Table 1.
Mean annual (±1 SD) minimum (Tmin) and maximum (Tmax) air temperatures, wind speed, and total rainfall at each site. The number of years indicates the period of record, and total days (N) and wet days (Nwet) are counts across all years for each site. The sites were grouped for model calibration and validation.
Fig 1.
A total of 23 meteorological stations across Malaysia were selected for this study, comprising 16 sites in Peninsular Malaysia and seven in East Malaysia.
Of these, 12 and 11 stations were used for the model calibration and validation, respectively.
Fig 2.
Summary of goodness-of-fit metrics (NMAE, NMBE, KGE) and Anderson-Darling test p-values (AD p-level) for mean monthly minimum air temperature (Tmin) (a-d), maximum air temperature (Tmax) (e-h), wind speed (i-l), and total monthly rainfall (m-p), shown separately for the calibration (C) and validation (V) sites (n = 46 and 55 site-years, respectively).
NMAE: Normalized Mean Absolute Error; NMBE: Normalized Mean Bias Error; KGE: Kling-Gupta Efficiency.
Fig 3.
Comparison of correlations between the mean monthly maximum air temperature (Tmax) and monthly rainfall for the model-generated (gen.) and observed (obs.) data, shown separately for the calibration and validation sites (n = 46 and 55 site-years, respectively).
The dashed line indicates perfect agreement (1:1 line).
Fig 4.
Comparison of observed and model-generated monthly rainfall transition probabilities at calibration (a,b) and validation (c,d) sites (n = 552 and 660 site-months, respectively).
PWW (a,c) is the probability of a wet day following a wet day; PWD (b,d) is the probability of a wet day following a dry day. Each point represents one site-month. Dashed line indicates perfect agreement (1:1). Values in brackets are goodness-of-fit metrics Normalized Mean Absolute Error (NMAE), Normalized Mean Bias Error, and Kling-Gupta Efficiency (KGE).
Fig 5.
Quantile-Quantile (QQ) plots comparing daily model-generated (gen.) and observed (obs.) Values for minimum air temperature (Tmin) (a), maximum air temperature (Tmax) (b), wind speed (c), and rainfall (d), shown separately for the calibration and validation sites.
Rainfall (d) includes wet days only. The dashed line indicates perfect agreement (1:1 line). Sample sizes for calibration and validation sites, respectively, are: n = 16797 and 20097 for Tmin, Tmax, and wind speed; n = 8389 and 10587 for rainfall.
Fig 6.
Hexagonal bin plots showing agreement between observed (obs.) and model-generated (gen.) daily values of minimum air temperature (Tmin) (a), maximum air temperature (Tmax) (b), and wind speed (c).
The color intensity indicates the number of data points within each hexagonal bin, from low (blue) to high (red). The dashed line indicates perfect agreement (1:1 line). Data were combined from all calibration and validation sites (n = 36894).
Fig 7.
Empirical cumulative distribution function (ECDF) plots of daily weather generation errors at a representative calibration site (Banting, left) and validation site (Alor Setar, right).
Panels show (a,b) minimum (Tmin, blue dashed) and maximum (Tmax, red solid) air temperature, (c,d) wind speed, and (e,f) rainfall. Vertical dashed lines indicate zero error. Values in brackets show the percentage of days within specified error thresholds. Sample sizes for each site are provided in Table 1. Complete ECDF results for all 23 sites are in S1–S3 Figs.
Fig 8.
Distribution of observed daily rainfall classes (a) and cumulative percentage of model-generated values within increasing error thresholds for each rainfall class (b).
Data were aggregated from all calibration and validation sites (n = 18976).
Fig 9.
Quantile-Quantile (QQ) plots comparing daily minimum air temperature (Tmin) (a, b), maximum air temperature (Tmax) (c, d), wind speed (e, f), and rainfall (g, h) between model-generated (gen.) and observed (obs.) values for the Kerayong (first row) and Kalumpong (second row) sites.
The rainfall (g, h) included only wet days. The dashed line indicates perfect agreement between the generated and observed values (1:1 line). Sample sizes for Kerayong and Kalumpong sites, respectively, are: n = 8035 and 8400 for Tmin, Tmax, and wind speed; n = 4234 and 2779 for rainfall.
Fig 10.
Empirical cumulative distribution function (ECDF) plots comparing observed and model-generated daily minimum air temperature (Tmin) (a, b), maximum air temperature (Tmax) (c, d), wind speed (e, f), and rainfall (g, h) between model-generated and observed values for the Kerayong (first row) and Kalumpong (second row) sites.
Values in parentheses at the top left of each panel give the percentage of generated weather values that are within successive error bands. For Tmin and Tmax, the bands are ± 0.5, 1, and 2 °C (a-d); for wind speed, they are ± 0.5 and 1 m s ⁻ ¹ (e-f); and for rainfall, they are ± 5, 10, and 20 mm (g-h). The vertical dashed line indicates zero error. Sample sizes for Kerayong and Kalumpong sites, respectively, are: n = 8035 and 8400 for Tmin, Tmax, and wind speed; n = 4234 and 2779 for rainfall.
Fig 11.
Comparison of observed (obs.) (circles) and simulated (sim.) (red and blue lines) annual oil palm fresh fruit bunch (FFB) yield in kg dry matter (DM). Simulations at Kerayong (first row, a–c) were for planting densities of 136 (a), 160 (b), and 185 (c) palms ha-1, whereas simulations at Kalumpong (second row, d–f) were for 124 (d), 138 (e), and 150 (f) palms ha-1.
For each site, the simulated yields were driven by either the observed daily weather (solid blue line) or model-generated daily weather (red dashed line).
Fig 12.
Oil palm mean yield responses to rainfall intensity manipulations at Kerayong (left) and Kalumpong (right) sites for different planting densities.
Panels show (a,b) graduated reduction of higher-intensity (H) rainfall, (c,d) graduated H-to-lower-intensity (L) redistribution, and (e,f) four scenarios manipulating 50% of H rainfall: L loss, H loss, L → H, and H → L. Pink background indicates rainfall deficit; green indicates rainfall was conserved. Note: YAP is the number of years after field planting.
Table 2.
Root mean square error (RMSE, in mm) comparison of parametric distributions fitted to observed daily rainfall across all 23 sites. RMSE is calculated from differences between empirical and fitted quantiles across 100 percentiles. Gamma was fitted using method of moments (MoM) as required by MsiaGen, while Generalized Pareto Distribution (GPD) and Mixed Exponential (MixExp) were fitted using maximum likelihood estimation (MLE) to evaluate their best-case performance. Lower RMSE values indicate closer agreement between observed and predicted rainfall distributions.