<p>This study presents an innovative, non-destructive remote sensing method for estimating sugarcane height using Landsat 8 imagery and meteorological and crop properties from 2018 to 2021 in the Shoeibeyeh area of Khuzestan province (Iran), at the field scale. Spectral bands from the visible to mid-infrared, along with thirty vegetation indices, meteorological data (cumulative temperature and sunny hours), and crop variety and age were used to model sugarcane height. The height of sugarcane was measured weekly across 178 fields, each with an average area of approximately 25 hectares. The dataset was rebalanced using the Synthetic Minority Over-Sampling Technique (SMOTE). Three popular machine learning (ML) methods were evaluated for regression models: Artificial Neural Networks (ANN), Random Forest (RF), and Support Vector Machines (SVM). Fifteen scenarios were analyzed, based on the dataset partitioning and these ML algorithms. Although all three ML methods performed well in estimating sugarcane height, RF performed slightly better than ANN and SVM. Based on model accuracy statistics and variable importance coefficients, the 5-RF-100-7 mode, with R² = 0.86 and NRMSE = 21.96% is selected without using the temperature variable. The innovative combination of scenarios 5 and 14 was used to adjust the importance coefficient of the temperature variable. The accuracy of the combined model improved, with R² = 0.87 and NRMSE = 19.45%, and the distribution of variable importance was more balanced. Therefore, the combined model, with a good sample distribution, effective selection of vegetation indices, and effective use of ML, produced an accurate estimate of sugarcane height at the field scale and can replace other methods.</p>