A drone flies across a field and returns a map of bright green, yellow and red rectangles. Someone calls it a crop-health map, but the colours alone do not establish plant health. They are often a display applied to an index calculated from selected spectral bands. If the camera was poorly calibrated, if shadows crossed the field, or if bare soil filled many pixels, the map can be visually persuasive while answering the wrong biological question.
This article explains how to read UAV multispectral vegetation products, including normalized difference vegetation index (NDVI), a common normalized difference red-edge index (NDRE), reflectance calibration panels, illumination effects and field validation. We will move from first principles to carefully checked numerical examples, then to interpretation, errors and independent problems. The goal is to translate the map into statements that respect its sensors and its limits.
The method is relevant to agronomy, environmental monitoring, forestry and science education. It is also an exercise in vocabulary: reflectance is not the same as digital brightness; an index is not the same as a concentration; a colour class is not the same as a diagnosis; and a pretty orthomosaic is not automatically a trustworthy quantitative dataset. Clear language begins by keeping those ideas separate.
Start with photons, bands and reflectance
Sunlight contains many wavelengths. A multispectral sensor records reflected light in a set of defined wavebands rather than recording one human-style colour photograph. A typical vegetation payload may include green, red, red-edge and near-infrared bands, and sometimes blue. Different instruments choose different centre wavelengths and bandwidths. A number called red-edge can therefore represent a somewhat different spectral region on two devices.
Reflectance is a ratio of reflected to incident radiation under specified geometry and measurement conventions. It is dimensionless, often written from zero to one, although noisy corrected products may occasionally fall outside the ideal range. A raw digital number from a camera is different. It includes the effects of exposure, sensor gain, optical transmission, illumination and processing. One cannot simply substitute raw pixel values for calibrated reflectance in a scientific calculation without understanding the implications.
Green leaves absorb much of the red light for photosynthesis while leaf internal structure strongly influences near-infrared reflectance. This broad spectral contrast supports familiar vegetation indices. Yet leaf water, pigment composition, canopy architecture, soil, shadows and view angle all affect the recorded signal. A vegetation index is useful because it summarizes part of that complexity, not because it eliminates it.
A reliable caption should name the sensor, band definitions, flight timing, whether reflectance calibration was performed, the pixel resolution and the processing product. If it says merely “drone analysis showed low crop health,” it has skipped the measurement chain. The reader should first learn which spectra were combined and which interpretation was subsequently tested.
Calculate NDVI, with units and signs checked
NDVI is defined as (NIR − Red)/(NIR + Red), with the denominator nonzero and both bands measured on compatible scales. For reflectances between zero and one, the ratio lies between minus one and plus one. NDVI is dimensionless. It is not a percent of green leaves, a measured nitrogen concentration, a chlorophyll assay or a yield prediction by itself.
Consider a hypothetical canopy pixel with red reflectance 0.12 and near-infrared reflectance 0.48. The numerator is 0.36; the denominator is 0.60; NDVI = 0.600. Check that the result is positive because near-infrared reflectance exceeds red. Also notice that the ratio is not 0.48 minus 0.12, which would be a difference of 0.36 rather than the normalized difference.
Now imagine red reflectance 0.28 and near-infrared reflectance 0.32. The numerator is 0.04 and the denominator 0.60, giving NDVI approximately 0.0667. Such a low value might arise from sparse vegetation, soil contamination or another non-green surface. It is insufficient evidence to declare a particular nutrient deficiency. Many very different conditions can produce the same simple ratio.
If the denominator is nearly zero, measurement noise makes the quotient unstable. If software scales reflectance to integers, the arithmetic may still work when both bands use the same linear scale and offset handling, but an added offset can alter the index. Check product metadata and any scale factors before reading a reported NDVI. A value stored as 6000 could represent 0.6000 or an entirely different encoding, depending on the dataset.
What the red-edge band adds
The red-edge region lies near the steep transition between strong red absorption and elevated near-infrared reflectance in green vegetation. Some sensors measure one or more narrow bands there. A normalized difference red-edge index is commonly written NDRE = (NIR − RedEdge)/(NIR + RedEdge). The exact bands and naming convention must be reported because there is more than one red-edge index in the literature.
Suppose near-infrared reflectance is 0.48 and a red-edge band gives 0.30. Then NDRE = 0.18/0.78, approximately 0.2308. The number is not directly comparable with NDVI 0.600 from the same pixel as though both represented the same scale of plant vigour. They use different denominators and respond differently to the leaf and canopy spectrum.
