Agriculture Monitoring with SAR at Scale

Over the last 40 years, agriculture monitoring with electro-optical satellite imagery has moved from a research topic to an operational practice. Most commonly, the normalized difference vegetation index (NDVI) is used as a proxy for vegetation health in determining management areas for variable rate application of crop inputs. We are also now starting to see crop yield estimation becoming a common practice. As with all electro-optical applications, agricultural monitoring can be interrupted by cloud. Is this a place where global high resolution SAR can enhance current operations?

Agriculture Monitoring Today

The use of satellite imagery for monitoring agriculture has been around for a long time. It seems like almost every agronomic service provider and large agrochemical company is buying satellite imagery, converting it to a vegetation index of some sort, building statistical or machine learning models relating the vegetation index to vegetation biophysical characteristics, and either using the information for business intelligence or passing it along to agronomists and farmers to help inform their precision agriculture operations. This operational use has expanded in the last 7 years or so with the introduction of higher resolution, high-cadence, global multispectral datasets from RapidEye and Planet.

The most commonly used vegetation index is the good ol’ normalized difference vegetation index or NDVI. The use of NDVI has stood the test of time, having been in use since at least the early 1970’s. This is for good reason since the NDVI reveals some information about photosynthetic activity and cellular structure of vegetation in a single numerical value. In other words, it’s easy to use and understand. Of course there are some limitations in what NDVI can truly reveal about vegetation structure, but it’s a good tool if you only have red and near infrared spectral bands.

NDVI of an agriculture area in Argentina. Greener areas have higher NDVI values (healthier vegetation). Produced from Landsat data using Google Earth Engine.

Like all applications that rely on seeing change over time with optical imagery, agriculture monitoring suffers from limitations due to cloud. This issue is most problematic in areas closer to the equator, though it is also a bit of a problem in northern hemisphere agriculture.

Clouds make it a bit difficult to get good NDVI signal over arable land in Mato Grosso, Brazil. NDVI time series from Sentinel-2 via Sentinel Hub by Synergise.
Sentinel-2 NDVI time series over Minnesota, USA. Still lots of clouds reducing the frequency of observation and measurement. Via Sentinel Hub by Synergise.

SAR for Agriculture Monitoring

So if cloud is an inhibitor to consistent crop monitoring, perhaps it’s possible to get more information about crops by using synthetic aperture radar (SAR). However, I’m not aware of many examples of using SAR in regular large scale commercial operations to monitor crops. Perhaps this is because it is a bit less easy to use and interpret than optical data. It could also be that there aren’t many systems besides Sentinel-1 that have the capacity to frequently observe global agriculture areas. Perhaps it’s also because some practitioners want higher resolution data (< 5m) for their applications.

Sentinel-1 time series over a US agriculture area. Using the orthorectified VH from ascending passes. Via Synergise Sentinel Hub.

I’m certain that with an increase in new datasets from commercial providers like Capella and Iceye, as well as new government sponsored systems like the Radarsat Constellation Mission, we will see more operational use of SAR data in agriculture monitoring. There is a wealth of very exciting research going on to support this future operational use. A quick search on Google Scholar for “Sentinel-1 Agriculture” yields a wide array of agriculture applications that can be addressed with C-band data at 10m resolution. There is an even deeper pool of applications that can be addressed with combinations of X-Band, C-Band, and L-Band data, which reveal different characteristics of crop structure, moisture content, and soil conditions.

Some of the most interesting research on the topic of agriculture monitoring with SAR is happening within the Joint Experiment for Crop Assessment and Monitoring (JECAM). I‘m particularly excited about the potential of monitoring crop biomass with multi-frequency SAR, which is being studied by researchers at Agriculture and Agrifood Canada.

Biomass and Soil Moisture from SAR. From Hosseini M., McNairn H., 2017, Using multi-polarization C- and L-band synthetic aperture radar to estimate wheat fields biomass and soil moisture. International Journal of Earth Observation and Geoinformation, 58, 50–64.

Simple Crop Monitoring Experiment with SAR

While the research into biophysical parameter estimation is fascinating and critical to improving future global agriculture insights, I was curious whether there was some simple vegetation health metric akin to NDVI that could be derived from SAR. It turns out that there have been a number of studies looking into whether SAR backscatter correlates with NDVI and if a “Radar Vegetation Index (RVI)” could be produced from polarimetric SAR data. I also came across some investigations into the “Cross Ratio” and a normalized difference ratio of backscatter from VV and VH polarizations, and thought it might be fun to try this out in Google Earth Engine, then compare the results to an NDVI from Sentinel-2.

Normalized difference backscatter ratio from Sentinel-1. Values averaged over a month in the early season. Produced using Google Earth Engine.
NDVI from Sentinel-2. Values averaged over a month in the early season. Produced using Google Earth Engine.

Well, this is interesting. The normalized difference ratio from the SAR data certainly looks like an NDVI image and has some correlation in the data, but it’s not providing the same information as an NDVI. That should be no surprise, as the physical principles of SAR and optical are quite different. While NDVI is most influenced by leaf internal structure and biochemical constituents (assuming limited shadow and soil influence), SAR backscatter is affected most by canopy structure and by other things like vegetation moisture content, wind altering the canopy structure, as well as the properties of the underlying soil (depending on the wavelength). A fascinating review of these dynamics can be found in “Characterization of Crop Canopies and Water Stress Related Phenomena using Microwave Remote Sensing Methods: A Review”.

So, the SAR data is providing some information on the structure of the vegetation that is correlated with what we can derive from optical data, and it’s certainly better to have this information from SAR than no information at all when there is cloud obscuring optical observations. This is especially true in agriculture areas with frequent cloud cover.

Wrapping Up

There is clearly a lot of potential value in the use of SAR data for agriculture monitoring. SAR data provides information that supplements optical monitoring and in some cases can be a replacement. Understanding what we are seeing in SAR and derivative products requires an understanding of the nature of how the microwaves interact with the canopy structure of plants, plant water content, and the soil. This can be less intuitive than it is with reflected light in optical applications. However, with additional research and some creativity in displaying derived information to users, I’m sure we will see SAR products in more commercial agriculture contexts over the next few years. We can expect this use to be multiplied when higher resolution, higher cadence commercial SAR systems enter into operations.