…

Evapotranspiration sits at the intersection of the water and energy cycles, and its accurate quantification has long challenged hydrologists and agronomists alike. Because transpiration is physically coupled to photosynthesis through the stomata, a crop that conserves water by closing its stomata simultaneously curtails carbon uptake, which is why water stress manifests so quickly in canopy temperature. Thermal infrared radiometry exploits this coupling: when stomata close, latent heat flux falls, a larger share of absorbed solar energy is dissipated as sensible heat, and the canopy warms relative to a well-watered reference. The study by Bates and colleagues leverages this physical link, but rather than treating the surface as a single thermal entity, it adopts the two-source framework that explicitly separates soil and canopy contributions to the surface energy budget. This distinction matters because bare soil in a row crop can be many degrees warmer than the foliage it surrounds, and conflating the two temperatures introduces systematic bias into evapotranspiration estimates, particularly early in the season or in clumped canopies where soil is directly exposed to solar radiation.

The two-source formulation traces back to the work of Norman and colleagues in the mid-1990s, who recognized that satellite pixels and even field-scale observations almost always contain a mixture of soil and vegetation signals. By assigning distinct temperatures, aerodynamic resistances, and radiation regimes to the soil and canopy layers, the model partitions net radiation into sensible, latent, and ground heat fluxes for each component. The Priestley–Taylor variant used in this study sidesteps one of the thorniest problems in land surface modeling: the specification of near-surface vapor pressure. The Penman–Monteith variant requires aerodynamic conductance to water vapor and assumptions about humidity at the canopy and soil surfaces, inputs that are notoriously difficult to constrain from remote sensing alone. By instead applying the Priestley–Taylor coefficient to partition the energy available to the canopy, the model relies primarily on radiometric temperature and canopy structure, both of which can be measured directly from an unmanned aerial vehicle. This makes the approach unusually well suited to drone deployment, where meteorological inputs are limited to what a small weather station can provide.

One of the more subtle contributions of this work lies in how it characterizes the canopy itself. Most evapotranspiration models consume a single vegetation index or leaf area estimate, implicitly assuming that all intercepted foliage is physiologically active. In reality, a maturing crop accumulates senescent tissue that still intercepts light and contributes to structural area but no longer transpires at the same rate. The authors address this by fusing multispectral green area index with LiDAR-derived plant area index to estimate the fraction of leaf area that remains green. LiDAR captures the full three-dimensional architecture of the canopy, including brown and woody material that spectral indices tend to overlook, while the multispectral signal isolates the photosynthetically functional component. The ratio of the two yields a dynamic estimate of functional canopy cover that feeds directly into the transpiration scaling within the model. The finding that green area index and plant area index perform comparably during peak growth but diverge during stress and senescence is consistent with the known behavior of spectral indices, which saturate in dense canopies and lose sensitivity precisely when functional decline begins.

The sensitivity analysis reported in the study carries a practical warning for anyone deploying thermal sensors on drones. Uncooled thermal cameras, while affordable and lightweight, require careful calibration because their radiometric output drifts with ambient temperature and detector temperature. Since land surface temperature exerts first-order control on the computed sensible heat flux, even modest radiometric errors propagate directly into evapotranspiration estimates. The strong sensitivity documented here reinforces the value of in-flight calibration targets or coincident ground-based temperature measurements, practices that remain inconsistently applied across the growing body of drone-based thermal literature. As uncooled microbolometer technology continues to improve, the gap between research-grade and operational thermal mapping will narrow, but the calibration discipline demonstrated in this study sets a standard that operational users would do well to follow.