Red-edge measurements can be useful when a red-band-based index becomes less responsive in dense vegetation, but the phrase “does not saturate” should not be treated as universal. Red-edge indices also have response limits, sensor differences, noise and confounding factors. Changes may reflect pigment, canopy structure, leaf area or geometry. Their usefulness depends on the specific crop, stage, treatment and validation objective.
Do not translate the red-edge band as a red border around the field. It is a spectral interval. Similarly, an index named NDRE is not a direct measurement of nitrogen uptake; it may be empirically correlated with nitrogen status in a particular dataset, but that relation must be established with appropriate ground measurements and tested beyond the data used for fitting.
Why calibration panels matter
A reflectance panel provides a physical reference whose spectral properties are known or measured. During a field campaign, the drone camera can photograph the panel under appropriate conditions so image processing has an anchor for estimating reflectance. This is more informative than assuming that a raw sensor value of 150 always means the same reflected fraction, regardless of flight or sunlight.
The U.S. Geological Survey’s published calibration-panel research explains an important limitation: a panel’s reflectance can change with exposure, dust and handling. A label stating that a panel was originally supplied with a known reflectance does not guarantee it remains unchanged after years in the field. Inspect panel condition and, where needed, verify its spectrum rather than treating the panel as a magical constant.
An idealized illustration helps. Suppose a detector’s dark-corrected signal from an unknown target is 480 counts, and a reference panel under matched conditions gives 800 counts with certified band reflectance 0.50. If the instrument behaves linearly and geometry and illumination are genuinely comparable, a simple relative estimate is 0.50 × (480/800) = 0.30. Real sensor pipelines may need exposure corrections, irradiance measurements and more elaborate calibration, but the arithmetic explains the basic role of a reference.
Now imagine the panel was dirty and its true reflectance had fallen to 0.45 although the analyst used the old value 0.50. The simplified conversion would be systematically biased upward, even if the arithmetic were perfect. Calibration improves physical traceability only when the reference material and processing assumptions are trustworthy.
Illumination changes during a flight
A drone may cover a large field over several minutes, during which clouds pass and solar angle changes. One part of the mosaic may be acquired in direct sun and another under diffuse light. Overlapping photographs may also have different exposure settings. An index normalizes some common effects, but it cannot guarantee identical measurements when the spectral illumination or viewing geometry changes.
A downwelling light sensor may help record incident illumination, but its readings must be time-aligned, properly oriented and consistent with the camera calibration. An irradiance value taken after a cloud passes may be inappropriate for an image captured before the cloud arrived. Software cannot correct information it did not record or identify correctly.
Before accepting a strip-shaped pattern as crop stress, compare its edges with flight lines and acquisition times. If the pattern follows the image swaths rather than the direction of planting or soil management, a radiometric seam is plausible. Conversely, if a stress pattern follows irrigation zones or measured soil boundaries and appears across independently calibrated flights, it merits further investigation.
Field notes are valuable. Record weather, solar conditions, panel images, exposure settings, time, crop stage and recent operations. That context enables an analyst to identify a problem without inventing a biological cause. A map with consistent radiometry often requires as much care in planning the flight as in drawing the final colour legend.
Orthomosaics are not just photographs pasted together
An orthomosaic combines overlapping images into a georeferenced map, typically using camera geometry, feature matching and a surface model. It corrects for much of the perspective difference between photographs, yet residual misregistration can remain. A small positioning error matters when rows, narrow paths or individual plant canopies are only a few pixels wide.
If the camera bands are housed in separate optical channels, their images must also be registered with one another. NDVI subtracts values from matching ground locations. If a red pixel represents soil just outside a leaf while the near-infrared pixel represents the leaf itself, the index can be wrong even when both spectral measurements are individually plausible.
Resampling is another hidden step. Nearest-neighbour, bilinear and cubic procedures can produce different mixtures of values at edges, especially in fine-resolution crop images. An index calculated before resampling may not be identical to one calculated after combining resampled bands. That difference becomes important near canopy boundaries and bare patches.
The correct plain-language interpretation of an orthomosaic includes the concept of spatial registration. A coloured edge is not necessarily an exact physical border. Before reporting a precise stressed area in square metres, verify georeferencing, pixel size, classification rule and whether narrow or mixed pixels were included.
Soil background changes the index
Early in a growing season, many UAV pixels contain more exposed soil than leaves. Soil reflectance is shaped by texture, minerals, moisture, organic matter and surface condition. A dark wet soil and bright dry soil can shift vegetation indices differently even if the crop leaves are unchanged. That is why a low-index field edge may be a bare patch rather than a physiologically stressed plant.
