A machine learning–coupled APSIM model pipeline for projected oil palm yield in Surat Thani, Thailand

Napat Jantaraprasit, Parichart Promchote, Shih–Yu Simon Wang, Sugontee Daengnui, Sajad Khoshnood Motlagh, Andre Geraldo de Lima Moraes, Luthiene França, Jin–Ho Yoon, Chalermpol Phumichai and Piya Kittipadakul

PLOS ONE, 2026, vol. 21, issue 6, 1-19

Abstract: Accurate and timely forecasts of oil palm yield are essential for both short-term farm management and long-term adaptation planning, yet their reliability is often constrained by the coarse spatial resolution of climate datasets and structural biases in process-based crop models. To address these challenges, we developed an end-to-end modeling framework that integrates spatially refined climate information with a hybrid process–machine-learning approach. Our method employs Spatial Interactions Downscaling to convert reanalysis, seasonal forecasts, and CMIP6 climate projections into fine-scale datasets anchored to the CHELSA baseline. These downscaled drivers are then coupled with the Agricultural Production Systems sIMulator (APSIM) and a Random Forest (RF) model to correct residual errors and improve predictive accuracy. A case study in Surat Thani, Thailand, demonstrates the framework’s performance and utility. Downscaled climate variables showed strong agreement with CHELSA, with minimal bias and compact error distributions, especially for temperature. Stand-alone APSIM overestimated yields (RMSE = 15.51 t ha ⁻ ¹), whereas the APSIM + RF hybrid significantly improved accuracy (RMSE = 5.52 t ha ⁻ ¹ at observed sites; 2.74 t ha ⁻ ¹ when averaged across sites). Seasonal forecasts based on downscaled data achieved skill levels comparable to those driven by reanalysis, enabling reliable yield predictions up to eight months in advance. On centennial scales, CMIP6 projections suggest stable to slightly higher yields in the early 21st century, a modest mid-century decline, and late-century stabilization across scenarios. These results indicate that oil palm production in southern Thailand is relatively resilient to projected climate change. More broadly, the framework offers a transferable approach for integrating fine-scale climate information and hybrid modeling to improve crop forecasting, support climate-risk assessment, and inform adaptation strategies across agricultural systems.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0349782 (text/html)
https://journals.plos.org/plosone/article/file?id= ... 49782&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pone00:0349782

DOI: 10.1371/journal.pone.0349782

Access Statistics for this article

More articles in PLOS ONE from Public Library of Science
Bibliographic data for series maintained by plosone ().