Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images

Abstract

Forest canopy cover (FCC) plays an important role in many ecological, hydrological and forestry applications. For large-scale applications, FCC is usually estimated from remotely sensed data by inverting radiative transfer models (RTMs) or using data-driven regressions. In this study, we proposed a hybrid model, which combines a 3D RTM and transfer learning-based convolutional neural network (T-CNN), to estimate FCC from very high-resolution satellite images (e.g., Chinese GaoFen-2, 1 m resolution with 4 bands). Unlike common hybrid models that are purely trained with simulation data, T-CNN combines simulation data-based pre-training and actual data-based transfer learning, which is a widely used technique in artificial intelligence for finetuning models. The performance of T-CNN was compared with a random forest (RF) model and two general CNN models, including CNN trained with actual dataset only (Data-CNN) and CNN trained with RTM simulation data only (RTM-CNN). Results on the independent validation dataset (not used in training stage) showed that T-CNN had higher accuracy (RMSE = 0.121, R 2 = 0.83), compared with RF (RMSE = 0.26, R 2 = 0.61), Data-CNN (RMSE = 0.142, R 2 = 0.81), and RTM-CNN (RMSE = 0.144, R 2 = 0.73), which indicates that T-CNN has a strong transferability. Tests on different training sizes showed that T-CNN (0.084 < RMSE < 0.108) provided constantly better performances than RF (0.116 < RMSE < 0.122) and Data-CNN (0.103 < RMSE < 0.128), which demonstrates the potential of T-CNN as an alternative to RTM-based inversion and data-driven regressions to estimate FCC, especially when training data is imbalanced and inadequate.

Estimation of Forest Canopy Cover by Combining ICESat-2/ATLAS Data and Geostatistical Method/Co-Kriging

“…Nonparametric and parametric models are commonly employed for developing FCC estimation at the region scale [5], [20], [71]- [76]. Previously, studies of FCC estimation using remote sensing techniques in this study area have been carried out by academics: Fu et al [75] used PLSR to construct an FCC predictive model of Abies fabri Forest in the study area based on the texture factor of TM images in 2006, with rRMSE = 0.17 and accuracy of 83.30%, and the estimation results showed that the FCC was mostly distributed in the range of 0.6 ~ 0.7; Hu et al [76] extracted features after selecting bands in different ways and used PLSR to build an FCC estimation model based on Hyperion hyperspectral data for the main tree species, with R 2 = 0.83, rRMSE = 0.18, and an accuracy of 82.90%.…”

“…It is an important parameter of forest inventory, reflects forest structure characteristics and growth state [3], [4]. However, obtaining timely and reliable forest inventory data is often labor-intensive and time-consuming, especially across extensive areas [5], [6]. Remote sensing technology, which offers a practical and economical approach for measuring and monitoring vegetation cover and structure in large areas [7].…”

Prediction approach of larch wood density from visible–near-infrared spectroscopy based on parameter calibrating and transfer learning

“…In agriculture and forestry, applications of transfer learning include the following: first, the most common application was the identification of tree species, including the rapid identification of economical woods ( Li et al., 2022 ), pests, and quality defects ( Chen et al., 2020 ; Ahmad et al., 2021 ; Alencastre-Miranda et al., 2021 ). Second, TL was used for forest and farmland management and ecosystem status assessment ( Astola et al., 2021 ; Jin et al., 2021 ). Third, TL is used for the prediction of the properties of wood and agricultural products ( Singh et al., 2021 ).…”

An Integrated Method for Estimating Forest-Canopy Closure Based on UAV LiDAR Data

“…Forest-canopy closure (FCC) is defined as the proportion of the vertical projection area of the tree crown [1], which is an important parameter for the monitoring of forest resources often used to assess forest disturbance [2,3], forest structure characteristics and growth state [4,5], wildlife habitat and wildfire risk [6,7], and species richness [8]. The traditional method of FCC estimation relies on field measurement, which involves considerable labor, time, and energy [9,10]. Remote-sensing data can monitor land surfaces globally and promptly.…”