Search

Deep learning (DL) models are gaining popularity in forest variable prediction using Earth observation (EO) images. However, in practical forest inventories, reference datasets are often represented by plot- or stand-level measurements, while high-quality representative wall-to-wall reference data for end-to-end training of DL models are rarely available. Transfer learning facilitates expansion of the use of deep learning models into areas with sub-optimal training data by allowing pretraining of the model in areas where high-quality teaching data are available. In this study, we perform a "model transfer" (or domain adaptation) of a pretrained DL model into a target area using plot-level measurements and compare performance versus other machine learning models. We use an earlier developed UNet based model (SeUNet) to demonstrate the approach on two distinct taiga sites with varying forest structure and composition. The examined SeUNet model uses multi-source EO data to predict forest height. Here, EO data are represented by a combination of Copernicus Sentinel-1 C-band SAR and Sentinel-2 multispectral images, ALOS-2 PALSAR-2 SAR mosaics and TanDEM-X bistatic interferometric radar data. The training study site is located in Finnish Lapland, while the target site is located in Southern Finland. By leveraging transfer learning, the SeUNet prediction achieved root mean squared error (RMSE) of 2.70 m and R2 of 0.882, considerably more accurate than traditional benchmark methods. We expect such forest-specific DL model transfer can be suitable also for other forest variables and other EO data sources that are sensitive to forest structure. [ABSTRACT FROM AUTHOR]

Copernicus Sentinel-1 images are widely used for forest mapping and predicting forest growing stock volume (GSV) due to their accessibility. However, certain important aspects related to the use of Sentinel-1 time series have not been thoroughly explored in the literature. These include the impact of image time series length on prediction accuracy, the optimal feature selection approaches, and the best prediction methods. In this study, we conduct an in-depth exploration of the potential of long time series of Sentinel-1 SAR data to predict forest GSV and evaluate the temporal dynamics of the predictions using extensive reference data. Our boreal coniferous forests study site is located near the Hyytiälä forest station in central Finland and covers an area of 2500 km2 with nearly 17,000 stands. We considered several prediction approaches and fine-tuned them to predict GSV in various evaluation scenarios. Our analyses used 96 Sentinel-1 images acquired over three years. Different approaches for aggregating SAR images and choosing feature (predictor) variables were evaluated. Our results demonstrate a considerable decrease in the root mean squared errors (RMSEs) of GSV predictions as the number of images increases. While prediction accuracy using individual Sentinel-1 images varied from 85 to 91 m3/ha RMSE, prediction accuracy with combined images decreased to 75.6 m3/ha. Feature extraction and dimension reduction techniques facilitated the achievement of near-optimal prediction accuracy using only 8–10 images. Examined methods included radiometric contrast, mutual information, improved k-Nearest Neighbors, random forests selection, Lasso, and Wrapper approaches. Lasso was the most optimal, with RMSE reaching 77.1 m3/ha. Finally, we found that using assemblages of eight consecutive images resulted in the greatest accuracy in predicting GSV when initial acquisitions started between September and January. [ABSTRACT FROM AUTHOR]

Time series of SAR imagery combined with reference ground data can be suitable for producing forest inventories. Copernicus Sentinel-1 imagery is particularly interesting for forest mapping because of its free availability to data users; however, temporal dependencies within SAR time series that can potentially improve mapping accuracy are rarely explored. In this study, we introduce a novel semi-supervised Long Short-Term Memory (LSTM) model, CrsHelix-LSTM, and demonstrate its utility for predicting forest tree height using time series of Sentinel-1 images. The model brings three important modifications to the conventional LSTM model. Firstly, it uses a Helix-Elapse (HE) projection to capture the relationship between forest temporal patterns and Sentinel-1 time series, when time intervals between datatakes are irregular. A skip-link based LSTM block is introduced and a novel backbone network, Helix-LSTM, is proposed to retrieve temporal features at different receptive scales. Finally, a novel semisupervised strategy, Cross-Pseudo Regression, is employed to achieve better model performance when reference training data are limited. CrsHelix-LSTM model is demonstrated over a representative boreal forest site located in Central Finland. A time series of 96 Sentinel-1 images are used in the study. The developed model is compared with basic LSTM model, attention-based bidirectional LSTM and several other established regression approaches used in forest variable mapping, demonstrating consistent improvement of forest height prediction accuracy. At best, the achieved accuracy of forest height mapping was 28.3% relative root mean squared error (rRMSE) for pixel-level predictions and 18.0% rRMSE on stand level. We expect that the developed model can also be used for modeling relationships between other forest variables and satellite image time series. [ABSTRACT FROM AUTHOR]