Predicting Growing Stock Volume of Scots Pine Stands Using Sentinel-2 Satellite Imagery and Airborne Image-Derived Point Clouds

Paweł Hawryło,Piotr Wężyk

doi:10.3390/f9050274

Abstract

Estimation of forest stand parameters using remotely sensed data has considerable significance for sustainable forest management. Wide and free access to the collection of medium-resolution optical multispectral Sentinel-2 satellite images is very important for the practical application of remote sensing technology in forestry. This study assessed the accuracy of Sentinel-2-based growing stock volume predictive models of single canopy layer Scots pine (Pinus sylvestris L.) stands. We also investigated whether the inclusion of Sentinel-2 data improved the accuracy of models based on airborne image-derived point cloud data (IPC). A multiple linear regression (LM) and random forest (RF) methods were tested for generating predictive models. The measurements from 94 circular field plots (400 m2) were used as reference data. In general, the LM method provided more accurate models than the RF method. Models created using only Sentinel-2A images had low prediction accuracy and were characterized by a high root mean square error (RMSE%) of 35.14% and a low coefficient of determination (R2) of 0.24. Fusion of IPC data with Sentinel-2 reflectance values provided the most accurate model: RMSE% = 16.95% and R2 = 0.82. However, comparable accuracy was obtained using the IPC-based model: RMSE% = 17.26% and R2 = 0.81. The results showed that for single canopy layer Scots pine dominated stands the incorporation of Sentinel-2 satellite images into IPC-based growing stock volume predictive models did not significantly improve the model accuracy. From an operational point of view, the additional utilization of Sentinel-2 data is not justified in this context.

Highlights

Remote sensing has many applications in the field of forestry
This study investigated the usefulness of Sentinel-2 satellite images and their inclusion in image-derived point clouds (IPC)-based predictive models for growing stock volume modeling of single canopy layer Scots pine (Pinus sylvestris L.) stands
We examined the accuracy of two different regression methods—multiple linear regression (LM) and random forest (RF) and analyzed the robustness of variables used in created predictive models

Summary

Introduction

Remote sensing has many applications in the field of forestry. It supports activities related to forest inventory, planning, management and monitoring at a local, regional and up to a global scale.Many different applications are powered by remote sensing data. Remote sensing has many applications in the field of forestry. It supports activities related to forest inventory, planning, management and monitoring at a local, regional and up to a global scale. Many different applications are powered by remote sensing data. The widespread operational use of remote sensing-based methods for forest stand volume estimation has often been limited by the relatively high costs of Airborne Laser Scanning (ALS) data and commercial high-resolution satellite images. The growing popularity of relatively low-cost airborne image-derived point clouds (IPC) and open access to medium-resolution multispectral Sentinel-2 satellite images have created an opportunity to make remote sensing-based inventory methods even more applicable in practice. The two twin optical satellites Sentinel-2A and Sentinel-2B developed by the European Space Agency (ESA) were launched in 2015 and 2017, respectively, as a part of Europe’s Global Monitoring for the Environment

Methods

Results

Discussion

Conclusion