Consider two hypothetical endmembers. A leaf-dominated target has red reflectance 0.12 and NIR 0.48, giving NDVI 0.60. A soil-dominated target has red 0.24 and NIR 0.30, giving NDVI approximately 0.111. Suppose an unresolved pixel captures an idealized equal reflectance mixture: red 0.18 and NIR 0.39. Its NDVI is 0.21/0.57, approximately 0.3684.
The mixed-pixel value is not simply the arithmetic mean of the two endmember NDVI values: (0.60 + 0.111)/2 is about 0.3555. Normalization is nonlinear. Real mixing can be more complicated because canopies cast shadows, scatter light and expose soil unevenly. The simple calculation demonstrates why mean index and index of mean reflectances should not be confused.
If crop cover increases, NDVI may rise even with no increase in chlorophyll concentration per leaf. The camera is seeing a larger share of leaf material and a smaller share of soil. A biological interpretation should ask whether the observed variation represents leaf physiology, canopy coverage, plant number or some combination.
Dense vegetation and index response
In dense canopies, red reflectance may become very low while near-infrared reflectance varies with structure and multiple scattering. Under some conditions, further increases in leaf area or pigment produce relatively small NDVI changes. Analysts call this a saturation tendency, but the precise behaviour depends on canopy architecture, sensor bands and observation conditions.
A red-edge index can sometimes remain more responsive to changes in dense vegetation, yet it may also respond to leaf nitrogen, chlorophyll, phenology and illumination effects. It is not automatically a more accurate measure of every target variable. The appropriate choice follows from a question: are we mapping cover, pigment-related variability, a validated nitrogen surrogate or spatial anomalies requiring inspection?
A product with NDVI 0.85 should not be read as 85 percent healthy. Similarly, a drop from 0.82 to 0.78 should not be announced as a four-percent drop in biomass. It is a difference of 0.04 index units, and the biological meaning requires empirical interpretation. Translating an index number into ordinary percentages is a common and consequential error.
One useful way to communicate the limitation is to show index values alongside ground observations for several canopy densities. If measured leaf area continues increasing while NDVI flattens, readers can see the saturation phenomenon directly. The demonstration is stronger than a blanket assertion that one spectral index is always superior.
Shadow, angle and the same crop on two dates
On a sunny day, canopy shadows can be very dark and spectrally different from sunlit leaves. If a second flight occurs at a different time of day, the shadow fraction changes. A pixel may then produce a different NDVI without a corresponding change in plant condition. This is particularly problematic in orchards, tall crops and forest stands with complex geometry.
Flight altitude and camera view angle influence the fraction of row gaps and canopy sides visible to the sensor. The bidirectional reflectance distribution describes how reflected radiation varies with illumination and view direction. An oblique pixel is not necessarily directly comparable to a near-nadir pixel, even after georeferencing, because the physical surfaces contributing to the signal can differ.
Comparing dates requires consistent acquisition geometry where possible, stable calibration and careful masks. It can be useful to compare flights at similar solar times and phenological stages, but weather and biological differences still need explicit consideration. A repeat survey is not automatically a controlled experiment merely because a drone followed the same route.
A reader should look for flight metadata before trusting fine date-to-date contrasts. Where a document offers only a false-colour image and no acquisition conditions, it may be a qualitative illustration rather than a basis for quantitative claims about crop change.
Why a coloured map needs its actual scale
Map software may render low index values red, middle values yellow and high values green. Those colours are arbitrary display categories. The underlying spectral data do not contain red polygons that inherently mean diseased plants. Changing the colour stretch may make a field appear dramatically worse or better without changing any measured value.
A continuous legend should show actual NDVI or NDRE numbers and state whether the values were clipped to a display range. If one map uses 0.2 to 0.8 and another uses 0.6 to 0.8, identical colours may correspond to different data. A comparison across maps is misleading unless the colour scale and processing are harmonized.
Classification thresholds are also contextual. A researcher may choose a low-index class to target scouting, but that class should not be labelled nutrient deficiency without ground truth. Shade, soil, missing plants, waterlogging and senescence could produce similar values. The map may tell the team where to investigate, not what conclusion to write in advance.
For a useful public explanation, describe the numerical meaning before the emotional meaning of the colours. State, for example, that darker green on the legend denotes higher NDVI under the chosen calibration, rather than automatically calling it healthier soil or higher yield.
