Accurate automated segmentation of remote sensing data could benefit applications from land cover mapping and agricultural monitoring to urban development surveyal and disaster damage assessment. While convolutional neural networks (CNNs) achieve state-of-the-art accuracy when segmenting natural images with huge labeled datasets, their successful translation to remote sensing tasks has been limited by low quantities of ground truth labels, especially fully segmented ones, in the remote sensing domain. In this work, we perform cropland segmentation using two types of labels commonly found in remote sensing datasets that can be considered sources of “weak supervision”: (1) labels comprised of single geotagged points and (2) image-level labels. We demonstrate that (1) a U-Net trained on a single labeled pixel per image and (2) a U-Net image classifier transferred to segmentation can outperform pixel-level algorithms such as logistic regression, support vector machine, and random forest. While the high performance of neural networks is well-established for large datasets, our experiments indicate that U-Nets trained on weak labels outperform baseline methods with as few as 100 labels. Neural networks, therefore, can combine superior classification performance with efficient label usage, and allow pixel-level labels to be obtained from image labels.
The advent of multiple satellite systems capable of resolving smallholder agricultural plots raises possibilities for significant advances in measuring and understanding agricultural productivity in smallholder systems. However, since only imperfect yield data are typically available for model training and validation, assessing the accuracy of satellite-based estimates remains a central challenge. Leveraging a survey experiment in Mali, this study uses plot-level sorghum yield estimates, based on farmer reporting and crop cutting, to construct and evaluate estimates from three satellite-based sensors. Consistent with prior work, the analysis indicates low correlation between the ground-based yield measures (r = 0.33). Satellite greenness, as measured by the growing season peak value of the green chlorophyll vegetation index from Sentinel-2, correlates much more strongly with crop cut (r = 0.48) than with self-reported (r = 0.22) yields. Given the inevitable limitations of ground-based measures, the paper reports the results from the regressions of self-reported, crop cut, and (crop cut-calibrated) satellite sorghum yields. The regression covariates explain more than twice as much variation in calibrated satellite yields (R2 = 0.25) compared to self-reported or crop cut yields, suggesting that a satellite-based approach anchored in crop cuts can be used to track sorghum yields as well or perhaps better than traditional measures. Finally, the paper gauges the sensitivity of yield predictions to the use of Sentinel-2 versus higher-resolution imagery from Planetscope and DigitalGlobe. All three sensors exhibit similar performance, suggesting little gains from finer resolutions in this system.
Machine learning and satellite data of crops shows that farms that till the soil less can increase yields of corn and soybeans and improve the health of the soil. Farmers have resisted a switch to reduced tilling because it was believed to reduce yields. Instead, it may increase yields while lowering production costs and reducing soil erosion.
The economic costs of Indonesia’s 2015 forest fires are estimated to exceed US $16 billion, with more than 100,000 premature deaths. On several days the fires emitted more carbon dioxide than the entire United States economy. Here, we combine detailed geospatial data on fire and local climatic conditions with rich administrative data to assess the underlying causes of Indonesia’s forest fires at district and village scales. We find that El Niño events explain most of the year-on-year variation in fire. The creation of new districts increases fire and exacerbates the El Niño impacts on fire. We also find that regional economic growth has gone hand-in-hand with the use of fire in rural districts. We proceed with a 30,000-village case study of the 2015 fire season on Sumatra and Kalimantan and ask which villages, for a given level of spatial fire risk, are more likely to have fire. Villages more likely to burn tend to be more remote, to be considerably less developed, and to have a history of using fire for agriculture. Although central and district level policies and regional economic development have generally contributed to voracious environmental degradation, the close link between poverty and fire at the village level suggests that the current policy push for village development might offer opportunities to reverse this trend.
By monitoring crops through machine learning and satellite data, Stanford scientists have found farms that till the soil less can increase yields of corn and soybeans and improve the health of the soil – a win-win for meeting growing food needs worldwide.
Agriculture degrades over 24 million acres of fertile soil every year, raising concerns about meeting the rising global demand for food. But a simple farming practice born from the 1930’s Dust Bowl could provide a solution, according to new Stanford research. The study, published Dec. 6 in Environmental Research Letters, shows that Midwest farmers who reduced how much they overturned the soil – known as tilling – increased corn and soybean yields while also nurturing healthier soils and lowering production costs.
“Reduced tillage is a win-win for agriculture across the Corn Belt,” said study lead author Jillian Deines, a postdoctoral scholar at Stanford’s Center on Food Security and the Environment. “Worries that it can hurt crop yields have prevented some farmers from switching practices, but we found it typically leads to increased yields.”
The U.S. – the largest producer of corn and soybeans worldwide – grows a majority of these two crops in the Midwest. Farmers plucked about 367 million metric tons of corn and 108 million metric tons of soybeans from American soil this past growing season, providing key food, oil, feedstock, ethanol and export value.
Farmers generally till the soil prior to planting corn or soybeans – a practice known to control weeds, mix nutrients, break up compacted dirt and ultimately increase food production over the short term. However, over time this method degrades soil. A 2015 report from the Food and Agriculture Organization of the United Nations found that in the past 40 years the world has lost a third of food-producing land to diminished soil. The demise of once fertile land poses a serious challenge for food production, especially with mounting pressures on agriculture to feed a growing global population.
In contrast, reduced tillage – also known as conservation tillage – promotes healthier soil management, reduces erosion and runoff and improves water retention and drainage. It involves leaving the previous year’s crop residue (such as corn stalks) on the ground when planting the next crop, with little or no mechanical tillage. The practice is used globally on over 370 million acres, mostly in South America, Oceania and North America. However, many farmers fear the method could reduce yields and profits. Past studies of yield effects have been limited to local experiments, often at research stations, that don’t fully reflect production-scale practices.
