Times are displayed in (UTC-05:00) Central Time (US & Canada) Change

248-5 Integrating Earth Intelligence Into Climate-Driven Estimation of Multi-Cropping Rice Planting Dates Using a Geographical Random Convolutional Kernel Transform

Session: Expanding Geology’s Horizons: Geoinformatics, Open Science, and Open Data

Presenting Author:

Hanchen Zhuang

Authors:

Zhuang, Hanchen¹, Wu, Sensen², Gao, Song³, Du, Zhenhong⁴

(1) School of Earth Sciences, Zhejiang University, Hangzhou, Zhejiang, China; Department of Geography, University of Wisconsin–Madison, Madison, WI, USA, (2) School of Earth Sciences, Zhejiang University, Hangzhou, Zhejiang, China, (3) Department of Geography, University of Wisconsin–Madison, Madison, WI, USA, (4) School of Earth Sciences, Zhejiang University, Hangzhou, Zhejiang, China,

Abstract:

Accurate, dynamic prediction of rice planting dates under changing climates is essential for optimizing yield, managing stress exposure and informing adaptive strategies in multi-cropping systems. We introduce Geo-ROCKET, a spatially explicit, data-driven framework that couples random convolutional kernel transforms with adaptive neighborhood sampling to learn climate-driven planting dynamics directly from high-resolution agrometeorological reanalysis and gridded rice-calendar observations. Unlike static rule-based calendars, Geo-ROCKET models successive planting windows across single, double and triple cropping seasons, explicitly linking each season’s harvest to the subsequent planting decision. Applied across monsoon Asia, Geo-ROCKET reduced area-weighted mean absolute error to 7.0 days in single-crop systems, 6.5 days in double-late crops and 8.5–12.5 days in triple-crop seasons, outperforming six benchmark algorithms (ROCKET, LightGBM, XGBoost, KNN, Random Forest, Ridge). Sensitivity analyses revealed that optimal climate-window lengths and neighborhood sizes vary by season, and that even small harvest-date perturbations can amplify downstream errors by over 60%. Geo-ROCKET maintains robust performance under data sparsity and scales efficiently over heterogeneous grids. By providing dynamic, climate-compatible planting-date inputs, Geo-ROCKET offers a generalizable tool for integrating adaptive planting strategies into global and regional crop simulation models, thereby enhancing the capacity of biological systems models to capture climate-driven phenological shifts.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-8588

Integrating Earth Intelligence Into Climate-Driven Estimation of Multi-Cropping Rice Planting Dates Using a Geographical Random Convolutional Kernel Transform

Description

Session Format: Oral

Presentation Date: 10/22/2025

Presentation Start Time: 09:35 AM

Presentation Room: HBGCC, 301C

Back to Session