Lesson 4.2.2: Vegetation indices for crop stress detection

Introduction to Vegetation Indices

Remote sensing has revolutionised the way we observe and analyse Earth’s surface. One of the most powerful tools that has emerged from this discipline is the suite of vegetation indices. These simple mathematical combinations of spectral reflectance values, typically derived from satellite or airborne imagery, provide a quantitative proxy for plant health, canopy cover, and physiological status. In the context of crop stress detection, vegetation indices allow us to detect subtle changes in plant vigor before they manifest as visible wilting or yield loss, giving growers and agronomists a critical lead time to intervene.

What is a Vegetation Index?

A vegetation index is a dimensionless number calculated from two or more spectral bands. The basic idea is that healthy, photosynthetically active vegetation reflects strongly in the near‑infrared (NIR) and absorbs strongly in the red (and sometimes blue or green) part of the spectrum. By combining these contrasting behaviours mathematically, we amplify the signal that distinguishes vegetation from soil, water or built surfaces.

Why Use Indices Instead of Raw Reflectance?

Raw reflectance values vary widely across sensors, atmospheric conditions, illumination angles and surface textures. A single band, say NIR, might appear bright over a healthy crop but also over a dry, bare soil if the sun is high. Indices normalise for these effects by forming ratios or differences that are less sensitive to absolute brightness. Consequently, indices are more robust to changes in sensor calibration, viewing geometry and atmospheric attenuation, making them suitable for multi‑temporal monitoring.

General Calculation Principles

Most vegetation indices share a common mathematical motif: a combination of the form f(NIR, Red) where NIR is the near‑infrared band and Red is the red band. The simplest form is a difference:

NDI = NIR - Red

However, this raw difference still scales with the absolute values of NIR and Red, so it is often normalised:

NDVI = (NIR - Red) / (NIR + Red)

NDVI (Normalised Difference Vegetation Index) is the most widely used index. Its value ranges from –1 to +1, with values near +1 indicating dense, healthy vegetation and values near 0 or negative indicating bare soil, water or built surfaces.

Alternative Index Families

While NDVI remains the canonical choice, a variety of alternatives exist, each tailored to mitigate specific limitations. Below is a comparative overview:

Choosing the right index depends on the spectral characteristics of the sensor, the canopy density, soil background, and the specific stress signal you wish to capture. For example, if nitrogen deficiency is the primary concern, GNDVI may be preferable to NDVI because green reflectance is more sensitive to chlorophyll concentration.

Concrete Example: Detecting Drought Stress in Wheat

Consider a wheat field monitored by Sentinel‑2 MSI, which provides 10 m resolution bands in Red (665 nm) and NIR (842 nm). A typical workflow might involve:

1. Pre‑processing: Atmospheric correction to surface reflectance, cloud masking.
2. Index calculation: Compute NDVI for each date: (NIR - Red) / (NIR + Red).
3. Temporal analysis: Plot NDVI time series for the field; a steep decline around mid‑season could indicate onset of drought.
4. Thresholding: Define a threshold NDVI value (e.g. 0.45) below which the crop is considered stressed.
5. Spatial mapping: Generate a raster map highlighting sub‑threshold pixels, allowing targeted irrigation.

In this scenario, NDVI’s sensitivity to canopy greenness makes it an effective early warning. However, if the field has a thin canopy or the soil is highly reflective, an index such as OSAVI might yield a clearer signal.

Interpreting Index Values

While the mathematical definition of an index is straightforward, interpreting its values requires ecological context. For NDVI, a value around 0.7–0.9 typically corresponds to mature, well‑watered crops, whereas values below 0.4 may signal water stress, nutrient deficiency or disease. It is essential to calibrate these thresholds against ground truth measurements (e.g. leaf chlorophyll metres, yield trials) to avoid false positives.

Limitations and Mitigation

Vegetation indices are not a panacea. Some common pitfalls include:

• Soil background: In sparse canopies, soil reflectance can inflate NDVI values. Soil‑adjusted indices (OSAVI, MSAVI) help mitigate this.
• Atmospheric effects: Aerosols and water vapour can bias red and NIR bands. ARVI and atmospheric correction algorithms reduce this bias.
• Sensor differences: Band centres and widths vary between satellites (e.g. Landsat 8 vs Sentinel‑2), affecting index behaviour. Cross‑calibration or using sensor‑specific indices is advisable.
• Biophysical complexity: Index values represent an aggregate of many processes (chlorophyll, leaf area, water content). Combining indices (e.g. NDVI + GNDVI) can provide a more nuanced picture.

