4.0 - Crop Sensing

CHECKLIST FOR COLLECTING REMOTE SENSING IMAGERY 

• Satellite imagery is collected during set orbit times irrespective of cloud cover. 

• The pricing of satellite imagery is set using specific scene sizes, which are often significantly larger than the area of interest. 

• Aerial imagery offers more flexible acquisition of data and is often more cost-effective. When selecting an image source consider the spatial and spectral resolution needed for the application. 

• For small areas, biomass data can be acquired using active optical sensors

In agriculture, optical sensing is commonly used to measure variability in soil and vegetation. Optical imaging utilises the visible, near-infrared (NIR) and thermal portions of the electromagnetic spectrum. Variations in the surface of the earth cause sunlight to be reflected absorbed or transmitted.

Vegetation and soil exhibit all 3 energy exchanges and these interactions vary across the EM spectrum. Knowledge of which wavelengths are absorbed by different land features and the intensity of the reflectance can help one to understand the state of an object. Optical sensors will generally use at least two different bands of light, most commonly the red and NIR.

Using the distinct spectral properties of plants with low reflectance in the visible and very high reflectance in the NIR region of the solar spectrum, the spectral contrast can be used for identifying the presence of green vegetation and evaluating some characteristics (e.g. cover and biomass) through various vegetation indices.

One example is plant cell density (PCD), which is the ratio of infrared to red reflectance. This provides a measure of crop vigour. The PCD values cannot be used to indicate specific levels of biomass. This indicates a level of biomass variability within the field.

When a high level of variability is indicated by the PCD image this can provide a basis for differential management of inputs such as fertiliser, water and growth regulators.

Ratios between other spectral bands provide information about other physical information such as plant water content and chlorophyll concentration or absorption. Ratios of narrow spectral bands will generally provide more specific information than those created from sensors with very broad spectral bands.

The spatial resolution determines the level of detail which can be distinguished in an image. It is determined by the size of the pixels within an image. As spatial resolution increases, each pixel represents a smaller area on the ground which increases the level of detail contained within an image.

The example below shows the effect of increasing spatial resolution of a PCD image:

(a) 2 m resolution; (b) 5 m resolution; (c) 10 m resolution; and (d) 25 m resolution.

UTILISING REMOTE-SENSED IMAGERY

• Enhancement tools help make the imagery more interpretable for specific applications. Enhancement and classification tools are often used to highlight features.

• Classified images and vegetation indices (e.g. NDVI & PCD) are frequently used as a substitute for biomass. The analysis of imagery in conjunction with other ancillary data helps enhance one's understanding of within-field variability and its causes.

• Imagery is only a surrogate for physical plant characteristics at a specific time. Field validation is essential to measure the attribute of interest.

More reading

Vegetation Indices:

This is a quick summary of the vegetation indices available in PCT Agcloud.  

NDVI, PCD & SVI 

These indices produce similar results with a few caveats and have been used for such purposes as: 

• Photosynthetic capacity of plant canopies 
• General ‘condition’ or ‘health’ of vegetation 
• Other correlations/estimations made from NDVI include: 
• Biomass 
• Leaf area index 
• Chlorophyll concentration of leaves 
• Others 

Note that all these correlations cannot hold at once as they are not all completely related to each other. 

Normalized Difference Vegetation Index (NDVI) 

The granddaddy of vegetation Indices. NDVI, conceived around 1970, normalizes the difference between red and near-infrared reflectance.  

• Operates in a functional range of 0 to 1 
• Tends to saturate easily on higher biomass crops 
• A good starting point 

The NDVI Wikipedia page is very good at detailing performance and limitations: https://en.wikipedia.org/wiki/Normalized_difference_vegetation_index 

Plant Cell Density (PCD) 

A non-normalized vegetation index which is simply NIR / Red.  

• Operations in a functional range of 0 to infinity. Often values 0 to 20. 
• Can be useful to use as extremes are not normalized out as they are for NDVI so it can appear ‘less noisy’ than that SVI and ‘less saturated’ than NDVI  
• Can be hard to work with as the range is never really known without examining the data whereas normalized indices will have a functional range of 0 to 1.  

