Comparison of GF2 and SPOT6 Imagery on Canopy Cover Estimating in Northern Subtropics Forest in China

Abstract

Canopy cover is an important vegetation attribute used for many environmental applications such as defining management objectives, thinning and ecological modeling. However, the estimation of canopy cover from high spatial resolution imagery is still a difficult task due to limited spectral information and the heterogeneous pixel values of the same canopy. In this paper, we compared the capacity of two high spatial resolution sensors (SPOT6 and GF2) using three ensemble learning models (Adaptive Boosting (AdaBoost), Gradient Boosting (GDBoost), and random forest (RF)), to estimate canopy cover (CC) in a Chinese northern subtropics forest. Canopy cover across 97 plots was measured across 41 needle forest plots, 24 broadleaf forest plots, and 32 mixed forest plots. Results showed that (1) the textural features performed more importantly than spectral variables according to the number of variables in the top ten predictors in estimating canopy cover (CC) in both SPOT6 and GF2. Moreover, the vegetation indices in spectral variables had a lower relative importance value than the band reflectance variables. (2) GF2 imagery outperformed SPOT6 imagery in estimating CC when using the ensemble learning model in our data. On average across the models, the R2 was almost 0.08 higher for GF2 over SPOT6. Likewise, the average RMSE and average MAE were 0.002 and 0.01 lower in GF2 than in SPOT6. (3) The ensemble learning model showed good results in estimating CC, yet the different models performed a little differently in the results. Additionally, the GDBoost model performed the best of all the ensemble learning models with R2 = 0.92, root mean square error (RMSE) = 0.001 and mean absolute error (MAE) = 0.022.

Evaluating the Performance of Hyperspectral Leaf Reflectance to Detect Water Stress and Estimation of Photosynthetic Capacities

“…Although SVM has the advantage of handling high-dimensionality data and does well with a limited training dataset [86], it performed poorly in comparison with RF in this study. GDboost exhibited severe overfitting and returned unreliable predictions, although the literature reported the GDBoost model performed better than RF when estimating forest coverage [87]. It may be concluded that GDBoost presented major flaws in modeling highly correlated hyperspectral data.…”

Hyperspectral sensing of photosynthesis, stomatal conductance, and transpiration for citrus tree under drought condition

“…Although SVM has advantage of handling high dimensionality data and do well with a limited training dataset (Ma et al 2019), it performed poorly in comparison with RF in this study. GDboost exhibited severe overfitting and returned unreliable prediction although literature reported GDBoost model performed better RF when estimating forest coverage (Zhou et al 2020a). It may be concluded that GDBoost present major flaws in modeling highly correlated hyperspectral data.…”

“…The GDboost Algorithm tries to match the residual error of the new predictor with the previous one, while the Adaboost algorithm corrects the unfitness of the training instance through the previous training. The main difference between GDboost and Adaboost was how they deal with the underfitted values (Zhou et al 2020a). The number of trees (ntree) and learning rate (learning_rate) were tuned in AdaBoost and GDBoost models.…”

ClouDet: A Dilated Separable CNN-Based Cloud Detection Framework for Remote Sensing Imagery

“…Table Ⅰ shows the relevant information of each band of Landsat 8. However, many platforms on satellites are only equipped with visible and near-infrared bands, such as ZY-3 [34] and GF-2 [35], which provides limited feature information to detect clouds in remote sensing imagery and make the task more difficult. In order to cover the cloud detection tasks on most of optical satellite platforms, only four spectral bands-Band 2 to Band 5 are used in this work.…”