Generation of Lidar-Predicted Forest Biomass Maps from Radar Backscatter with Conditional Generative Adversarial Networks

Abstract

This paper studies construction of above-ground biomass (AGB) prediction maps from synthetic aperture radar (SAR) intensity images. The purpose is to improve traditional regression models based on SAR intensity, trained with a limited amount of AGB in situ measurements. Although it is costly to collect, data from airborne laser scanning (ALS) sensors are highly correlated with AGB. Therefore, we propose using AGB predictions based on ALS data as surrogate response variables for SAR data in a sequential modelling fashion. This increases the amount of training data dramatically. To model the regression function between SAR intensity and ALS-predicted AGB we propose to utilise a conditional generative adversarial network (cGAN), i.e. the Pix2Pix convolutional neural network. This enables the recreation of existing ALS-based AGB prediction maps. The generated synthesised ALS-based AGB predictions are evaluated qualitatively and quantitatively against ALS-based AGB predictions retrieved from a traditional non-sequential regression model trained in the same area. Results show that the proposed architecture manages to capture characteristics of the actual data. This suggests that the use of ALS-guided generative models is a promising avenue for AGB prediction from SAR intensity. Further research on this area has the potential of providing both large-scale and low-cost predictions of AGB.

On the Potential of Sequential and Nonsequential Regression Models for Sentinel-1-Based Biomass Prediction in Tanzanian Miombo Forests

“…2) We change the activation function in the output layer from a hyperbolic tangent (tanh) function used in [35] to a rectified linear unit (ReLU) activation function. In an earlier phase of this work [96], we noticed that the tanh activation function we used in the output layer generated AGB values that overestimated the ALS-based AGB predictions from [22], and particularly failed to predict AGB values close to zero. An essential criterion for our cGAN regression model is that it should be able to predict zero biomass to correlate well with AGB ground reference data, z, in non-vegetated areas.…”

“…An essential criterion for our cGAN regression model is that it should be able to predict zero biomass to correlate well with AGB ground reference data, z, in non-vegetated areas. The overprediction observed in [96] can be explained by the nature of the tanh activation function. As the range of the tanh function is [0.0, 1.0], it implies that all data introduced to the cGAN need to be normalised to the same range.…”

“…IV-D2 against each other to assess their effect on cGAN performance for data generation. For all experiments in this section, we trained the cGAN for 200 epochs using a ResNet-6 network, WGAN-GP objective function, batch size (BS) of 2, layer normalisation (LN) for D and batch normalisation (BN) for G. These settings were determined by the model validation results presented in [96]. Overall, Fig.…”

Synthetic SAR Data Generator Using Pix2pix cGAN Architecture for Automatic Target Recognition

“…For the case of Sentinel's optical and SAR images, they proposed a novel method based on a cGAN. Bjork et al studied the generation of LiDAR-predicted above-ground biomass maps from SAR intensity images using cGAN[41]. In[42], Huang et al proposed a novel fusion dehazing method to directly restore haze-free images by using an end-to-end cGAN.…”