Soil mulching can effectively modify the crop growth environment and increase crop productivity in rainfed agriculture. Accurate estimation of crop evapotranspiration (ET), especially its transpiration (T) component, is crucial for understanding the crop water use and predicting crop yield in agricultural ecosystems. Nevertheless, direct measurement of T in the field is often difficult, expensive, destructive and time-consuming. Daily rainfed maize T under four mulching methods (NM: non-mulching, SM: straw mulching, RPBF: plastic-mulched ridge with bare furrow, and RPSF: plastic-mulched ridge with straw-mulched furrow) was obtained from sap flow measurements over four maize growing seasons (2015–2018) in Northwest China. A modified Jarvis-Stewart model (MJS) and a support vector machine model optimized by the whale optimization algorithm (SVM-WOA) were further proposed to estimate daily maize T based on solar radiation (R_s), vapor pressure deficit (VPD), soil water content (SWC) and leaf area index (LAI), which were compared to the simple multiple linear regression model (MLR). The three models were calibrated using data obtained in 2015 and 2017, and validated using data from 2016 and 2018. The measured seasonal T under SM, RPBF and RPSF was increased by 6.9–19.1%, 12.1–31.3% and 15.3–36.7% compared to that under NM, respectively. The SVM-WOA model (R² = 0.83–0.89, RMSE = 0.55–0.73 mm d⁻¹, MAE = 0.42–0.53 mm d⁻¹) was superior to the MJS model (R² = 0.61–0.79, RMSE = 0.75–1.12 mm d⁻¹, MAE = 0.58–0.88 mm d⁻¹) during validation, both of which greatly outperformed the MLR model (R² = 0.57–0.60, RMSE = 1.28–1.41 mm d⁻¹, MAE = 0.99–1.09 mm d⁻¹) under various mulching methods. The prediction accuracy of the SVM-WOA and MJS models was improved by 47–57% and 19–41% in terms of RMSE compared with that of the MLR model, respectively. Although the physically-based MJS model satisfactorily described the dynamics of rainfed maize T under various mulching methods, the blackbox-type SVM-WOA model was more suitable for estimating daily maize T after a careful calibration with adequate experimental data due to its advantage in modeling complex nonlinear relationships between T and its driving variables.