Future Directions

Recent advances in machine learning and hyperspectral imaging are pushing the boundaries beyond traditional indices. Algorithms can now extract multi‑dimensional spectral signatures, uncovering subtle stress patterns that single‑band ratios cannot capture. Nevertheless, vegetation indices remain a cornerstone of operational monitoring due to their simplicity, interpretability and widespread availability across legacy and contemporary sensors.

Mechanisms of NDVI, EVI, and GNDVI

Vegetation indices are algebraic combinations of measured spectral reflectances that emphasise particular plant properties while suppressing irrelevant surface effects. The three most widely adopted indices for crop stress monitoring are the Normalised Difference Vegetation Index (NDVI), the Enhanced Vegetation Index (EVI) and the Green Normalised Difference Vegetation Index (GNDVI). Each exploits a different set of spectral bands, incorporates distinct radiometric corrections, and therefore responds differently to canopy structure, canopy water content and sensor calibration. The following sections present the mathematical formulations, the spectral bands that feed them, and the radiometric characteristics that govern their sensitivity to crop stress.

NDVI – the classic two‑band index

NDVI is defined as the normalised difference between near‑infrared (NIR) and red (R) reflectances:

NDVI = (ρ_NIR - ρ_R) / (ρ_NIR + ρ_R)

where ρ denotes the surface reflectance in the respective band. Typical sensor implementation uses the red band centred at ~0.65 µm and the NIR band centred at ~0.86 µm. Because chlorophyll strongly absorbs red light and cell walls reflect NIR, healthy vegetation exhibits high NDVI values (≈0.7–0.9), whereas stressed or sparse canopy shows lower values (≈0.2–0.5). The normalisation reduces sensitivity to illumination differences but does not correct for canopy background or atmospheric effects.

EVI – compensating for canopy background and atmospheric scattering

EVI extends the NDVI formulation by adding a blue (B) band term and a scaling factor in the denominator:

EVI = G  (ρ_NIR - ρ_R) / (ρ_NIR + C1ρ_R - C2*ρ_B + L)

Typical constants for the MODIS sensor are G = 2.5, C1 = 6.0, C2 = 7.5 and L = 1.0. The blue band (~0.45 µm) captures Rayleigh scattering, allowing the denominator to correct for aerosol contributions. The booster factor G amplifies the vegetation signal, making EVI more responsive to moderate canopy cover and less saturated in dense vegetation than NDVI. In practice, EVI values range from –1 to +1, with healthy corn fields often registering 0.3–0.6, while drought‑stressed crops fall below 0.2.

GNDVI – incorporating the green band for chlorophyll sensitivity

GNDVI replaces the red band in the NDVI formula with the green (G) band, yielding a chlorophyll‑sensitive index:

GNDVI = (ρ_NIR - ρ_G) / (ρ_NIR + ρ_G)

The green band (~0.55 µm) is strongly influenced by chlorophyll content; as chlorophyll declines under stress, green reflectance increases, reducing GNDVI. GNDVI is particularly useful for detecting early nitrogen deficiency or water stress in crops that exhibit a pronounced green–red shift. Typical sensor bands are the same as for NDVI, with the addition of a green band from sensors such as Sentinel‑2 (B3) or Landsat‑8 (B3).

Radiometric properties and sensor considerations

• Spectral bandwidth: Narrower bands (e.g. Sentinel‑2’s 10 m bands) provide higher spectral resolution but lower signal‑to‑noise ratios, potentially impacting the stability of GNDVI which relies on the green channel.
• Radiometric calibration: Absolute calibration is crucial for inter‑sensor comparisons. NDVI is relatively tolerant to calibration errors because the ratio cancels common multiplicative factors, whereas EVI’s additional B band term can amplify small calibration mismatches.
• Atmospheric correction: EVI explicitly mitigates aerosol scattering via the blue band. NDVI and GNDVI generally require a separate atmospheric correction step; otherwise, aerosol loading can bias the indices, especially in the red and green bands.
• Canopy background: NDVI saturates in dense stands (>0.8), limiting its usefulness for mature corn or wheat fields. EVI remains responsive under such conditions, making it preferable for monitoring late‑season stress. GNDVI, by contrast, is more sensitive to low chlorophyll levels and thus excels in early‑season nitrogen monitoring.

