Critical Analysis of Machine Learning Approaches for Vegetation Fractional Cover Estimation Using Drone and Sentinel-2 Data

The accuracy of estimated fractional vegetation cover (FVC) depends on selecting the best suitable input features, precise ground information, and a prediction model. Therefore, in this paper, four machine learning (ML) algorithms, namely Support Vector Regression (SVR), Random Forest Regression (RFR), K-Nearest Neighbors (KNN), and Linear Regression (LR), are used for FVC estimation using different vegetation indices (VI) as input features. Estimated FVC is compared with the ground truth FVC, which has been calculated with the high-resolution drone images, and their R-square values are calculated. R-square value is used for the assessment of the best input features and model. RFR and KNN emerge as the best suitable ML models in comparison to SVR and LR. The obtained R-square values for the RFR model with the input features used as NDVI, PAVI, SAVI, and MSAVI are 0.873, 0.869, 0.869, and 0.862, respectively. NDVI, PAVI, SAVI, and MSAVI perform better in comparison to other features. Hence, they are permuted together and used as input features, and their results on all the algorithms are moderately better. The highest R-square value obtained is 0.878 when SAVI and PAVI are used as input features for the RFR model.