Lazin et al. (2025) Climate-Informed flood damage assessment in the cropland area across the midwestern USA

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-12
• Authors: Rehenuma Lazin, Xinyi Shen, Emmanouil N. Anagnostou
• DOI: 10.1038/s41598-025-27288-z

Research Groups

• Atmospheric, Earth, and Energy Division (AEED), Lawrence Livermore National Laboratory, Livermore, CA, USA
• School of Freshwater Sciences, University of Wisconsin-Milwaukee (UWM), WI, USA
• Department of Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA

Short Summary

This study developed a climate-informed convolutional neural network (CNN) model to assess flood damages in corn and soybean croplands across the midwestern USA, projecting future damages ranging from a 40% decrease to a 120% increase by mid-century under CMIP5 climate scenarios.

Objective

• To develop and apply a climate-informed convolutional neural network (CNN) model to estimate and project flood damages (in hectares) in corn and soybean cropland areas across the midwestern USA for historical (1976-2005) and future mid-century (2041-2070) periods under CMIP5 climate scenarios.

Study Configuration

• Spatial Scale: Midwestern USA (Iowa, Illinois, Minnesota, and Missouri counties). Meteorological data downscaled to 4 km spatial resolution. Damage estimated at the county level.
• Temporal Scale:
  • Model training and evaluation: 2008-2020.
  • Historical baseline period: 1976-2005.
  • Future mid-century period: 2041-2070.
  • Daily meteorological data aggregated to 3-day intervals for model input, analyzed for early (May-June) and late (July-November) seasons.

Methodology and Data

• Models used:
  • Climate-informed Convolutional Neural Network (CNN) model.
  • Coupled Model Intercomparison Project 5 (CMIP5) global climate models (e.g., CNRM-CM5, CanESM2, IPSL-CM5B-LR, MIROC-ESM-CHEM, MIROC-ESM, BNU-ESM).
  • Multivariate Adaptive Constructed Analogs (MACA) for downscaling CMIP5 projections.
• Data sources:
  • Meteorological variables (precipitation, maximum/minimum temperature, solar radiation, specific humidity): gridMET dataset (2008-2020) and MACA-downscaled CMIP5 projections.
  • Topographical predictor: Height Above Nearest Drainage (HAND).
  • Cropland masks: Annual cropland maps (2008-2020).
  • Flood-damaged corn and soybean area (target data): Risk Management Agency (RMA) at the U.S. Department of Agriculture (USDA RMA) (2008-2020).

Main Results

• The CNN model demonstrated robust performance, with relative errors within ±25% for most counties in unseen years and high correlation coefficients (R > 0.8) for high-damage counties. Root Mean Square Error (RMSE) for early-season corn prediction was approximately 3,237 hectares, and for early-season soybean, it was around 1,619 hectares.
• Future flood damages (2041-2070) in Midwestern USA croplands are projected to vary significantly, ranging from a 40% decrease to a 120% increase compared to the historical baseline (1976-2005), depending on the CMIP5 model and Representative Concentration Pathway (RCP) scenario (4.5 and 8.5).
• Most climate models project higher damages for both corn and soybean in Iowa counties during the early season.
• Conversely, a consensus among models suggests decreasing damages in Minnesota counties during late-season floods for both crops.
• During the late season, temperature, humidity, and radiation are critical factors influencing crop losses. Higher temperatures and radiation are detrimental to both crops, while high humidity increases corn damage and low humidity increases soybean damage.
• Model agreement maps indicate that over 70% of models project increasing early-season flood damages for both corn and soybean in Iowa.

Contributions

• Developed and applied a climate-informed Convolutional Neural Network (CNN) model to project future flood-related crop damages at the county level, overcoming limitations of traditional physical models.
• Provided quantitative projections of future flood damage trends (percentage change) for corn and soybean in the Midwestern USA under CMIP5 climate change scenarios (RCP 4.5 and 8.5).
• Identified significant regional and seasonal disparities in projected damages and elucidated the influence of specific meteorological variables (temperature, humidity, radiation) on late-season crop losses.
• Offers valuable insights for farmers, stakeholders, and policymakers to inform future risk management, adaptation strategies (e.g., adjusting crop calendars, insurance planning), and initiatives to enhance food security.

Funding

• Eversource Energy Center (partnership of Eversource and the University of Connecticut).
• United States Department of Agriculture (USDA) for data support.
• Lawrence Livermore National Laboratory (LLNL) under Contract DE-AC52-07NA27344.
• IM release number: LLNL-JRNL-871284.

Citation

@article{Lazin2025ClimateInformed,
  author = {Lazin, Rehenuma and Shen, Xinyi and Anagnostou, Emmanouil N.},
  title = {Climate-Informed flood damage assessment in the cropland area across the midwestern USA},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-27288-z},
  url = {https://doi.org/10.1038/s41598-025-27288-z}
}