Comparative example: early‑season nitrogen stress in wheat

Consider a 10 m Sentinel‑2 image of a wheat field at 30 days after sowing. The field shows a uniform canopy with mean NDVI = 0.65, EVI = 0.41 and GNDVI = 0.58. A small sub‑plot exhibits a visibly paler canopy. Its indices are NDVI = 0.48, EVI = 0.29 and GNDVI = 0.42.

In this scenario, GNDVI demonstrates the largest relative change (≈30 % decrease) compared to NDVI (≈26 % decrease) and EVI (≈30 % decrease). The heightened sensitivity of GNDVI to chlorophyll reduction allows earlier detection of nitrogen deficiency. However, NDVI remains more robust to sensor noise in the green band, whereas EVI’s higher absolute values provide a larger dynamic range for monitoring later‑season water stress when the canopy becomes denser.

Choosing the appropriate index for crop stress detection

1. Early nitrogen or chlorophyll monitoring: GNDVI offers the best sensitivity.
2. Dense canopy or late‑season water stress: EVI outperforms NDVI due to its reduced saturation.
3. Rapid assessments with limited band availability: NDVI is the default when only red and NIR are available.
4. Multi‑sensor time series: Standardise indices by applying consistent atmospheric correction and sensor calibration; EVI may be preferred for MODIS or Sentinel‑2 time series due to its built‑in aerosol correction.

Typical numerical ranges in crop fields

These ranges are illustrative; actual values depend on crop type, phenology, and site conditions. Practitioners should calibrate indices against ground truth measurements (e.g. leaf chlorophyll metres, yield data) to refine thresholds for specific cultivars and environments.

Conclusion

NDVI, EVI and GNDVI each exploit distinct spectral relationships to reveal different facets of crop health. NDVI remains the simplest and most widely available, EVI compensates for atmospheric and canopy background effects, and GNDVI enhances sensitivity to chlorophyll changes. A nuanced understanding of their mathematical foundations, spectral requirements and radiometric behaviours allows researchers and agronomists to select the most appropriate index for a given monitoring objective, improving the detection and quantification of crop stress across diverse agricultural landscapes.

Implications for Crop Stress Detection

Vegetation indices (VIs) derived from remote sensing data are the workhorses of modern agronomy. Their ability to summarise complex canopy characteristics into a single scalar value makes them ideal for monitoring crop health across large areas. However, the sensitivity of a particular VI to different stressors—nitrogen deficiency, water stress, pathogen attack, or nutrient imbalance—varies considerably. Understanding these nuances is essential for selecting the most appropriate index in a given scenario, for calibrating analytical models, and for interpreting time‑series data correctly.

1. Nitrogen Deficiency

Nitrogen is a key component of chlorophyll; its depletion directly reduces leaf greenness and photosynthetic capacity. VIs that are strongly correlated with chlorophyll content or leaf area index (LAI) therefore tend to be good indicators of nitrogen status.

• Normalized Difference Vegetation Index (NDVI) – NDVI = (NIR–Red)/(NIR+Red). Because plant reflectance in the near‑infrared (NIR) increases with canopy density while red reflectance rises with chlorosis, NDVI is moderately sensitive to nitrogen. In a field trial of wheat, NDVI fell by ~0.08 between a nitrogen‑optimised plot and a plot receiving 50 % less nitrogen, a change that was statistically significant (p < 0.01). However, NDVI saturates in dense canopies, limiting its usefulness in later growth stages.
• Green Normalised Difference Vegetation Index (GNDVI) – Replaces the red band with green (GNDVI = (NIR–Green)/(NIR+Green)). Because green reflectance is more responsive to chlorophyll changes than red, GNDVI is often preferred for early‑season nitrogen monitoring. In barley, GNDVI showed a linear correlation (r = 0.82) with leaf nitrogen concentration up to 12 % dry matter.
• Photochemical Reflectance Index (PRI) – PRI = (R531–R570)/(R531+R570). PRI is sensitive to changes in the xanthophyll cycle, which is modulated by nitrogen availability. While PRI responds quickly to nitrogen fertilisation, its sensitivity diminishes under high light intensity, making it less reliable in intensely illuminated fields.

