Crop-Parameter Identification through Bayesian analysis for Regional-Scale Yield Estimation with AquaCrop

In building an environmentally harmonized food system, it is essential to examine measures to stabilize food production under climate change, grounded in quantitative evidence. This requires a framework that quantifies the impacts of climate change on crop production and utilizes those results to design crop production strategies. Approaches that estimate changes in yield with crop production models are widely used and are effective for identifying national-scale trends. However, uncertainty in regional-scale future projections persists because region-specific characteristics, such as soil conditions and farming practices, are not sufficiently represented. In addition, in process-based models, it is often difficult to define the numerous parameters governing crop growth in ways that are consistent with local soil and management conditions, and this parameter uncertainty strongly influences projections.

In this study, we performed Bayesian calibration of crop-file parameters in AquaCrop [1] using the Markov Chain Monte Carlo (MCMC) method to address the practical difficulty of specifying model parameters consistent with local conditions. To support this calibration, we developed a method to generate a region-specific characteristic dataset by integrating climate drivers with soil information as regional characteristics. AquaCrop, a process-based crop model developed by the Food and Agriculture Organization of the United Nations (FAO), was driven by daily maximum and minimum temperature, precipitation, and ETo. Climate forcing was obtained from historical data, and soil characteristics were derived from the Japanese Soil Inventory [2] provided by the National Agriculture and Food Research Organization (NARO), including soil maps and gridded property layers (e.g., saturated hydraulic conductivity and water-retention metrics such as pF-based water contents and available water capacity). We harmonized the coordinate systems of the climate and soil datasets and implemented a data-generation procedure to produce climate-grid-consistent regional inputs. This enables multi-year, multi-site calibration of crop parameters and subsequent yield simulations under local conditions.

Furthermore, to estimate key crop-file parameters from observations, we developed robust procedures for preparing daily weather time series. This includes standardizing the required variables and performing Bayesian estimation via MCMC. Using multi-year observed yields, we estimate posterior distributions of major crop parameters and quantify associated uncertainty. Using the developed system, we validate it with historical data and conduct yield projections under temperature-increase conditions, enabling evaluation of the contributions of climate warming and soil-mediated regional differences to yield changes.  By combining an integrated regional input data foundation with explicit treatment of crop-parameter uncertainty, this framework provides a basis for improving the reliability of regional-scale future projections.

[1] FAO (2024) AquaCrop, Version 7.2. Food and Agriculture Organization of the United Nations

[2] NARO, https://soil-inventory.dc.affrc.go.jp/