Vegetation Indices Compared: NDVI vs EVI vs SAVI

Vegetation indices are mathematical combinations of spectral reflectance values measured by satellites or sensors, designed to enhance the signal from vegetation while suppressing confounding factors like soil brightness, atmospheric effects, and illumination geometry. Raw spectral band values — the actual reflectance percentages measured by Sentinel-2 at specific wavelengths — contain vegetation information, but it is mixed with other signals. Vegetation indices isolate the vegetation component, making them far more reliable indicators of crop health, biomass, and physiological status. The choice of vegetation index directly affects the quality of agricultural decisions. Using the wrong index for your conditions — for example, applying NDVI to a field with bright, sandy soil or during early growth stages with low canopy cover — can lead to inaccurate assessments and misguided management actions. Different indices are optimized for different conditions: some excel at high biomass, others at low canopy cover, and others in atmospherically challenging situations. Understanding these trade-offs allows farmers and agronomists to select the most informative index for each specific situation. Over 150 vegetation indices have been proposed in the scientific literature, but only a handful are routinely used in agriculture. This article focuses on the three most practical indices for crop monitoring with Sentinel-2 data: NDVI, EVI, and SAVI. Together, these three indices cover the major use cases encountered in European arable farming, from early-season crop establishment monitoring to peak-season biomass assessment.

The Normalized Difference Vegetation Index (NDVI) is calculated as (NIR - Red) / (NIR + Red), where NIR is near-infrared reflectance (Sentinel-2 Band 8, 842 nm) and Red is red reflectance (Band 4, 665 nm). Values range from -1 to +1, with healthy vegetation typically producing values between 0.3 and 0.9. NDVI exploits the fundamental spectral contrast of green vegetation: chlorophyll absorbs red light for photosynthesis while leaf mesophyll cells scatter near-infrared light. The normalization (dividing by the sum) reduces the effects of overall illumination intensity, making NDVI comparable across different sun angles and atmospheric conditions. NDVI's greatest strength is its simplicity and universal applicability. It requires only two spectral bands available on every optical satellite sensor, it is well-understood after 50 years of research, and extensive look-up tables exist relating NDVI values to crop parameters like leaf area index, biomass, and yield potential. For routine crop health monitoring during the active growth phase (when canopy cover exceeds 40-50%), NDVI provides reliable, actionable information. The vast majority of agricultural remote sensing applications worldwide use NDVI as their primary index. However, NDVI has two well-known limitations. First, it saturates at high leaf area index (LAI above approximately 3-4), meaning it cannot distinguish between a good crop and an excellent crop once the canopy is dense. A wheat field with LAI 4 and a wheat field with LAI 7 may both show NDVI of 0.85, despite the second field having significantly more biomass. Second, NDVI is sensitive to soil background brightness, meaning that the same sparse crop growing on dark, organic-rich soil will show a higher NDVI than the same crop on bright, sandy soil. This soil sensitivity is particularly problematic during early growth stages when significant soil is visible between crop rows.

The Enhanced Vegetation Index (EVI) was developed specifically to overcome NDVI's saturation and atmospheric sensitivity issues. The formula is EVI = G * (NIR - Red) / (NIR + C1 * Red - C2 * Blue + L), where G is a gain factor (2.5), C1 and C2 are atmospheric correction coefficients (6.0 and 7.5 respectively), L is a canopy background adjustment (1.0), and Blue is the blue band reflectance (Sentinel-2 Band 2, 490 nm). The inclusion of the blue band provides atmospheric correction capability because blue light is most affected by aerosol scattering, and the additional parameters reduce soil background influence. EVI's primary advantage over NDVI is its superior performance in high-biomass conditions. Where NDVI saturates around LAI 3-4, EVI maintains sensitivity up to LAI 6-8, making it the preferred index for crops with dense canopies like corn at full development, sugar beet, or rapeseed before flowering. Research comparing NDVI and EVI for corn yield prediction consistently shows that EVI explains 10-20% more yield variance than NDVI during the reproductive growth phase. For precision nitrogen management in high-yield situations, this additional sensitivity translates directly into better rate recommendations. The atmospheric correction built into EVI's formula makes it more stable across time series than NDVI, which can show artificial variation due to changing atmospheric conditions between satellite passes. This stability is valuable for monitoring seasonal crop trajectories and detecting genuine changes in crop condition. However, EVI's complexity is also a drawback: it requires three spectral bands instead of two, the fixed coefficients (C1, C2, G, L) are optimized for the MODIS sensor and may not be perfectly calibrated for Sentinel-2, and EVI values are less intuitive to interpret than NDVI values for users without remote sensing training. Typical EVI values for crops range from 0.1 (bare soil) to 0.6 (dense vegetation), a narrower range than NDVI.