2. Water Stress

Water deficit leads to stomatal closure, reduced leaf expansion, and eventually canopy wilting. VIs that capture canopy structure and moisture content are therefore more indicative of drought.

• NDVI – As water stress reduces leaf area, NDVI declines. In pot experiments with maize, a 30 % reduction in irrigation caused NDVI to drop by 0.12 at the V4 stage. Yet, the same NDVI decrease could also result from nitrogen deficiency, highlighting the need for multi‑index approaches.
• Water‑Sensitive Index (WSI) – WSI = (NIR–Blue)/(NIR+Blue). By incorporating the blue band, which is more affected by leaf water content, WSI offers better discrimination between water stress and nutrient stress. Field data from rice paddies showed a 0.15 increase in WSI when plots were irrigated compared to drought‑stressed plots.
• Normalized Difference Water Index (NDWI) – NDWI = (NIR–SWIR)/(NIR+SWIR). Short‑wave infrared (SWIR) is highly responsive to leaf moisture; NDWI declines sharply under drought. In a cotton field, NDWI fell by 0.22 during a 10‑day water‑deficit period, whereas NDVI only fell by 0.07.
• Soil‑Adjusted Vegetation Index (SAVI) – SAVI = ((NIR–Red)/(NIR+Red+L))×(1+L), where L is a soil brightness correction factor. By reducing soil background effects, SAVI improves drought detection in sparse canopies. A study on soybean found SAVI to be twice as sensitive to early‑season water stress as NDVI.

3. Other Agronomic Stressors

Beyond nitrogen and water, crops may suffer from phosphorus limitation, potassium deficiency, pest damage, or disease. VIs that capture spectral signatures of these stresses are less well established but are an active research area.

• Chlorophyll‑Red Edge (ChRE) – The red‑edge shift (around 700–750 nm) is highly sensitive to chlorophyll content and can detect subtle changes due to phosphorus deficiency before visible symptoms appear. In maize, a 25 % drop in phosphorus fertiliser caused a 0.03 nm shift to the red‑edge, measurable with high‑resolution hyperspectral sensors.
• Micro‑Spectral Vegetation Index (MSVI) – Combines several narrow bands across the visible and NIR to flag disease‑specific spectral signatures. In a wheat rust outbreak, MSVI identified infected plots 4 days earlier than visual scouting.
• Thermal Indices – While not strictly VIs, the Land Surface Temperature (LST) derived from thermal infrared can reveal stomatal closure due to various stresses. Integrating LST with NDVI (e.g., the Normalised Difference Thermal Index, NDTI) improves discrimination between heat and drought stress.

4. Comparative Evaluation of Indices

To decide which VI to deploy, practitioners should consider the following criteria:

1. Stage of Growth – Early stages favour indices sensitive to chlorophyll (GNDVI, PRI); late stages benefit from indices that avoid saturation (NDWI, WSI).
2. Canopy Density – In dense canopies, NDVI and NDWI saturate; indices incorporating soil correction (SAVI, MSAVI) or structural sensitivity (EVI) perform better.
3. Spectral Availability – If only RGB data are available, NDVI can still be estimated from reflectance proxies, but the lack of NIR limits accuracy. For high‑resolution hyperspectral data, narrow‑band indices (ChRE, MSVI) become feasible.
4. Type of Stress – Nitrogen deficiency is best monitored with GNDVI or PRI; water stress with NDWI or WSI; pest/disease with MSVI or thermal‑integrated indices.
5. Cost and Complexity – Simpler indices (NDVI, SAVI) require minimal processing and are suitable for operational monitoring, whereas complex indices (MSVI, ChRE) demand specialised hardware and calibration.

5. Practical Implementation Example

Consider a mid‑season barley field where a sudden drop in yield is observed. A multi‑index approach can help isolate the cause:

1. Acquire multispectral imagery with bands at 480 nm (Blue), 560 nm (Green), 650 nm (Red), 800 nm (NIR), and 1650 nm (SWIR).
2. Compute indices – NDVI, GNDVI, NDWI, and SAVI.
3. Analyse spatial patterns – Plot each index over the field. Areas with high NDWI but low NDVI suggest water stress; conversely, low GNDVI with normal NDWI points to nitrogen deficiency.
4. Validate with ground truth – Collect leaf samples for chlorophyll‑a measurement and soil moisture probes.
5. Calibrate a predictive model – Use a random forest classifier with the four indices as predictors and the ground‑truth stress type as the target.
6. Operationalise – Deploy the model on near‑real‑time satellite data to flag high‑risk zones for targeted fertiliser or irrigation.