A checked example of pixel-area accounting
Suppose a georeferenced UAV raster contains 2,500 valid pixels, each representing a ground square 0.20 metres on a side. The nominal area per pixel is 0.20 × 0.20 = 0.04 square metres. The total nominal sampled area is 2,500 × 0.04 = 100 square metres. If 400 valid pixels lie in a low-index class, their mapped area is 400 × 0.04 = 16 square metres.
The low-index class therefore occupies 16 percent of the valid mapped pixels, not necessarily 16 percent of the crop canopy. Some pixels may be bare soil or mixed canopy. If the raster contains no-data borders, those pixels should not silently enter the denominator. If the mapping grid has been resampled or transformed, the per-cell ground area needs to be determined from the actual coordinate system.
A further check asks whether the pixel size is nominal ground sample distance or the actual chosen output grid. A five-centimetre native image resampled to twenty centimetres has larger map cells, but resampling does not create new independent measurements. A field boundary may also cut through cells, and accurate area accounting may require a polygon-intersection method.
This example illustrates the difference between a digital class percentage and a biological coverage percentage. A label such as “16 percent of the field is diseased” is not justified merely by counting pixels below a spectral threshold. The defensible statement is that 16 square metres of the analysed raster met the defined index class, subject to the masking and georegistration.
Calibration uncertainty and meaningful change
Suppose a sensor produces NDVI 0.62 on one date and 0.65 on another. The measured difference is 0.03 index units. If radiometric calibration, geometric registration and scene variability introduce uncertainty of comparable size, the observed contrast may not support a claim of real canopy change. Significant-looking map colours do not establish statistical or practical significance.
Uncertainty can arise from reflectance calibration, band alignment, sensor noise, illumination mismatch, spatial sampling and choice of mask. Some errors may affect both dates similarly, while others vary independently. It is not always appropriate to combine all errors by simply adding percentages; the way they propagate through the NDVI ratio matters.
For the general index f = (N − R)/(N + R), sensitivity to each input depends on both reflectances. The partial derivative with respect to N is 2R/(N + R) squared; with respect to R it is −2N/(N + R) squared. These sensitivities help explain why errors in red and near-infrared reflectance do not contribute identically. A proper uncertainty calculation also needs covariance if errors are correlated.
A practical field report can use repeated calibration targets, independent plots and withheld ground measurements to quantify reliability rather than relying exclusively on a formal propagation equation. Where the uncertainty has not been measured, describe the difference as an observed spectral contrast requiring confirmation instead of using invented error bars.
Field validation: turn a map into a tested interpretation
A vegetation index is often most useful as a spatial predictor to be checked against a field measurement. Depending on the research question, ground measurements might include leaf chlorophyll readings, crop height, leaf-area index, tissue nitrogen, visual disease scoring or final harvested yield. These are not interchangeable targets. A model validated for one should not automatically be transferred to another.
A robust workflow reserves independent sampling locations for testing. If the same observations are used to select an index, adjust thresholds and report prediction accuracy, the result can be optimistically biased. Cross-validation helps, but geographically nearby measurements may be strongly correlated. A spatially independent test can be more revealing than random sampling within the same homogeneous field.
Ground-truth sample placement matters. Sampling only the greenest and reddest spots may exaggerate a relation while missing ordinary variability. Sampling by management zone, soil texture, slope or other stratification can improve coverage. The report should state when and how the field observations were collected relative to the drone flight.
A useful claim is specific: “Within these sampled plots, calibrated red-edge values were associated with independently measured leaf nitrogen under this crop and season.” It avoids claiming a universal nitrogen formula and invites validation when the crop, sensor, soil background or environment changes.
From spectral pattern to a management decision
A good map can make scouting more efficient. Instead of walking every square metre blindly, a manager may inspect representative areas of unusual index values, compare them with irrigation records or identify locations where sensor values changed over time. Such a decision-support role does not require pretending that spectral colours are diagnostic laboratory results.
If a low-index strip follows a broken irrigation line, it is reasonable to investigate moisture supply. If it matches a flight swath, investigate calibration first. If it follows tractor wheel tracks, compaction may deserve inspection. If it appears only in one processing version, a software or masking choice may be responsible. These patterns create hypotheses, not automatic diagnoses.
Actions with significant cost or risk should use corroborating evidence. Fertilizer application, disease treatment and harvest decisions require more than one unexplained number. The map should shorten the path from broad observation to targeted verification, rather than replacing agronomic judgment with a colour label.