The Soil Adjusted Vegetation Index (SAVI) addresses NDVI's soil background sensitivity through a simple but effective modification. The formula is SAVI = (NIR - Red) / (NIR + Red + L) * (1 + L), where L is a soil brightness correction factor. The value of L ranges from 0 (equivalent to NDVI, suitable for dense vegetation) to 1 (maximum soil correction, suitable for very sparse vegetation), with 0.5 commonly used as a general-purpose compromise. By adjusting L based on vegetation density, SAVI reduces the soil background influence that distorts NDVI in low-cover conditions. SAVI is most valuable during early growth stages (crop establishment, tillering, early stem elongation) when canopy cover is below 30-40% and soil spectral properties significantly influence index values. At these growth stages, SAVI provides a more accurate representation of actual vegetation amount than NDVI. This is particularly important for monitoring crop emergence uniformity, detecting establishment failures, and assessing early-season crop vigor when management interventions (re-seeding, fertilizer adjustment) are still possible. In the German context, SAVI is especially useful for spring crops like sugar beet, corn, and spring barley during May and June when rows are visible and inter-row soil is exposed. A key practical limitation of SAVI is the L factor, which ideally should be adjusted based on actual vegetation density — but knowing the vegetation density requires the index value, creating a circular dependency. In practice, L = 0.5 provides good results across a wide range of conditions and is the standard default. Modified versions like MSAVI (Modified SAVI) use iterative or self-adjusting approaches to optimize L automatically. For routine agricultural monitoring, however, the simple SAVI with L = 0.5 is often sufficient, and the improvement over NDVI in low-cover conditions justifies the marginal additional complexity.

When comparing these three indices across key performance dimensions, clear patterns emerge. For sensitivity at high biomass (LAI above 4), EVI ranks best, SAVI is moderate, and NDVI is poorest due to saturation. For soil background resistance, SAVI and EVI both outperform NDVI, with SAVI specifically designed for this purpose. For atmospheric stability, EVI is best due to its built-in blue-band correction, while NDVI and SAVI are equally susceptible to atmospheric noise. For simplicity and interpretability, NDVI leads, SAVI is moderately complex, and EVI is most complex. Computational requirements are a minor consideration with modern processing platforms but worth noting. NDVI and SAVI require two spectral bands (red and NIR), while EVI requires three (adding blue). All three can be calculated at Sentinel-2's native 10-meter resolution since Bands 2, 4, and 8 are all available at 10 meters. Processing time differences are negligible for field-level analysis. Correlation with crop yield varies by growth stage: in early season, SAVI typically shows the highest correlation; at peak biomass, EVI outperforms the others; across the full season, all three indices show similar cumulative correlations with final yield. Data availability and historical archives favor NDVI because it can be calculated from any multispectral sensor, including Landsat (available since 1984) and even AVHRR (available since 1981). EVI requires a blue band that some older sensors lack or measure at lower quality. SAVI has the same band requirements as NDVI and therefore the same long historical archive potential. For multi-decadal time series analysis — such as studying long-term yield trends or climate change impacts on agriculture — NDVI remains the only option with consistent data spanning more than 30 years.

For routine, all-purpose crop monitoring throughout the growing season, NDVI remains the best default choice. Its simplicity, universality, and extensive supporting literature make it the most practical index for day-to-day management decisions. Use NDVI as your primary monitoring tool and switch to specialized indices only when NDVI's limitations are relevant to your specific situation. Most agronomic decisions — identifying stress zones, timing fungicide applications, assessing harvest readiness — can be made effectively with NDVI alone. Switch to SAVI (or MSAVI) during early-season monitoring when canopy cover is below 30-40%. This is critical for monitoring crop emergence in spring-planted crops (corn, sugar beet, soybeans, spring cereals) from planting through early vegetative growth (typically April through June in Central Europe). SAVI's soil correction provides more accurate vegetation assessment when a significant proportion of each pixel contains bare soil. On fields with particularly bright or variable soil backgrounds — sandy soils, chalky soils, or fields with strong soil color differences — SAVI may be preferable to NDVI even at moderate canopy cover levels. Use EVI when you need to differentiate crop conditions in dense canopies. This is most relevant for corn from late July through August, sugar beet from July through September, and rapeseed during vegetative growth before flowering. EVI's extended sensitivity at high LAI values makes it the best choice for late-season nitrogen management decisions, yield estimation in high-yield environments, and any situation where NDVI shows uniformly high values (above 0.8) across a field that you know has within-field variability. If your platform supports it, consider using EVI alongside NDVI: display NDVI for general monitoring and switch the map to EVI when a field shows NDVI values approaching saturation.