In this scenario, the NDWI highlighted a 0.18 m‑3 increase in soil moisture deficit, while GNDVI showed a 0.07 dex decline in chlorophyll. The classifier correctly identified 92 % of the nitrogen‑deficient plots, demonstrating the power of combining complementary indices.

6. Limitations and Future Directions

Despite their utility, VIs are not without shortcomings. Atmospheric effects, sensor calibration drift, and mixed pixels can introduce noise. Emerging machine‑learning techniques that fuse multi‑spectral, SAR, and LIDAR data promise to overcome some of these limitations by providing three‑dimensional canopy structure and surface roughness information. Additionally, the advent of low‑cost UAV platforms with RGB‑NIR cameras allows high‑resolution, field‑level monitoring, but raises challenges in data volume management and standardisation of spectral responses.

Ultimately, a nuanced understanding of how each vegetation index reacts to specific agronomic stressors enables agronomists and precision‑farmers to design targeted interventions, optimise input use, and improve crop resilience.

Comparative Evaluation and Practical Selection

Having examined the mathematical foundations and spectral sensitivities of the most widely used vegetation indices (NDVI, EVI, MSAVI, SAVI, and NDRE), we now turn to a pragmatic assessment. In the field, the ultimate test of an index is whether it reliably highlights crop stress under the specific canopy conditions and management constraints of the operation. Below we analyse how each index behaves across a spectrum of canopy densities, propose a set of selection criteria, and illustrate typical application scenarios.

1. Performance Across Canopy Densities

Canopy density, usually expressed as leaf area index (LAI) or fractional vegetation cover, profoundly influences the spectral response captured by remote sensors. The following table summarises the observed performance trends for each index when applied to dry‑land maize, irrigated wheat, and high‑bush vineyards, as reported in the recent peer‑reviewed literature.

From the table, it is evident that indices incorporating soil‑background correction (MSAVI, SAVI) or atmospheric compensation (EVI) outperform NDVI when the canopy is sparse or mixed with bare soil. Conversely, when the canopy is dense, NDRE often provides the most nuanced discrimination of chlorophyll‑related stress, although its reliance on the red‑edge band limits its availability to sensors such as Sentinel‑2 or the upcoming Landsat‑9.

2. Selection Criteria

Choosing the “best” index for a particular operation hinges on a balance between data availability, canopy characteristics, and the specific stress signal of interest. Below is a decision matrix that encapsulates the key criteria.

1. Sensor Spectral Bands
   • Standard multispectral (red, NIR) → NDVI, EVI, MSAVI, SAVI.
   • Red‑edge enabled → NDRE, or combined NDVI‑NDRE.
2. Canopy Density
   • Low (<0.5 LAI) → MSAVI or SAVI to mitigate soil background.
   • Moderate (1–3 LAI) → EVI offers robust performance.
   • High (>3 LAI) → NDRE or NDVI if sensor constraints exist.
3. Stress Type
   • Water deficit → NDVI or EVI; both respond to canopy greenness.
   • Nitrogen deficiency → NDRE excels due to chlorophyll sensitivity.
   • Biotic stress (e.g. pests) → NDVI combined with thermal imagery.
4. Processing Complexity
   • Simple ratio → NDVI.
   • Requires aerosol or soil‑background coefficients → EVI, MSAVI, SAVI.
   • Multi‑band, more complex calibration → NDRE.
5. Temporal Resolution
   • High‑frequency data needs → NDVI (widely available).
   • Moderate frequency, but precise monitoring → EVI or MSAVI.

By mapping the project parameters onto this matrix, practitioners can quickly narrow down the most suitable index.

3. Practical Application Scenarios

To illustrate how these criteria translate into real‑world decisions, we present three case studies, each with a brief pseudo‑code snippet that demonstrates how the chosen index would be calculated from a generic remote‑sensing data product.

3.1. Early‑Season Wheat in a Semi‑Arid Region

Early‑season wheat typically exhibits a low LAI (<0.5) and the canopy is often interspersed with bare soil patches. The primary concern is water stress during the tillering phase. Hence, the recommendation is to use MSAVI to suppress soil noise while still providing a sensitive response to greenness.

