Showing papers by "James S. Gerber published in 2016"

Random Forests for Global and Regional Crop Yield Predictions.

Abstract: Accurate predictions of crop yield are critical for developing effective agricultural and food policies at the regional and global scales. We evaluated a machine-learning method, Random Forests (RF), for its ability to predict crop yield responses to climate and biophysical variables at global and regional scales in wheat, maize, and potato in comparison with multiple linear regressions (MLR) serving as a benchmark. We used crop yield data from various sources and regions for model training and testing: 1) gridded global wheat grain yield, 2) maize grain yield from US counties over thirty years, and 3) potato tuber and maize silage yield from the northeastern seaboard region. RF was found highly capable of predicting crop yields and outperformed MLR benchmarks in all performance statistics that were compared. For example, the root mean square errors (RMSE) ranged between 6 and 14% of the average observed yield with RF models in all test cases whereas these values ranged from 14% to 49% for MLR models. Our results show that RF is an effective and versatile machine-learning method for crop yield predictions at regional and global scales for its high accuracy and precision, ease of use, and utility in data analysis. RF may result in a loss of accuracy when predicting the extreme ends or responses beyond the boundaries of the training data.

Subnational distribution of average farm size and smallholder contributions to global food production

Abstract: Smallholder farming is the most prevalent form of agriculture in the world, supports many of the planet's most vulnerable populations, and coexists with some of its most diverse and threatened landscapes. However, there is little information about the location of small farms, making it difficult both to estimate their numbers and to implement effective agricultural, development, and land use policies. Here, we present a map of mean agricultural area, classified by the amount of land per farming household, at subnational resolutions across three key global regions using a novel integration of household microdata and agricultural landscape data. This approach provides a subnational estimate of the number, average size, and contribution of farms across much of the developing world. By our estimates, 918 subnational units in 83 countries in Latin America, sub-Saharan Africa, and South and East Asia average less than five hectares of agricultural land per farming household. These smallholder-dominated systems are home to more than 380 million farming households, make up roughly 30% of the agricultural land and produce more than 70% of the food calories produced in these regions, and are responsible for more than half of the food calories produced globally, as well as more than half of global production of several major food crops. Smallholder systems in these three regions direct a greater percentage of calories produced toward direct human consumption, with 70% of calories produced in these units consumed as food, compared to 55% globally. Our approach provides the ability to disaggregate farming populations from non-farming populations, providing a more accurate picture of farming households on the landscape than has previously been available. These data meet a critical need, as improved understanding of the prevalence and distribution of smallholder farming is essential for effective policy development for food security, poverty reduction, and conservation agendas.

Nitrogen use in the global food system: past trends and future trajectories of agronomic performance, pollution, trade, and dietary demand

Abstract: Nitrogen (N) limits crop and grass production, and it is an essential component of dietary proteins. However, N is mobile in the soil-plant system and can be lost to the environment. Estimates of N flows provide a critical tool for understanding and improving the sustainability and equity of the global food system. This letter describes an integrated analysis of changes in N in human diets, N use efficiency (NUE) of cropping and livestock systems, N pollution and N in traded food and feed products for 12 world regions for the period 1960–2050. The largest absolute change in consumption of animal proteins during the period 1960–2009 is seen in China, while the largest share of animal protein per capita is currently observed in North America, Europe and Oceania. Due to the substantial growth of the livestock sector, about three quarters of contemporary global crop production (expressed in protein and including fodder crops and bioenergy byproducts) is allocated to livestock. Trends and levels of NUE and N surpluses in crop production are also diverse, as some regions show soil N depletion (developing regions, e.g. Africa), improving efficiency (industrialized regions, e.g. USA and Europe) and excessive N use (e.g. China, India). Global trade between the 12 regions has increased by a factor of 7.5 for vegetable proteins and by a factor of 10 for animal proteins. The scenarios for 2050 demonstrate that it would be possible to feed the global population in 2050 with moderate animal protein consumption but with much less N pollution, and less international trade than today. In such a scenario, optimal allocation of N inputs among regions to maximize NUE would further decrease pollution, but would require increased levels of N trade comparable to those in a BAU scenario.

Spatially explicit estimates of N2O emissions from croplands suggest climate mitigation opportunities from improved fertilizer management

Abstract: With increasing nitrogen (N) application to croplands required to support growing food demand, mitigating N2 O emissions from agricultural soils is a global challenge. National greenhouse gas emissions accounting typically estimates N2 O emissions at the country scale by aggregating all crops, under the assumption that N2 O emissions are linearly related to N application. However, field studies and meta-analyses indicate a nonlinear relationship, in which N2 O emissions are relatively greater at higher N application rates. Here, we apply a super-linear emissions response model to crop-specific, spatially explicit synthetic N fertilizer and manure N inputs to provide subnational accounting of global N2 O emissions from croplands. We estimate 0.66 Tg of N2 O-N direct global emissions circa 2000, with 50% of emissions concentrated in 13% of harvested area. Compared to estimates from the IPCC Tier 1 linear model, our updated N2 O emissions range from 20% to 40% lower throughout sub-Saharan Africa and Eastern Europe, to >120% greater in some Western European countries. At low N application rates, the weak nonlinear response of N2 O emissions suggests that relatively large increases in N fertilizer application would generate relatively small increases in N2 O emissions. As aggregated fertilizer data generate underestimation bias in nonlinear models, high-resolution N application data are critical to support accurate N2 O emissions estimates.

Learning large-scale plantation mapping from imperfect annotators

Abstract: Plantation mapping is important for understanding deforestation and climate change. Most existing plantation products rely heavily on visual interpretation of satellite imagery, which results in both false positives and false negatives. In this paper we aim to design an automatic framework that map plantations in large regions. Conventional classification methods cannot be directly applied due to the lack of ground-truth data. To this end, we propose a novel method that learns from multiple imperfect annotators. Since each annotator's labeling accuracy varies across different land covers due to his expertise and reference imagery, we model the annotator's reliability level to be associated with different types of locations. On the other hand, the temporal variation of land covers also greatly impacts the performance of conventional learning model. Therefore we utilize the remote sensing data which are available at multiple periods of a year and extend our proposed method by incorporating multi-instance learning. Finally, we show the superiority of the proposed method over multiple baselines in both synthetic dataset and real-world dataset. In addition, through several case studies we demonstrate that our method can achieve a better balance of precision and recall than the existing plantation products.