“Feet on the ground, eyes in the sky”: measuring yields by satellite

Whether forecasting harvest quantities or assessing farm income, accurately measuring crop yields is crucial. Yet, many developing countries struggle to do so, as shown in a paper to be published soon in the American Journal of Agricultural Economics.[1]The authors of the article highlight both the weakness of the most common measurement methods and the quality of the predictions provided by satellites.

They draw on the findings of a 2016 study in eastern Uganda, in which smallholder maize yields were estimated in four distinct ways. Surveyors asked approximately 500 farmers to report their perceived productivity of the crop on each of the plots they cultivated. In addition, technicians randomly distributed sample sections within each of the 460 selected plots. They also exhaustively determined the yields of more than 200 entire plots. Finally, maize productivity was calculated using vegetation indices derived from images taken by a Sentinel-2A satellite from the European Copernicus program. The resolution of this type of satellite is 10 meters in the observable light spectrum and near infrared.

The study first examines the results of field measurements. Average maize yields are not significantly different depending on whether they are assessed on the basis of sample cuts or entire plots. On the other hand, farmers tend to significantly overestimate the productivity of their crop, as the average yield they report (1.83 t/ha) is 2.7 times higher than the objectively determined yield of entire plots (0.68 t/ha). This is particularly worrying given that the self-reported method is widely used in many low-income countries.

The correlation between the yields of the sample sections and those of the entire plots is 50 %. It is limited by the great heterogeneity of the intra-plot yields, because the size of the "sub-plots" in which the sample sections are carried out (a square of 8 m on each side) represents only 4 % of the average surface area of the plots. But above all, we note that there is almost no correlation between the yields estimated by the farmers and those of the sub-plots, which confirms the shortcomings of the declarative approach.

The researchers then compared the ground yield measurements with yields predicted from satellite images. These are calculated using a linear equation linking, for each plot, the vegetation indices provided by Sentinel-2A on May 30 and June 19, 2016, and the yields of entire plots or sample sections. The findings are encouraging. On the one hand, the yields assessed by satellite explain the variability of entire plot yields as well, if not better. On the other hand, the satellite-based predictive model calibrated on sample section yields can estimate entire plot yields almost as well as the satellite model calibrated on entire plot yields. From a practical point of view, this conclusion is very important because sample sections are much less expensive to carry out than determining entire plot yields.

The study thus clearly demonstrates the usefulness of satellites for measuring yields, even when crops are grown on very small plots, as is the case in eastern Uganda. There are, however, obstacles. Below 0.10 hectares, the median size of maize plots in this region, the quality of satellite observations decreases significantly. It is also less good when farmers produce several crops simultaneously in the same field, which complicates image interpretation. In the example considered, three-quarters of the maize plots are grown in association with cassava, beans, or peanuts. Each of these plants, depending on the extent of its foliage and its height, "blurs" the maize vegetation index to a greater or lesser extent. Advances in remote sensing should help remove these obstacles.

These results open up new perspectives for stakeholders in the agri-food sector, as well as for governments. According to Airbus' contribution to the white paper on agricultural insurance recently published by the CICA (International Confederation of Agricultural Credit) in collaboration with FARM,[2], geospatial information, combined with digital technologies and big data, is poised to revolutionize the entire agricultural insurance value chain. The challenge is to make it affordable for small farmers in developing countries.

[1] David B. Lobell, George Azzari, Marshall Burke, Sydney Gourlay, Zhenong Jin, Talip Kilic, and Siobhan Murray, “Eyes in the Sky, Boots on the Ground: Assessing Satellite- and Ground-Based Approaches to Crop Yield Measurement and Analysis,” American Journal of Agricultural Economics, aaz051, article posted online October 26, 2019.

[2] “A contribution to the Sustainable Development Goals. For a more resilient and better protected agriculture against climate hazards. White paper on agricultural insurance,” downloadable document on www.cica.ws/conferences