Global greenhouse gas emissions from agriculture: pathways to sustainable reductions
Data files
Dec 29, 2024 version files 35.75 MB
-
dryad3.zip
35.75 MB
-
README.md
4.33 KB
Abstract
We synthesized global data from the Food and Agriculture Organization (FAO) of the United Nations between 1990 and 2021 to analyze the impacts of agricultural activities on global GHG emissions from agricultural land. We obtained predictive estimates of agricultural GHG emissions for the future period of 2022 to 2050 using a deep learning model.
README: Global greenhouse gas emissions from agriculture: pathways to sustainable reductions
https://doi.org/10.5061/dryad.73n5tb36k
Description of the data and file structure
Files and variables
File: dryad.zip
Description: Agricultural greenhouse gas emissions (reported as CO2 equivalents) are the sum of CO2, CH4, and N2O emissions from agricultural land, including emissions from land use change. Forest loss is the net conversion of forestland area due to agricultural uses. Livestock includes buffalo, camels, cattle, chickens, donkeys, goats, horses, mules, hinnies, sheep, and swine/pigs. Inorganic nitrogen fertilizers are the sum of the nitrogen content from urea, ammonium sulfate, ammonium nitrate, calcium ammonium nitrate, and other mixtures with calcium carbonate, sodium nitrate, urea and ammonium nitrate solutions, anhydrous ammonia, and other not elsewhere classified inorganic nitrogen (N) fertilizers. Crop residues are the sum of N content in crop residues and forage left on agricultural land. Irrigation is the land area equipped for irrigation.
Variables, units, and FAO database:
livestock, billion livestock unit, https://www.fao.org/faostat/en/#data/EK
fert, inorganic nitrogen fertilizer, million ton, https://www.fao.org/faostat/en/#data/RFN
residue, crop residue nitrogen left on the field, million tons, https://www.fao.org/faostat/en/#data/GCE
irg, irrigation area, million hectares, https://www.fao.org/faostat/en/#data/RL
forest, net forest loss, million hactares, https://www.fao.org/faostat/en/#data/RL
GHG, greenhouse gas, GtCO2eq, https://www.fao.org/faostat/en/#data/GT
Model Architecture
The input layer corresponds to the number of features and time steps in the dataset, forming the entry point for sequential data in the model. A series of LSTM layers are implemented to capture the complex patterns in the data. The number of neurons in the LSTM layers decreases progressively (256, 128, 64, 32) to refine the patterns as the data moves deeper into the network. Dropout layers are incorporated after each LSTM layer to reduce the risk of overfitting by randomly disabling a fraction (30%) of the neurons during training. The output from the final LSTM layer is further refined through two dense layers. The second dense layer outputs the final value.
Software and Libraries
The model was implemented using the TensorFlow and Keras libraries in Python. Required libraries were listed at the beginning of the script.
input.csv is the input dataset for modeling (Data source: FAO). The FAO data are under a CC-BY-4.0 license. Dryad cannot publish data that is under a CC-BY-4.0 license. The input data were removed upon Dryad editor's request. The input data can be downloaded from FAO or be available upon request to the corresponding author.
predicting model.ipynb is the Python code for the deep learning model;
best_model folds 1-10.keras are the trained deep learning models;
future_ghg predictions sets 1-3.csv are the predicted values;
loss function plot.jpg and observed vs predicted.jpg are the visualization of the model performance;
SEM.ipynb is the Python code for structural equation modeling;
results.csv and stats.csv are the output of the structural equation modeling;
SEM.jpg is the visualization of the structural equation model;
bootstrap_results.csv is the bootstrap resampling results of the structural equation model.
Code/software
We used the Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN) as the algorithm for modeling using Keras in TensorFlow (Python 3.12). The required libraries were listed at the beginning of the script.
Access information
Data was derived from the following sources:
- Food and Agriculture Organization (FAO) of the United Nations (https://www.fao.org/faostat/en/#data)
Acknowledgment
I thank FAO for the publicly available database and Dr. Francesco Tubiello for guidance on the use of FAO data.
Methods
The historical data for the period of 1990–2021 were obtained from FAO. Data were accessed in September 2024. Data are available by year and country, with global coverage. We compiled agricultural GHG emissions and the related agricultural activities (net forest loss to agricultural use, livestock production, inorganic N fertilizer application, crop residue left on agricultural land, and irrigation) for up to 214 countries and territories.