Accurate monitoring of agriculture is vital in discovering negative changes in crop condition that can severely impact local populations’ food security as well as global market stability. One international group of researchers came together in 2017 to develop new crop monitoring methods that could be utilized across diverse agricultural systems.
The team tested models based on the leaf area index (LAI), a measurement of the crop canopy, and a variable that allows for improved crop condition monitoring and yield prediction. Running and comparing the models for 4 different crops and across 7 countries, the team found that their simplified, single-predictor model performed favorably well against more data heavy models. These results provide a pathway for future researchers, policymakers, and stakeholders to more easily ascertain crop productivity and provide humanitarian or market stabilizing efforts as needed.
The ability to monitor crop conditions in a local area as well as aggregate conditions across the globe is important for both individual farmer security and agricultural commodity markets.
In East Africa, swarms of locusts devastated crops in 2020. With a swarm of only around one square kilometer in size (and many swarms are far larger) able to consume an equivalent amount of food as 35,000 people in a single day, the locust crisis ravaged agriculture in the region. With agriculture making up around 30% of regional GDP and employing 65% of the working population, these losses threatened severe consequences not just for regional food security but for individual economic stability.
In another story of crop production instability, 2011 saw intense spikes in food commodity prices. An analysis of food prices in late 2010 by JP Morgan Chase found that over the previous year corn, wheat, and soybeans had increased in price by 63%, 84%, and 24% respectively. These price hikes (caused by failed crop harvests across multiple countries) continued through 2011 and led to countries instituting export restrictions and softening import controls in an attempt to prevent high prices for their populations.
Both of these situations stress the importance of effective crop monitoring so that responsible parties can proactively implement policies to limit the damage instead of reactively responding. One project, the Joint Experiment for Crop Assessment and Monitoring (JECAM), was started in 2017 with the goal of comparing crop monitoring methods across diverse agricultural settings to evaluate the robustness of each method.
Data used for this study came from 7 JECAM countries including Argentina, Canada, Germany, India, Poland, Ukraine, and the United States. The team conducting the study consisted of 22 researchers from institutions spread across the world, including NASA Harvest’s Dr. Mehdi Housseini and Program Director Dr. Inbal Becker-Reshef.
Figure 1. Locations of the Joint Experiment for Crop Assessment and Monitoring (JECAM) sites contributing to this research in (a) a world map, and land cover of each site including (b) Argentina, (c) Canada, (d) Germany, (e) India, (f) Poland, (g) Ukraine and (h) USA-North Dakota (ND). The white polygons are the boundaries of the test sites
One effective variable for monitoring crop condition and productivity is leaf area index (LAI), which is a measure of the crop canopy development. Observing LAI in the early stages of crop growth is a good indicator of the health and eventual yield of a given field. Given its ability to penetrate clouds and provide unimpeded views of the ground, synthetic-aperture radar (SAR) satellite data is exceptionally useful for LAI measurement.
One common monitoring technique utilizing SAR data is the Water Cloud Model (WCM), which has produced highly accurate estimates in recent studies. As effective as the WCM is however, successfully implementing it is difficult as it requires temporally and spatially coincident LAI and soil moisture training data collections. The difficulty arises from the highly variable nature of the latter variable, soil moisture, which can change significantly from day to day, unlike LAI which changes at a slower pace. This means that soil moisture collections used for model training data must be collected very near the time a SAR satellite is passing overhead. This creates complicated labor and resource issues.
The JECAM research team wanted to develop a method that could rely solely on LAI training data and be trained without using soil moisture data. To do this, the team created a Support Vector Machine (SVM) model that only required LAI as input. The team compared the predictive power of the known-to-be-effective WCM to the single variable SVM with four different crops (corn, soy, wheat, and rice) across 7 countries on 4 continents to ensure that each method was evaluated in a number of agricultural systems.
They found that estimations from SVM were on par with estimations derived from the WCM, indicating the former’s utility in monitoring LAI. Corn and wheat LAI estimates from the SVM had greater accuracy than the WCM for those crops. WCM performed slightly better than SVM for soybeans, however the difference was small and SVM was still quite successful. Rice was not tested with both models, as its method of planting by flood renders soil moisture measurements useless and therefore the WCM was not run. While unable to compare between models for rice, the SVM performed very well and LAI estimates had high accuracy.
One particularly promising result from the study was the performance of the SVM in estimating soy LAI. For soybeans, soil moisture data was unable to be collected from the Argentinian and United States’ JECAM sites, so those sites were excluded from the calibration of the WCM and SVM models. Despite this, the SVM model (using only LAI training data) was able to estimate with high accuracy LAI in the two excluded countries despite having no training data from either one. This suggests great robustness of the SVM to estimate LAI in areas where researchers are unable to collect data.
Despite the promising results, the JECAM research team do note that their method comes with some limitations. As the SVM models are crop specific, crop type maps for the study area will be necessary to ensure that model users are using the correct model for a given field. Since the crop grown in a field can change from year to year, ideally one would use in-season crop type maps to make this determination. However, in-season crop type maps are not as accurate as end of season crop type maps. Further work in developing more accurate in-season crop type maps would aid in the deployment of the JECAM team’s SVM models.
Overall, the JECAM team was able to develop a simplified, single-variable model capable of estimating crop LAI with comparable accuracy to previously proven, but more resource intensive methods. These developed models will aid policymakers and stakeholders in making more informed decisions going forward and definitely assist in future humanitarian and market stabilization efforts. Further the successful work of the JECAM team speaks to the benefits of international cooperation, cross-border sharing of data, and improved access to countries for research.
More information on JECAM and their mission can be found here.
The complete study can be found here.