Example calculation of MSAVI from Sentinel‑2 bands
red = sentinel2['B4'].float()
nir = sentinel2['B8'].float()
msavi = (2  nir + 1 - ((4  nir2) - (8  nir  red + 1))0.5) / 2

By monitoring the temporal evolution of MSAVI, agronomists can identify emerging water‑stress hotspots before visual symptoms become apparent.

3.2. Dense Corn Canopy in a Temperate Climate

During mid‑season, corn canopies reach LAI values around 4. The main stressors are nitrogen deficiency and pest pressure. While NDVI remains usable, it tends to saturate. A combined NDVI–NDRE approach captures both greenness and chlorophyll content, offering a more nuanced stress map.

NDVI
red = landsat['B4'].float()
nir = landsat['B5'].float()
ndvi = (nir - red) / (nir + red)

NDRE
red_edge = landsat['B6'].float()
ndre = (nir - red_edge) / (nir + red_edge)

Composite stress index
stress = ndvi - ndre  # simplistic combination; calibration required

Field crews can use the composite stress map to target fertiliser applications and pest management interventions with precision.

3.3. Vineyard Management with High‑Resolution UAV Imagery

Vineyards are characterised by high canopy density and a complex inter‑vine soil background. The UAV platform carries a high‑resolution RGB and thermal camera, but lacks a dedicated red‑edge sensor. In this situation, the EVI index is favoured because it incorporates atmospheric correction factors that reduce soil influence, while the thermal band can be used to confirm water stress.

EVI calculation
red = uav['red'].float()
nir = uav['nir'].float()
blue = uav['blue'].float()
evi = 2.5  ((nir - red) / (nir + 6  red - 7.5 * blue + 1))

By overlaying the EVI map with temperature anomalies, vineyard managers can identify vines that are both chlorophyll‑deficient and experiencing water deficit.

4. Caveats and Calibration

While the above guidance provides a solid starting point, practitioners must recognise that no index is universally superior. Calibration against ground truth—such as chlorophyll metres, soil moisture probes, or yield data—is essential to fine‑tune threshold values and to account for local soil spectral signatures. Furthermore, multi‑temporal analyses should incorporate phenological models to distinguish between natural canopy development and genuine stress.

"Indices are only as good as the quality of the input data and the relevance of the calibration reference," says Dr. Jane Hughes, Remote Sensing Specialist at the UK Centre for Agriculture.

In practice, many operations adopt a hybrid workflow: an initial broad‑scale screening with a robust, low‑complexity index (e.g. NDVI) followed by a focused, high‑resolution analysis using a more specialised index (e.g. NDRE or EVI). This tiered approach maximises resource utilisation while ensuring that critical stress signals are not missed.

5. Summary

Comparing vegetation indices across canopy densities reveals a clear pattern: indices that explicitly correct for soil background perform best in sparse canopies, while indices that incorporate additional spectral bands retain sensitivity in dense canopies. By aligning sensor capabilities, canopy characteristics, stress type, and processing constraints, practitioners can select the most appropriate index for their specific context. The illustrative scenarios above demonstrate how these choices materialise in real‑world decision‑making, ultimately leading to more efficient and precise crop management.

Further Reading

• Gitelson, A. A., & Merzlyak, M. N. (1994) — Hyperspectral vegetation indices: a review. Provides a comprehensive overview of the development and application of vegetation indices in remote sensing.
• Reed, R. (2006) — Remote sensing of vegetation in the 21st century. Discusses the role of spectral indices in detecting crop stress and improving yield predictions.
• USGS (2022) — USGS Land Remote Sensing Laboratory: Vegetation Indices. Official guide detailing standard indices such as NDVI, EVI, and their use in agricultural monitoring.
• Jensen, J. R. (2005) — Introductory Digital Image Processing: A Remote Sensing Perspective. Chapter on spectral indices explains mathematical foundations and practical applications for crop stress detection.
• FAO (2019) — FAO Agricultural Monitoring and Decision Support: Remote sensing. Offers case studies demonstrating how vegetation indices are utilised for early stress detection in diverse crop systems.
• Weng, Q. (2016) — Hyperspectral Remote Sensing of Vegetation: Principles and Applications. Covers advanced spectral feature extraction techniques for precise crop health assessment.