Predicting Nitrous Oxide Emission From China's Waterbodies With Multiple Deep Learning Algorithms

Abstract Many studies have been conducted on the prediction of nitrous oxide (N 2 O) emissions from soils. Comparably, prediction of N 2 O water–air emissions is much more limited, especially at the national level. Here, we collected published N 2 O emission data across China's watersheds and analyzed spatiotemporal patterns during dry and wet seasons. We predicted N 2 O emission fluxes from these waterbodies for 2026–2028 using a traditional gray prediction model (GM) coupled with several deep learning models: Long Short‐Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional Long Short‐Term Memory (BiLSTM). The study showed large regional variation in emissions from subtropical to boreal watersheds. Average emission rates varied from 13.95 (±27.15) μg m −2 h −1 in the Yellow River Basin to 68.71 (±102.62) μg m −2 h −1 in Southwest China. N 2 O emissions were clearly higher in the dry season than the wet season in all regions except the Yellow River Basin, indicating strong influence from wetland vegetation. Regarding model performance, higher accuracy was achieved by GRU and BiLSTM, which successfully predicted fluctuating increases of N 2 O emission fluxes in most regions from 2026 to 2028, reflecting seasonal changes. While LSTM performed less accurately, GRU and BiLSTM, evolved from LSTM, may be more appropriate for complex situations. These findings provide insights into national spatiotemporal patterns of N 2 O emissions and can guide regional and national mitigation strategies as well as future research.