The retrieval methods of LAI based on traditional neural networks require a large number of training samples constructed from remote sensing parameter products or simulation data using the radiative transfer models. Furthermore, the training samples for the neural networks are sensor-specific. Therefore, the existing training samples for a sensor can not be directly applied to estimate LAI from remote sensing data acquired by other sensors. In addition, currently available LAI ground measurements considered as "truth value" are not used to construct training datasets of the neural networks to further improve the accuracy of the retrieved LAI values. In this paper, the LAI retrieval method based on deep transfer learning is proposed. A transplantable network is constructed by the Deep Belief Networks (DBN) and the LAI ground measurements are used as label to participate in the inversion process. The DBN is composed of Restricted Boltzmann Machine (RBM) and Back Propagation (BP) network. The RBM is used to extract the characteristic informations from surface reflectances, and BP is used to fine-tune network parameters. Firstly, the DBN model is pre-trained by the MODIS surface reflectances and the LAI values fused from the MODIS and CYCLOPES LAI products. Then, the all RBM layers’ parameters of the pre-trained DBN model are frozen, and the parameters of BP are fine-tuned by the small samples composed of VIIRS surface reflectances and LAI ground measurements. Finally, the fine-tuned DBN model is used to retrieve LAI. In order to estimate the retrieved results, the retrieved results at sites with different vegetation types are directly evaluated with the ground measurements and the global MODIS and VIIRS LAI products are compared with the retrieved results. The results indicate that for different biomes, the LAI retrieved results show reasonable seasonality. The direct validation results show that the retrieved LAI values with transfer learning are obviously superior to the retrieved LAI values without transfer learning. This study demonstrates that deep transfer learning can effectively retrieve LAI values from the VIIRS surface reflectance with limited LAI ground measurement samples and the existing training dataset.
Paper
Full text
Estimation of Leaf Area Index from VIIRS Surface Reflectance Based on Deep Transfer Learning
Semantic Scholar · Environmental Science · 2023
Abstract
The retrieval methods of LAI based on traditional neural networks require a large number of training samples constructed from remote sensing parameter products or simulation data using the radiative transfer models. Furthermore, the training samples for the neural networks are sensor-specific. Therefore, the existing training samples for a sensor can not be directly applied to estimate LAI from remote sensing data acquired by other sensors. In addition, currently available LAI ground measurements considered as "truth value" are not used to construct training datasets of the neural networks to further improve the accuracy of the retrieved LAI values. In this paper, the LAI retrieval method based on deep transfer learning is proposed. A transplantable network is constructed by the Deep Belief Networks (DBN) and the LAI ground measurements are used as label to participate in the inversion process. The DBN is composed of Restricted Boltzmann Machine (RBM) and Back Propagation (BP) network. The RBM is used to extract the characteristic informations from surface reflectances, and BP is used to fine-tune network parameters. Firstly, the DBN model is pre-trained by the MODIS surface reflectances and the LAI values fused from the MODIS and CYCLOPES LAI products. Then, the all RBM layers’ parameters of the pre-trained DBN model are frozen, and the parameters of BP are fine-tuned by the small samples composed of VIIRS surface reflectances and LAI ground measurements. Finally, the fine-tuned DBN model is used to retrieve LAI. In order to estimate the retrieved results, the retrieved results at sites with different vegetation types are directly evaluated with the ground measurements and the global MODIS and VIIRS LAI products are compared with the retrieved results. The results indicate that for different biomes, the LAI retrieved results show reasonable seasonality. The direct validation results show that the retrieved LAI values with transfer learning are obviously superior to the retrieved LAI values without transfer learning. This study demonstrates that deep transfer learning can effectively retrieve LAI values from the VIIRS surface reflectance with limited LAI ground measurement samples and the existing training dataset.