The same reasoning applies to forests, wetlands and restoration projects. A remote-sensing proxy can reveal where spectral behaviour differs, while field surveys determine whether the difference concerns plant cover, species composition, stress or measurement conditions.
Compare camera GCC, UAV NDVI and red-edge correctly
A fixed ordinary RGB camera can produce a PhenoCam green chromatic coordinate, GCC = G/(R + G + B). A multispectral UAV measures selected reflectance bands and may calculate NDVI or NDRE. They share the idea of summarizing optical signals, but their formulas, band responses, viewing geometry and sampling scales differ.
One might see an early-season rise in PhenoCam GCC before a large increase in UAV NDVI, or the reverse, depending on leaf colour, canopy closure, soil visibility and observation frequency. This discrepancy is not proof that the instruments disagree in a technical sense. They may respond to different facets of seasonal development.
A comparison should align dates, vegetation type and field of view, then state which index and measurement definition each graph uses. Rather than collapsing every metric into “vegetation health,” preserve its observational meaning. A reader who can distinguish colour, near-infrared response and red-edge response can interpret the relationships more reliably.
A good set of related readings includes PhenoCam GCC greenness and seasonal thresholds and optical leaf-area index and clumping. They answer different parts of the puzzle: visible seasonality, multispectral response and canopy geometry.
A practical audit of any UAV index chart
Begin with the image source and exact sensor bands. Check whether the values are raw digital numbers, at-sensor radiances or calibrated reflectances. Find out whether the panel and irradiance records support the claimed conversion. Inspect acquisition times, cloud changes, exposure settings and camera geometry.
Next inspect the orthomosaic. Do field patterns coincide with flight swaths? Do bright seams cross crop rows? Are band edges aligned around single leaves and narrow pathways? Is bare soil masked or deliberately included? Are the map grid, projection and pixel area appropriate for the reported spatial precision?
Then check the formula, color legend and statistical aggregation. Were NDVI and NDRE calculated using compatible band values and meaningful denominators? Was a pixel classified by a threshold chosen before examining the results, or selected afterward to create a persuasive map? Are differences stated in index units rather than invented biomass percentages?
Finally compare with independent evidence. A useful interpretation identifies what the sensor found, why the pattern may matter and how it was checked. If those steps have not been completed, a cautious scientific claim is stronger than a dramatic diagnosis.
Transfer problem: a changing red band
A hypothetical plot is measured twice, and its near-infrared reflectance stays at 0.48. On the first date red reflectance is 0.12, giving NDVI 0.36/0.60 = 0.600. On the second date red reflectance is 0.16, giving NDVI 0.32/0.64 = 0.500. The observed index difference is minus 0.100 units. Which measurement actually changed? Only the recorded red reflectance in this simplified example; the biological cause is not specified.
A student might conclude that plant health fell by ten percent. That is incorrect twice: the index changed by 0.100 on its own scale, and there is no validated map from that change to a percentage of health. An alternative explanation might be an illumination or calibration difference. Independent leaf observations, matched calibration and field records would help determine whether leaf pigments or another canopy property changed.
A further check asks what happens if both red and NIR reflectance double. With compatible linear, offset-free scaling, NDVI remains the same because numerator and denominator double. If only one band doubles, NDVI changes. This is why a single common gain error behaves differently from a band-specific calibration error. The distinction helps diagnose sensor problems without treating every index shift as a crop response.
Transfer problem: area and denominators
Two UAV maps cover the same 1,000-square-metre field. In map A, valid imagery covers 900 square metres, with 180 square metres classified low NDVI. In map B, valid imagery covers only 600 square metres, with 150 square metres classified low NDVI. A naive report says the low area dropped from 180 to 150 square metres, suggesting improvement. But the valid observation footprint has also shrunk.
As a fraction of valid area, map A has 180/900 = 20 percent low-index coverage, while map B has 150/600 = 25 percent. That does not prove deterioration either: the missing 400 square metres may contain systematically different vegetation, and the two rasters may differ in masks or conditions. The simple comparison becomes ambiguous once the denominator changes.
A defensible comparison might restrict both maps to their overlapping valid area and apply a consistent index threshold, then report both percentage and area. If those areas are not spatially representative of the whole field, say so. Never announce a fieldwide benefit from a changing set of observed pixels without explaining what was excluded. The lesson transfers directly to satellite maps, classroom data tables and public dashboards.
