Rice systems as a stress test for hydrological and methane modelling: Developing rice water (RIWA) benchmark dataset for remote sensing of soil moisture and water levels in rice fields across Asia

 

 

Rice-based agricultural systems account for the largest share of global agricultural water use and are a major source of methane emissions, yet their hydrological dynamics remain among the least constrained water fluxes in hydrological, land-surface, and greenhouse gas models. This is largely due to difficulty of cloud cover in optical data use in monsoon regions and a lack of validation data for natural and irrigation related soil moisture dynamics due to insufficient ground truthing data and poor resulting satellite product quality.

The implications are becoming more acute due to climate change as major rice-growing regions in India are shifting their practices and adapting to new realities by either decreasing or increasing water use. These further increases the uncertainties in already coarse irrigation and soil moisture products that currently drive models. With rice being both severely affected by climate change and methane emissions from water management in rice fields a key mitigation opportunity – these uncertainties propagate into hydrological and emission assessment globally and locally. Fortunately, recent advances in remote sensing such as AI-driven embeddings such as a AlphaEarth, new satellites such as the NISAR L-Band, and continuously increasing computational power and deep learning, promise rapid improvements in filling this crucial data gap – but to materialize these promises, ground truthing benchmark datasets will be required to adequately validate and compare these new approaches.

Here, we present our Rice Water Benchmark (RIWA) dataset that were are currently developing across India, Cambodia, the Philippines and Vietnam with multiple colleagues and partners. The dataset contains sub-weekly soil moisture and water level readings from more than 300 rice fields across multiple seasons that capture spatial and temporal heterogeneity of soil water status. We further present initial results for how these datasets can be used to evaluate different remote sensing approaches for predicting soil moisture and water management in rice fields – that can also be applied to other crops – and how this matters for hydrological and methane modelling applications.

Besides, we discusses challenges for data quality that include consistency, deployment of low-cost devices, spatial representativeness and the need for auxiliary data such as irrigation events timings from regular phone surveys. By developing harmonized and transparent global datasets for water use in agriculture will be crucial to fully utilize the promise of advances in remote sensing, digital hydrology and digital agriculture and the use of AI for global cereal systems, of which rice provides an important stress test due to its complex water management regimes.