Estimating biomass of winter oilseed rape using vegetation indices and texture metrics derived from UAV multispectral images

Monitoring the above ground biomass (AGB) in winter oilseed rape is important for improving the agronomic management efficiency and predicting yield to ensure edible oil and biofuel supplies. The Yangtze River Basin in China accounts for one-fifth of the rapeseed yield in the world. However, the fragmented farming lands in the Yangtze River Basin make it difficult to accurately monitor winter oilseed rape growth at large scale using the medium-resolution satellite data. The low-altitude unmanned aerial vehicle (UAV) provides a feasible way to accurately and non-destructively estimate AGB in winter oilseed rape at the plot level. In this study, we evaluated the contributions of vegetation indices (VIs) and texture metrics, derived from multispectral images captured by the camera mounted on a UAV, to predict the AGB in winter oilseed rape. The AGB was estimated with (1) multiple VIs and (2) multiple VIs and texture metrics. The partial least square regression (PLSR) and random forest (RF) regression models were trained with datasets of the experiment in 2016-2017 and 2017-2018, and then applied to predict AGB of the 2018-2019 growth season. Results demonstrated that the incorporation of texture metrics to both PLSR and RE models provided more accurate estimations of AGB in winter oilseed rape than the models based solely on VIs. The accuracy of the AGB predicted by the RF regression model using VIs and texture metrics (RMSE = 274.18 kg/ha for the validation dataset) was slightly higher than the results of the PLSR model (RMSE = 284.09 kg/ha for the validation dataset). According to the evaluation of the important variables, the red edge chlorophyll index (CTred (edge)) and ratio vegetation index (RVI) were selected as the most important input features by PLSR and RF regression models. The normalized difference vegetation index (NDVI) contrast was selected as an important texture metrics for the AGB prediction, indicating that NDVI contrast could be a sensitive indicator of the spatial distribution of shadow caused by the different amount of biomass. This study suggested the great potential of UAV in estimating the plot-level AGB by combining VIs and texture metrics.