Showing papers by "James S. Gerber published in 2017"

Greenhouse gas emissions intensity of global croplands

Abstract: Global high-resolution crop-specific estimates of greenhouse gas emissions intensity (in 2000) reveal that certain cropping practices contribute disproportionately to emissions, making them suitable targets for climate mitigation policies. Stabilizing greenhouse gas (GHG) emissions from croplands as agricultural demand grows is a critical component of climate change mitigation1,2,3. Emissions intensity metrics—including carbon dioxide equivalent emissions per kilocalorie produced (‘production intensity’)—can highlight regions, management practices, and crops as potential foci for mitigation4,5,6,7. Yet the spatial and crop-wise distribution of emissions intensity has been uncertain. Here, we develop global crop-specific circa 2000 estimates of GHG emissions and GHG intensity in high spatial detail, reporting the effects of rice paddy management, peatland draining, and nitrogen (N) fertilizer on CH4, CO2 and N2O emissions. Global mean production intensity is 0.16 Mg CO2e M kcal−1, yet certain cropping practices contribute disproportionately to emissions. Peatland drainage (3.7 Mg CO2e M kcal−1)—concentrated in Europe and Indonesia—accounts for 32% of these cropland emissions despite peatlands producing just 1.1% of total crop kilocalories. Methane emissions from rice (0.58 Mg CO2e M kcal-1), a crucial food staple supplying 15% of total crop kilocalories, contribute 48% of cropland emissions, with outsized production intensity in Vietnam. In contrast, N2O emissions from N fertilizer application (0.033 Mg CO2e M kcal−1) generate only 20% of cropland emissions. We find that current total GHG emissions are largely unrelated to production intensity across crops and countries. Climate mitigation policies should therefore be directed to locations where crops have both high emissions and high intensities.

Incremental Dual-memory LSTM in Land Cover Prediction

Abstract: Land cover prediction is essential for monitoring global environmental change. Unfortunately, traditional classification models are plagued by temporal variation and emergence of novel/unseen land cover classes in the prediction process. In this paper, we propose an LSTM-based spatio-temporal learning framework with a dual-memory structure. The dual-memory structure captures both long-term and short-term temporal variation patterns, and is updated incrementally to adapt the model to the ever-changing environment. Moreover, we integrate zero-shot learning to identify unseen classes even without labelled samples. Experiments on both synthetic and real-world datasets demonstrate the superiority of the proposed framework over multiple baselines in land cover prediction.

Declining spatial efficiency of global cropland nitrogen allocation

Abstract: Efficiently allocating nitrogen (N) across space maximizes crop productivity for a given amount of N input and reduces N losses to the environment. Here we quantify changes in the global spatial efficiency of cropland N use by calculating historical trade-off frontiers relating N inputs to possible N yield assuming efficient allocation. Time series cropland N budgets from 1961 to 2009 characterize the evolution of N input-yield response functions across 12 regions and are the basis for constructing trade-off frontiers. Improvements in agronomic technology have substantially increased cropping system yield potentials and expanded N-driven crop production possibilities. However, we find that these gains are compromised by the declining spatial efficiency of N use across regions. Since the start of the Green Revolution, N inputs and yields have moved farther from the optimal frontier over time; in recent years (1994–2009), global N surplus has grown to a value that is 69% greater than what is possible with efficient N allocation between regions. To reflect regional pollution and agricultural development goals, we construct scenarios that restrict reallocation, finding that these changes only slightly decrease potential gains in nitrogen use efficiency. Our results are inherently conservative due to the regional unit of analysis, meaning a larger potential exists than is quantified here for cross-scale policies to promote spatially efficient N use.

Predict land covers with transition modeling and incremental learning

Abstract: Successful land cover prediction can provide promising insights in the applications where manual labeling is extremely difficult. However, traditional machine learning models are plagued by temporal variation and noisy features when directly applied to land cover prediction. Moreover, these models cannot take fully advantage of the spatio-temporal relationship involved in land cover transitions. In this paper, we propose a novel spatio-temporal framework to discover the transitions among land covers and at the same time conduct classification at each time step. Based on the proposed model, we incrementally update the model parameters in the prediction process, thus to mitigate the impact of the temporal variation. Our experiments in two challenging land cover applications demonstrate the superiority of the proposed method over multiple baselines. In addition, we show the efficacy of spatio-temporal transition modeling and incremental learning through extensive analysis.