Determining an appropriate time scale and lag time to predict coffee yield at the local scale based on vegetation health index and meteorological drought indices

Abstract

This study aimed to improve a procedure to determine an appropriate time scale and lag time to predict coffee yield at the local scale. The Vegetation Health Index (VHI), Effective Drought Index, Standardised Precipitation Index, Standardised Precipitation Evapotranspiration Index (SPEI), and soil moisture from 2000 to 2022 in Dak Lak, Vietnam, were selected for the analysis. The yield differences between drought and wet phases and the correlation coefficients between the indices and yield were analysed to determine potential predictors for the model. Then, a stepwise multiple linear regression model with the leave-one-out cross-validation was performed to select appropriate predictors with their time scales and lag times. The results showed that VHI with a lag time from seven to nine months before harvest (VHI7-9) and SPEI at a time scale of five months with a lag time of ten months before harvest (SPEI510) had essential contributions in predicting coffee yield. Meanwhile, soil moisture had a poor contribution. Coffee yield could be predicted from three to nine months before harvest based on meteorological drought indices, VHI, and soil moisture. With a reasonably long prediction time and relatively high accuracy, the proposed prediction procedures may be applied to mitigate climate change’s impacts, aiming for sustainable development.