Time-ordered empirical framework for forecasting climate-driven oil palm yield in southern Thailand

Abstract

Importance of the work: Oil palm yield in southern Thailand is strongly sensitive to
climatic variations; however, robust intra-annual forecasting frameworks are largely absent.
Objectives: To quantify lagged climate-yield relationships and to build province-scale
empirical models for reliable intra-annual yield forecasting.
Materials and Methods: Monthly yield data were compiled for 13 southern provinces
(2005–2022) to derive province-averaged reanalysis climate variables with lag adjustments.
Collinearity was reduced using variance inflation factor analysis and models were trained
through exhaustive search with expanding-window cross-validation.
Results: Robust models with 2–8 predictors were produced for 11 provinces. Across test
splits, the ranges were: coefficient of determination (R2
), 0.33–0.73; root mean square
error, 0.14–0.30 t/ha; coefficient of variation of the root mean square error, 8.38–23.75 %;
and mean bias error, within ±0.13 t/ha. Dew-point temperature, soil moisture at 1.6 m
and precipitation dominated the selected predictors, with common lags near 3 mth and
21 mth. Forecast lead time spanned 3–20 mth, most often 6–11 mth ahead of harvest.
The lowest errors occurred in Nakhon Si Thammarat and Surat Thani provinces, while
southernmost provinces had higher relative errors. Partial-R² attribution highlighted
mid-season soil moisture (often a 7 mth lag) as the most influential predictor in 7 of the
11 provinces.
Main finding: A time-ordered empirical framework successfully forecast province-scale oil
palm yields several months in advance using reanalysis climate and soil-moisture predictors.
The results could support operational early warning for mills and agencies, while also
clarifying the lagged climatic controls that shape yield variability in southern Thailand.