Satamap Vegetation Index (SVI)/Modified Triangular Vegetation Index 2 (MTVI2)

Historically Satamap took on MTVI2 and applied a fixed, unique colour scale and bundled up named it Satamap Vegetation Index (SVI). MTVI2 can still be used as a vegetation index with the rainbow colour scale to compare alongside the other indices. 

• MTVI2 / SVI is a solid choice in high biomass situations (resistant to saturation https://www.tandfonline.com/doi/abs/10.5589/m08-071) where you want the range to remain between 0 and 1 
• Alongside red and NIR, it exploits the green band. This attempts to account for soil colour. 
• MTVI2 is commonly used in narrowband and hyperspectral applications but we find it performs excellently with Sentinel 2 imagery, especially in high biomass situations. 
• Identical results to MCARI2 

Figure 1: Common vegetation indices compared for drought-stressed (low biomass) barley. 

Figure 2: Common vegetation indices compared for high biomass crop. Chemical damage (bottom left) and fertiliser trial (top right).

Leaf Area Index (LAI) 

“LAI is defined as half the developed area of photosynthetically active elements of the vegetation per unit horizontal ground area. It determines the size of the interface for the exchange of energy (including radiation) and mass between the canopy and the atmosphere.” – Sentinel2 ToolBox Level2 Products, 2016

Canopy Chlorophyll Content (CCC) 

The chlorophyll content is a very good indicator of stresses including nitrogen deficiencies. It is strongly related to leaf nitrogen content (Houles et al. 2001).  - Sentinel2 ToolBox Level2 Products, 2016

LAI & CCC estimations using neural networks 

For LAI & CCC we use an on-the-fly implementation of the biophysical variables in the Scientific Toolbox Exploitation Platform (STEP) developed by the European Space Agency (ESA). STEP takes a neural network approach to calculating these parameters. To train this model, the process uses simulated data to form a ‘generic’ algorithm. What this means is it should have reasonable performance in most geographic locations over several vegetation types but to use with caution. ESA have used this method extensively and successfully in the past for many other satellite missions.  

Other indices:

Normalized Difference Red Edge (NDRE) 

NDRE follows the same concept as NDVI, except in place of the near-infrared band. It uses what is called ‘Red Edge’.  

A healthy leaf will generally absorb red light and be highly reflective of NIR. By moving to a point somewhere between red and near-infrared in place of infrared, there is a lower chance of saturation in high biomass crops as top-of-canopy reflectance will be less intense. In addition, there is some evidence to suggest NDRE can correlate with leaf nitrogen content (reference).

NDRE can mitigate saturation with PCD and SVI and experiment with leaf nitrogen content with CCC. Also, the red edge band is lower resolution (20m as opposed to 10m). 

Moisture Stress Index (MSI) 

MSI is an estimation of leaf water content. NIR is dived by shortwave infrared (SWIR).  SWIR will reflect more as leaf water content decreases. NIR reflectance is not directly impacted by water content and is therefore used as a reference.  

Similar to PCD, MSI is not normalised. So, the exact range will not be known. Generally, values from 0.4 to 2 are seen.  

It is important to note that the values are inverted to a normal vegetation index. A high value indicates low water content/high plant stress.  

Comparing Indices: 

Below is a chart that compares the values of MSI (blue), SVI (red), NDVI (green), and NDRE (yellow) throughout the growing season. The inverse nature of MSI and the non-saturating characteristic of SVI is obvious.  

Conclusion: 

There are several different vegetation indices available and this document just scratches the surface. Each one has its own specific strengths and weaknesses. Having a brief practical understanding of how these can be applied will help in both interpreting the data and figuring out what to use in specific situations. 

References:

Sentinel2 ToolBox Level2 Products, 2016, https://step.esa.int/docs/extra/ATBD_S2ToolBox_L2B_V1.1.pdf

Houlès, V., Mary, B., Machet, J.M., Guérif, M., & Moulin, S. (2001). Do crop characteristics available from remote sensing allow to determine crop nitrogen status?

G. Grenier, & S. Blackmore (Eds.), 3rd European Conference on Precision Agriculture (pp. 917-922). Montpellier: Agro Montpellier