The Stanford team turned to machine learning and satellite datasets to address this knowledge gap. First, they identified areas of reduced and conventional tilling from previously published data outlining annual U.S. practices for 2005 to 2016. Using satellite-based crop yield models – which take into account variables such as climate and crop life-cycles – they also reviewed corn and soybean yields during this time. To quantify the impact of reduced tillage on crop yields, the researchers trained a comput
Average impacts on corn yields from conservation tillage across the U.S. Corn Belt from 2008 to 2017. Red colors indicate increased yields under conservation tillage, blue colors indicate yield declines.
The researchers calculated corn yields improved an average of 3.3 percent and soybeans by 0.74 percent across fields managed with long-term conservation tillage practices in the nine states sampled. Yields from the additional tonnage rank in the top 15 worldwide for both crops. For corn, this totals approximately 11 million additional metric tons matching the 2018 country output of South Africa, Indonesia, Russia or Nigeria. For soybeans, the added 800,000 metric tons ranks in between Indonesia and South Africa’s country totals.
Some areas experienced up to an 8.1 percent increase for corn and 5.8 percent for soybeans. In other fields, negative yields of 1.3 percent for corn and 4.7 for soybeans occurred. Water within the soil and seasonal temperatures were the most influential factors in yield differences, especially in drier, warmer regions. Wet conditions were also found favorable to crops except during the early season where water-logged soils benefit from conventional tillage that in turn dries and aerates.
“Figuring out when and where reduced tillage works best could help maximize the benefits of the technology and guide farmers into the future,” said study senior author David Lobell, a professor of Earth system science in the School of Earth, Energy & Environmental Sciences and the Gloria and Richard Kushel Director of the Center on Food Security and the Environment.
It takes time to see the benefits from reduced tillage, as it works best under continuous implementation. According to the researchers’ calculations, corn farmers won’t see the full benefits for the first 11 years, and soybeans take twice as long for full yields to materialize. However, the approach also results in lower costs due to reduced need for labor, fuel and farming equipment while also sustaining fertile lands for continuous food production. The study does show a small positive gain even during the first year of implementation, with higher gains accruing over time as soil health improves. According to a 2017 Agricultural Censuses report, farmers appear to be getting on board with the long-term investment and close to 35 percent of cropland in the U.S. is now managed with reduced tillage.
“One of the big challenges in agriculture is achieving the best crop yields today without comprising future production. This research demonstrates that reduced tillage can be a solution for long-term crop productivity,” Deines said.
To read all stories about Stanford science, subscribe to the biweekly Stanford Science Digest.
David Lobell is also the William Wrigley Senior Fellow at the Stanford Woods Institute for the Environment, a senior fellow at the Freeman Spogli Institute for International Studies and the Stanford Institute for Economic Policy and Research. Graduate student Sherrie Wang is also a co-author. Research was funded by NASA Harvest.
Understanding the determinants of agricultural productivity requires accurate measurement of crop output and yield. In smallholder production systems across low- and middle-income countries, crop yields have traditionally been assessed based on farmer-reported production and land areas in household/farm surveys, occasionally by objective crop cuts for a sub-section of a farmer’s plot, and rarely using full-plot harvests. In parallel, satellite data continue to improve in terms of spatial, temporal, and spectral resolution needed to discern performance on smallholder plots. This study evaluates ground- and satellite-based approaches to estimating crop yields and yield responsiveness to inputs, using data on maize from Eastern Uganda. Using unique, simultaneous ground data on yields based on farmer reporting, sub-plot crop cutting, and full-plot harvests across hundreds of smallholder plots, we document large discrepancies among the ground-based measures, particularly among yields based on farmer-reporting versus sub-plot or full-plot crop cutting. Compared to yield measures based on either farmer-reporting or sub-plot crop cutting, satellite-based yield measures explain as much or more variation in yields based on (gold-standard) full-plot crop cuts. Further, estimates of the association between maize yield and various production factors (e.g., fertilizer, soil quality) are similar across crop cut- and satellite-based yield measures, with the use of the latter at times leading to more significant results due to larger sample sizes. Overall, the results suggest a substantial role for satellite-based yield estimation in measuring and understanding agricultural productivity in the developing world.
Irrigation has been pivotal in wheat’s rise as a major crop in India and is likely to be increasingly important as an adaptation response to climate change. Here we use historical data across 40 years to quantify the contribution of irrigation to wheat yield increases and the extent to which irrigation reduces sensitivity to heat. We estimate that national yields in the 2000s are 13% higher than they would have been without irrigation trends since 1970. Moreover, irrigated wheat exhibits roughly one-quarter of the heat sensitivity estimated for fully rainfed conditions. However, yield gains from irrigation expansion have slowed in recent years and negative impacts of warming have continued to accrue despite lower heat sensitivity from the widespread expansion of irrigation. We conclude that as constraints on expanding irrigation become more binding, furthering yield gains in the face of additional warming is likely to present an increasingly difficult challenge.
Feeding a growing population while reducing negative environmental impacts is one of the greatest challenges of the coming decades. We show that microsatellite data can be used to detect the impact of sustainable intensification interventions at large scales and to target the fields that would benefit the most, thereby doubling yield gains. Our work reveals that satellite data provide a scalable approach to sustainably increase food production.
Not many people go into farming to get rich. Low commodity prices, high operational costs and limited profit opportunities cloud the outlook. William Wrigley Professor and FSE Founding Director ROSAMOND NAYLOR gave a keynote presentation on the path toward a more profitable future at an agricultural symposium hosted by the Federal Reserve Bank of Kansas City. See slides from Naylor’s presentation here.
View video of the presentation
Download PDF of the article