional as compared to handwritten digits or other object recognition datasets and hence more difficult to be shattered by neural networks. We then derive the mean distance from the centroid to the nearest and farthest sampling points in an n-dimensional manifold and show that the Relative Contrast of the sample data vanishes as dimensionality of the underlying vector space tends to infinity.

Global Distribution and Density of Constructed Impervious Surfaces.

We present the first global inventory of the spatial distribution and density ofconstructed impervious surface area (ISA). Examples of ISA include roads, parking lots,buildings, driveways, sidewalks and other manmade surfaces. While high spatialresolution is required to observe these features, the new product reports the estimateddensity of ISA on a one-km² grid based on two coarse resolution indicators of ISA - thebrightness of satellite observed nighttime lights and population count. The model wascalibrated using 30-meter resolution ISA of the USA from the U.S. Geological Survey.Nominally the product is for the years 2000-01 since both the nighttime lights andreference data are from those two years. We found that 1.05% of the United States landarea is impervious surface (83,337 km²) and 0.43 % of the world's land surface (579,703km²) is constructed impervious surface. China has more ISA than any other country(87,182 km²), but has only 67 m² of ISA per person, compared to 297 m² per person in theUSA. The distribution of ISA in the world's primary drainage basins indicates that watersheds damaged by ISA are primarily concentrated in the USA, Europe, Japan, China and India. The authors believe the next step for improving the product is to include reference ISA data from many more areas around the world.

Deploying a quantum annealing processor to detect tree cover in aerial imagery of California.

Quantum annealing is an experimental and potentially breakthrough computational technology for handling hard optimization problems, including problems of computer vision. We present a case study in training a production-scale classifier of tree cover in remote sensing imagery, using early-generation quantum annealing hardware built by D-wave Systems, Inc. Beginning within a known boosting framework, we train decision stumps on texture features and vegetation indices extracted from four-band, one-meter-resolution aerial imagery from the state of California. We then impose a regulated quadratic training objective to select an optimal voting subset from among these stumps. The votes of the subset define the classifier. For optimization, the logical variables in the objective function map to quantum bits in the hardware device, while quadratic couplings encode as the strength of physical interactions between the quantum bits. Hardware design limits the number of couplings between these basic physical entities to five or six. To account for this limitation in mapping large problems to the hardware architecture, we propose a truncation and rescaling of the training objective through a trainable metaparameter. The boosting process on our basic 108- and 508-variable problems, thus constituted, returns classifiers that incorporate a diverse range of color- and texture-based metrics and discriminate tree cover with accuracies as high as 92% in validation and 90% on a test scene encompassing the open space preserves and dense suburban build of Mill Valley, CA.

Vegetation Greening and Climate Change Promote Multidecadal Rises of Global Land Evapotranspiration.

Recent studies showed that anomalous dry conditions and limited moisture supply roughly between 1998 and 2008, especially in the Southern Hemisphere, led to reduced vegetation productivity and ceased growth in land evapotranspiration (ET). However, natural variability of Earth's climate system can degrade capabilities for identifying climate trends. Here we produced a long-term (1982-2013) remote sensing based land ET record and investigated multidecadal changes in global ET and underlying causes. The ET record shows a significant upward global trend of 0.88 mm yr(-2) (P < 0.001) over the 32-year period, mainly driven by vegetation greening (0.018% per year; P < 0.001) and rising atmosphere moisture demand (0.75 mm yr(-2); P = 0.016). Our results indicate that reduced ET growth between 1998 and 2008 was an episodic phenomenon, with subsequent recovery of the ET growth rate after 2008. Terrestrial precipitation also shows a positive trend of 0.66 mm yr(-2) (P = 0.08) over the same period consistent with expected water cycle intensification, but this trend is lower than coincident increases in evaporative demand and ET, implying a possibility of cumulative water supply constraint to ET. Continuation of these trends will likely exacerbate regional drought-induced disturbances, especially during regional dry climate phases associated with strong El Niño events.