As extreme weather becomes more frequent and increasingly impacts the price behavior of agricultural commodities, new and more sophisticated techniques of forecasting crop yields are emerging.
One of these is remote sensing, where typically satellites are used to help make early predictions of yield across a range of crops. This high-tech forecasting relies on measuring how vegetation reflects or absorbs electromagnetic radiation, most unseen by the naked eye, to assess likely future yield.
But turning remote-sensing information into useful and reliable forecasts is both an art as well as a science, as it requires highly trained analysts to process and interpret the imagery. Remotely sensed imagery is just one stream of evidence; it also requires weather and climate data as well as fieldwork to be added into the melting pot.
To explain how the esoteric world of remote sensing can be used to give an edge in predicting yield and resulting price of agricultural commodities, we sat down with Dr Daniel Redo, Manager, Agriculture Research with the Lanworth team at Thomson Reuters. He oversees the Lanworth team’s early season and long-range forecasts of corn, soybean, wheat, and rapeseed production in the United States, South America, Europe, Asia, and Australia.
Can you explain to us the basics of remote sensing in layman’s terms?
A good starting point is to consider Google Earth where we can see images that are compatible to the human eye. Water is in various shades of blue, vegetation appears green, and so forth. The visible colors are certainly useful, but there is so much more information that a sensor can capture that we humans can’t see. So the more advanced techniques in remote sensing also rely on analyzing energy that is not visible to the naked eye. Beyond physical light that we can see in the form of colors, there is a range of other wavelengths of energy ranging from electromagnetic, infrared, to even microwaves that are critical for forecasting crop yields and mapping area.
Sensors on satellites, drones or airplanes can detect how this energy is transmitted, absorbed or reflected depending on the shape and texture of the subjects. The result is that we can produce a grid-like picture of individual pixels that can be given digital markers to help identify differences among objects based on their reflectance properties. Detecting these differences can help us for example, distinguish agriculture from other land covers such as forest, water, and urban areas.
Has remote sensing developed to a degree that it is a mainstream part of the overall forecasting process?
Satellite technology is not particularly new and many images have been and still are freely available from the U.S. government for example. What I would say has changed the equation is the recent development of more efficient storage and processing capacity. The takeoff of remote sensing is partly about much greater server capacity to store the imagery and also the ability today to process vast amounts of satellite images.
Natural color (left) and false color (right) composite images taken from the Landsat 8 satellite on July 20, 2014 (Images were downloaded from USGS’ Global Visualization Viewer at http://glovis.usgs.gov/). Dominant features shown are the Missouri River, the city of Mobridge, South Dakota, and a mixture of farmland and other forms of natural vegetation. In the false color image on the right, healthy crops and forested areas are more pronounced due to the deep red color, which is possible by taking into account near-infrared radiation, a form of energy not detectable by the human eye.
Is it available in all geographic markets or dependent on satellite coverage?
We are able to cover nearly all major agricultural regions using NASA’s MODIS satellite. This is our workhorse satellite and is high enough in altitude to revisit the same point on the Earth’s surface everyday, which gives us plenty of opportunity to find cloud-free days. The MODIS sensor onboard the Terra and Aqua satellites orbit higher above the Earth than most satellites and the higher up you get, each pixel becomes coarser – representing a range of 250 meters by 250 meters with MODIS. While this is considered coarse, we are able to see entire regions on a frequent basis.
To see things below that resolution we use sensors onboard lower orbiting satellites such as Landsat, another U.S. satellite, or one could use aircraft and drones to get more details like in precision agriculture. MODIS though is good for global level remote sensing, say for instance, if we wanted to detect widespread drought in Australia.
Can anyone make sense of them or is it more complicated?
While we can all go to Google Earth and see vegetation or water and compare over time to see droughts or floods, you need to transform that into something useful, something quantitative. How much of a production impact will a change in weather have, or to be more precise, what are the quantifiable changes in yield? This is where experience of interpreting data matters. And because no two years are the same, it also helps to have a number of seasons under your belt.
Can you explain how images are interpreted?
While satellite images are by their nature giving us historic information, they can still be used as part of a forward-looking forecast. In addition, they are also an important cross-check. A lot of the groundwork is visiting fields during planting and yield formation but using satellite imagery allows us to directly monitor crop conditions on a continuous basis over areas the size of states and countries, unlike fieldwork or any other tool. Because they directly monitor crops on large scales, satellites are also our best tools for detecting new or novel changes in crops.
For example, over the past few years, satellites have shown us that South American farmers, particularly in Brazil and Paraguay, have been planting crops much earlier than in the past. This has substantially lowered some of the risks associated with drought and disease, as it helps ensure the crops have formed much of their yield prior to the time when droughts or diseases peak.
What types of problems can it detect – just droughts or other problems for example such as diseases or lack of nutrients?
There are many available options but it is predominately droughts (or lack thereof) that we detect. We would use a variety of techniques such as vegetation indices to map the amount of vegetation density present, which can be indirectly related to droughts or diseases through a lack of nutrients.
Who are the biggest users of remote sensing in agriculture?
Most of our clients are hedge fund managers or livestock feeders so they indirectly use remote sensing products. The biggest direct users are likely those in the precision agriculture business operating often at the scale of an individual farm.
Are there particular agricultural commodities that remote-sensing works best for or is most common? For instance in Asia, how important is it for predicting crop yields and the direction of markets such as soya, palm oil, etc?
There are some fundamental differences as no two crops are alike. For the major grains and oilseeds, remote sensing works well because we can easily detect the growth cycle of the plant. After the crop is planted, we can easily see the plant green-up, reach maturity, senesce, and then be harvested. We can then compare this season’s growth cycle at various times to other seasons.
But for other crops such as Indonesian palm oil, sugarcane or coffee, it is more complicated as these crops do not have similar growth cycles. These are tree crops and harvest occurs over several years. In the case of palm oil, for example, the trees remain after the fruit is removed. This presents a big challenge for the remote sensing community since we can’t see changes in vegetation density with the tree remaining after just the fruit has been removed.
How common is it for market participants trading agricultural derivative products to use remote sensing? Can you use this data to anticipate shifts in soft commodity prices?
Remote sensing is increasingly used by market participants seeking to gain an advantage. It is still the only tool available that can view and monitor large areas on a consistent basis without time-consuming and expensive fieldwork.
However, what still prevents ubiquitous use is that it requires trained, skilled interpreters and processing capability to distill an image into something actionable. The benefit of our modeling approaches is that it can generate a specific production number – we look at the area, quantity and yield to give actionable numbers. The other value add is timeliness. We can give our forecasts more frequently as well as ahead of official reports such as that by the U.S. Department of Agriculture or other national agencies.
For example this year in the U.S., soy and corn crops produced a record yield. Remotely sensed imagery, coupled with weather monitoring and fieldwork, helped us gain that advantage. We were seeing much higher yields evidenced by satellite imagery before nearly everyone else and our customers had access to that information before everyone else.