Satellite data has become an integral part of modern agricultural management to keep track of crop progress. Some satellite data, however, also can foretell how a crop will perform in the future, offering valuable insights to a wide range of industry players.
NDVI is probably the most important of the satellite data. By reading infrared light waves reflected from plants, NDVI can signal stresses to plant health, such as oncoming drought, as much as two weeks before problems are visible to the naked eye. Many agricultural industry participants can benefit from such advance alerts. Farmers can increase irrigation or add crop protection to forestall a pest infestation. And physical traders, processors, and food and beverage companies could seek out alternative supplies, or hedge their positions.
NDVI also can reveal positive indications about a crop, providing a heads-up to market participants, logistics companies, and others to prepare for a big harvest. Another type of satellite data, called evapotranspiration (ET), also sends early-warning signals about plants, based on measures of moisture evaporation and transpiration. But ET is available only on a monthly basis, much less frequent than the eight-day reports on NDVI.
NDVI is the key input to Gro’s machine-learning-based yield models. It plays a decisive role in Gro’s yield forecasts for corn in the US and Argentina, soybeans in the US, and wheat in countries including Ukraine, Russia, India, and the US.
In this Weekly Insight, we explain the science behind NDVI and the impact it has on Gro’s robust yield models. We also examine cases when awareness of NDVI signals could be put to valuable use by industry and government officials.
Early space exploration quickly led to atmospheric and meteorological studies. NASA in 1972 launched the Earth Resources Technology Satellite, the forerunner to Landsat, the world’s longest-running satellite imagery program. This first satellite was able to distinguish between visible red and near-infrared reflectance bands, which allowed it to identify vegetation, soil, water, and other features.
Light from the sun is present as visible light (reds, greens, and blues) and light not visible to our eyes (infrared and ultraviolet). These can be absorbed, transmitted, or reflected by an object. In healthy plants, most of the visible light is absorbed for use in photosynthesis, and much of the near-infrared radiation (NIR) is reflected. However, if the plant is stressed, because of dehydration, for instance, it reflects less NIR and absorbs less light in the visible spectrum, specifically in the red portion, since the plant is not using photosynthesis as efficiently.
Using these assumptions, researchers formulated the so-called normalized difference vegetation index, or NDVI. The equation looks like this:
NDVI = (NIR - Red)/(NIR + Red)
Simply put, NDVI measures plant greenness, a direct measurement of chlorophyll content and photosynthetic activity, which is a basic but reliable way to gauge plant health and biomass, or yield. Using the simple equation above, which returns values between 0 and 1, a higher value represents a healthier plant.
Numerous satellites capture light at these wavelengths. They differ by factors such as pixel size, or how much area is covered in a single image, and by how often the satellite returns to the same region, known as revisit time. Gro chose to use the MODIS sensor aboard the TERRA and AQUA satellite, which has a nominal spatial resolution of 250 meters by 250 meters. It also has a daily revisit time, which is aggregated to eight- and 16-day products in order to allow for cloud cover. The MODIS sensor also has an archive going back to 2000, which gives users the ability to model long-term trends.
Traditional survey-based yield modeling is slowly being supplanted by remote-sensing techniques as satellite technology continues to improve. Changes in crop condition can be detected more quickly using NDVI, which reads near-infrared light waves. Crop stress from pests, diseases, drought, flooding, and other factors won’t immediately be picked up in the visible-light spectrum, but will be reflected in the NIR spectrum.
Simply looking at a map of NDVI readings, however, can often be misleading, partly because the data doesn’t distinguish between planted crops and non-target vegetation, such as surrounding trees. Identifying the exact areas where crops are growing, called crop masking, is a critical first step, and NDVI plays an important role in this process, as well.
Yield models contain a variety of climate inputs such as temperature and precipitation that provide valuable insights about growing conditions. Other variables, like evapotranspiration, reveal information about the condition of the crop and soil moisture. But NDVI is the only variable, in addition to latitude and longitude, that is included in every one of Gro’s yield models. It is also the only yield model variable that directly measures plant activity, in this case photosynthesis, which ultimately determines yield. Other variables in our yield models act as proxies for plant health, but don’t directly measure production of biomass.
NDVI is most effective beginning in the pre-peak season of crop growth and continuing through harvest. Earlier in the season, when most crops are just starting to show shoots, NDVI doesn’t provide much useful information because it isn’t reading as much surface area.
To be sure, NDVI also is less effective with certain crops. In production of ratoon crops like sugarcane, for example, large fields of shoots are cut and harvested over four- to 12-month periods while the roots are left to regenerate. This practice results in significant spatial and temporal variation in growth over sugarcane acreage, which presents a challenge to monitoring yield using NDVI.
A good example of the usefulness of satellite imagery to identify crop risks was during the devastating Mexican drought of 2011-12. Evapotranspiration readings and other indices were signaling severely dry conditions in the spring of 2011. Although many farmers had already committed to ambitious plans for the year, some still had time to alter decisions. Meanwhile, the NDVI anomaly index—which runs from -1 to +1 and compares historical, or normal, conditions with the current condition—was below a neutral reading of 0, forecasting weak yields. Processors and consumers could have hedged against higher prices but, unaware of the alarming satellite signals, generally did not.
The fallout for Mexico’s agricultural sector was disastrous. Corn yields in Mexico dropped 11% between 2010 and 2011 and producer prices jumped 45%. Imports rose to meet demand for the country’s most consumed grain.
NDVI also can send positive signals. After a dismal 2012 harvest, producers in the US Corn Belt were hoping for an upbeat 2013. While initial forecasts looked promising, the USDA, in its monthly WASDE report, gradually reduced yield estimates each month from May through August. Finally, in November, the USDA sharply raised its yield forecast, more than retracing its reduced estimates since the previous spring. The USDA explained that “cooler-than-normal summer temperatures” had offset dryness across the Corn Belt.
NDVI readings, however, were signaling throughout the 2013 growing season that the crop was in better shape than the USDA estimated. NDVI anomaly measurements for the Corn Belt increased weekly from -0.38 in mid-June to 0.26 in the last week of August. The upshot: Final 2013 yield was 158.1 bushels per acre, a 28% jump from the previous year.
Similarly, Brazil’s record soybean crop in 2018 was anticipated by satellite data. For 17 straight weeks, between the end of September 2018 and the end of January 2019, NDVI was the highest it had been for the previous 10 years in Brazil’s major soybean-producing regions. In addition to acreage expansion, yield of 51.6 bushels per acre were the highest ever for the country. Brazil’s bumper soybean crop last year had a huge impact on global markets, as China sought alternative supplies amid a trade war with the US.
NDVI satellite data, a core part of Gro’s data platform, is a powerful indicator of plant health, detecting positive signals and plant stressors as much as two weeks before such issues are apparent to the naked eye. Simple maps of NDVI readings aren’t very helpful, partly because they fail to distinguish between crops and extraneous vegetation. With proper masking, however, NDVI becomes a vital component of reliable crop yield models, along with other variables such as soil moisture, temperature, and precipitation. Industry players across the agricultural sector, including traders and food, beverage, and chemical companies, can benefit from building predictive models which incorporate NDVI to better inform their decision-making.