Scaling Up Rice Methane Fluxes in Monsoon Asia Through Data-Driven Model
Abstract
Rice cultivation is an important anthropogenic methane source to the atmosphere, contributing about 8% of total global anthropogenic emissions, and large uncertainties still exist in bottom-up estimates. The geographical distribution of rice emissions has been assessed at regional-to-global scales by bottom-up inventories and land surface models over coarse spatial units. However, high-resolution flux estimates capable of capturing local climatic and management, as well as replicate in situ data remain challenging to produce. To fill this gap, we use rice methane flux data from 23 global eddy covariance sites (45 growing seasons) and geospatial datasets with machine learning to 1) evaluate data-driven model performance and predictor importance to predict rice CH4 fluxes; 2) produce gridded up-scaling estimates of rice CH4 emissions at high resolution (500-1000m) in Monsoon Asia, where >80% of global rice is cultivated.
Our random forest model is trained on CH4 and bioclimatic data at 8-day intervals. We used a leave-one-site-out procedure to reduce over-fitting. Our preliminary model selects 20 best predictors from an initial set of 210. Air temperature and land surface temperature at night are the most important predictor followed by MODIS indices that reflect greenness and surface water conditions (EVI, SRWI, and LSWI). The model reproduces much of the 8-day average flux variation, albeit with substantial error (R2 = 0.47, MAE = 63.8 nmol m-2 s-1, Bias = 1.6 nmol m-2 s-1). The model also predicts well the mean seasonal cycles (R2=0.54) and the site means (R2=0.52).- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMB038.0014Y
- Keywords:
-
- 0414 Biogeochemical cycles;
- processes;
- and modeling;
- BIOGEOSCIENCES;
- 0426 Biosphere/atmosphere interactions;
- BIOGEOSCIENCES;
- 0475 Permafrost;
- cryosphere;
- and high-latitude processes;
- BIOGEOSCIENCES;
- 0497 Wetlands;
- BIOGEOSCIENCES