Transfer problem: choose the right index, not the fashionable one
Imagine a dense crop with near-infrared reflectance 0.56, red reflectance 0.06 and red-edge reflectance 0.28. The NDVI is 0.50/0.62, approximately 0.8065. Using the defined NDRE formula gives 0.28/0.84, approximately 0.3333. The numbers have different scales because the red and red-edge bands measure different spectral regions. Calling 0.8065 “better” than 0.3333 would be meaningless.
Suppose the next flight shows NDVI 0.817 and NDRE 0.370. The red-edge result has moved by approximately 0.037 whereas NDVI has moved by approximately 0.011. It may be tempting to say NDRE detected a meaningful biological response that NDVI missed. Before doing so, check the calibration, scene geometry, measurement noise, ground sampling and actual quantity being predicted. Larger numeric movement does not necessarily equal better predictive validity.
The independent task is to propose a fair field test. Choose a specific measurable target such as laboratory leaf chlorophyll, sample plots across the full range, obtain synchronized UAV data, fit a calibration on part of the plots and evaluate both indices on withheld plots. The stronger index is the one that supports the target decision with credible independent accuracy, not the one whose colour map looks more dramatic.
How to repair a misleading conclusion
Original sentence: “The red zones are diseased crops because drone NDVI fell.” That claims a diagnosis the index alone cannot make. A corrected version is: “The calibrated UAV map identified zones with lower NDVI under the chosen threshold; field checks are needed to determine whether reduced canopy cover, plant stress, soil exposure or image artifacts explain them.” This preserves the observation while postponing an unsupported cause.
Original sentence: “NDRE showed 28 percent chlorophyll.” A corrected version is: “The selected red-edge index was 0.28, a dimensionless ratio of two calibrated bands; its relationship with chlorophyll must be evaluated against leaf measurements.” This repair matters because a student can complete a correct formula and still describe the result incorrectly.
Original sentence: “This year’s plants are 15 percent healthier because the map is greener.” A repaired statement names the dates, sensor, comparable acquisition conditions, actual index difference, uncertainty and field validation. It might conclude only that the vegetation produced a stronger spectral contrast on the second date. Such caution is not weakness; it tells the reader what the evidence can sustain.
Original sentence: “The sensor was calibrated because a white board appeared in one image.” A white object is not necessarily a measured reflectance reference. A correct methods statement identifies a suitable reference panel, band-specific reference characteristics, collection conditions and the radiometric processing applied. Where those records are missing, the resulting map may be useful for exploration but should not be described as rigorously calibrated reflectance.
Frequently asked questions about drone vegetation maps
Is NDVI a direct plant-health measurement? No. It is a red and near-infrared spectral ratio that can be associated with vegetation cover and other plant properties. Its biological interpretation depends on canopy, site, season, sensor and independent validation. It can efficiently identify patterns that deserve inspection but cannot diagnose every stressor.
Why use a red-edge camera rather than an ordinary RGB drone? A red-edge band adds spectral information unavailable from standard visible RGB imagery, and near-infrared reflectance is needed for conventional NDVI. RGB photographs can still provide valuable maps of visible crop cover and colour. The extra bands become useful when a specific task benefits from their spectral response and the acquisition is calibrated.
Will flying lower always improve accuracy? Lower flight altitude may make smaller features visible, but it can increase the number of images and sensitivity to motion, overlap problems and processing requirements. A tiny ground sample distance does not compensate for bad calibration or poor registration. Accuracy must be considered across spectral, geometric and biological parts of the workflow.
Can a map from a cloudy day be compared with a sunny-day map? Possibly, with appropriate radiometric methods, consistent calibration, acquisition metadata and uncertainty checks. A normalized index does not guarantee invariance under changing spectral illumination. If an apparent change matches a cloud boundary or swath seam, investigate that before attributing it to vegetation.
Does NDRE give a universal nitrogen rate prescription? No. Nitrogen recommendations depend on crop growth stage, soil supply, cultivar, expected yield, management, climate and validation. An index may enter a locally validated decision framework, but it is not by itself a fertilizer recommendation. Large management actions need independent agronomic evidence.
What should a report preserve for future analysts? Preserve raw images where permitted, camera and band metadata, calibration-panel records, acquisition time, irradiance information, ground control, geometric products, masks, processing versions, index formulas, independent ground measurements and the exact map legend. These records make a result reproducible and allow a different analyst to test whether a spectacular colour pattern represents a real environmental signal.
Primary sources and methodological references
NASA Earth Observatory: how NDVI uses red and near-infrared
NASA Harmonized Landsat and Sentinel-2 vegetation indices
USGS measurements of UAS radiometric calibration-panel reflectance
USGS field setup for multispectral camera calibration-panel imaging