A comparative analysis of SLR, MLR, ANN, XGBoost and CNN for crop height estimation of sunflower using Sentinel-1 and Sentinel-2

Sustainable monitoring and determining the biophysical characteristics of crops is of global importance due to the increase in demand for food. In this context, remote sensing data provide valuable information on crops. This study investigates the relationship between the variables determined from both Synthetic Aperture Radar (SAR) and optical images and crop height. For this purpose, backscatter (σVH, σVV, σVH / σVV) and coherence (ϒVH, ϒVV) of multi-temporal dual-polarized Sentinel-1 and vegetation indices of multi-temporal Sentinel-2 data are analyzed. Two indices, namely, Normalized Difference Vegetation Index (NDVI) and NDVI with the red-edge band (NDVIred), are interpreted to identify the contribution of the red-edge band over the near-infrared band. The Zile District of Tokat province in Turkey where dominantly sunflower cultivation is carried out, was selected as the study area. In the analysis of the data, Simple Linear Regression (SLR), Multiple Linear Regression (MLR), Artificial Neural Network (ANN), EXtreme Gradient Boosting (XGBoost), and Convolutional Neural Network (CNN) were used. In the results of the study, ANN showed the lowest RMSE = 3.083 cm (RMSE%= 11.284) in the stem elongation period. The CNN followed the lowest RMSE for the Inflorescence development and flowering stages 19.223 cm (RMSE%=15.458) and 8.731 cm (RMSE%=5.821), respectively. In the ripening period, XGBoost achieved the lowest RMSE = 8.731 cm (RMSE%=6.091). All the best models in four methods were created using common variables of σVH, σVV, ϒVH, ϒVV and NDVIred, except ANN which exclude coherence variables. The results concluded that NDVIred contributed more than NDVI which is widely interpreted in previous studies.