The validation against eddy covariance measurements deserves particular attention. Eddy covariance towers integrate fluxes over a footprint that shifts with wind direction and stability, typically spanning hundreds of meters, whereas the drone retrievals resolve the field at roughly fifteen centimeters. Comparing point-scale tower fluxes with spatially distributed model output therefore requires aggregation, and the reported agreement, with root mean square errors between 0.08 and 0.31 millimeters per hour at the instantaneous scale, is respectable for this class of model. The observation that performance degrades during senescence is physically interpretable: as canopies turn brown and heterogeneous, the assumptions underlying the resistance formulations and the partitioning of energy between soil and vegetation become increasingly strained. The finding that daily estimates improve when the tower energy balance closes is also telling, because residual energy in eddy covariance systems is a well-documented source of uncertainty that complicates validation across sites and seasons.

The upscaling step, in which instantaneous fluxes are converted to daily totals using a reference evapotranspiration fraction, is a pragmatic solution to a persistent problem. Energy balance models produce fluxes at the moment of image acquisition, typically late morning or midday, but irrigation decisions and water accounting operate on daily totals. The reference evapotranspiration fraction method assumes that the ratio of actual to reference evapotranspiration remains roughly constant across the day, an approximation that holds reasonably well under moderate conditions but can break down during rapid weather transitions or under severe afternoon stress. The daily errors of roughly one to two millimeters reported here are within the range typically accepted for irrigation scheduling, though they highlight the inherent difficulty of translating snapshot observations into continuous water use estimates without additional temporal information.

Perhaps the most operationally significant result is the early detection of spatial stress patterns that structural and spectral metrics alone failed to reveal. Canopy temperature responds to stomatal closure within minutes to hours, whereas changes in greenness or biomass unfold over days to weeks. This temporal asymmetry means that thermal imagery can flag developing stress before it becomes visible in optical data, providing a window for corrective irrigation that purely spectral approaches miss. The sub-field delineation enabled by the very high spatial resolution is equally important, since within-field variability in soil properties, compaction, and water availability creates zones of differential stress that coarser satellite products average away. For precision irrigation, where the goal is to apply water variably across a field rather than uniformly, this spatial acuity translates directly into potential water savings and yield protection.

The choice of three agronomically distinct crops, sugar beet, potato, and winter wheat, strengthens the generalizability of the findings. These species differ markedly in canopy architecture, rooting depth, growth habit, and seasonal phenology, and a modeling framework that performs consistently across them is more likely to transfer to other cropping systems than one tuned to a single crop. Winter wheat, with its rapid spring green-up and terminal senescence, tests the model’s ability to track functional decline; potato, with its partially closed canopy for much of the season, probes the soil-vegetation partitioning; and sugar beet, with its dense rosette structure, challenges the canopy characterization in high-biomass conditions. The consistent improvement from incorporating the dynamic green fraction across all three crops suggests that this refinement addresses a structural limitation rather than a crop-specific artifact.

Looking forward, the framework described here sits within a broader transition toward routine drone-based monitoring of field-scale water fluxes. As thermal sensor costs decline and flight automation improves, the barrier to repeated thermal surveys diminishes, and the bottleneck shifts to the modeling and interpretation side. The integration demonstrated in this study, fusing structural, spectral, and thermal observations within a physically grounded energy balance model, offers a template for that transition. It also underscores that no single sensor suffices: temperature reveals the physiological state, LiDAR reveals the architecture, and multispectral data reveal the functional green component, and it is their combination that unlocks reliable evapotranspiration estimation across the full spectrum of crop conditions from vigorous growth through stress and senescence.

Subject of Research: …

Article Title: Tracking crop evapotranspiration and water stress across seasons using multi-sensor UAV observations

Article References: Bates, J. S., Montzka, C., Vereecken, H., & Jonard, F. (2026). Tracking crop evapotranspiration and water stress across seasons using multi-sensor UAV observations. Biogeosciences, 23(20), 7253-7288. https://doi.org/10.5194/bg-23-7253-2026

Image Credits: AI Generated

DOI: 10.5194/bg-23-7253-2026

Keywords: Tracking, crop, evapotranspiration, water, stress, across, seasons, multi-sensor, observations, scientific research