Predicting Landscapes from Environmental Conditions Using Generative Networks

Abstract

Landscapes are meaningful ecological units that strongly depend on the environmental conditions. Such dependencies between landscapes and the environment have been noted since the beginning of Earth sciences and cast into conceptual models describing the interdependencies of climate, geology, vegetation and geomorphology. Here, we ask whether landscapes, as seen from space, can be statistically predicted from pertinent environmental conditions. To this end we adapted a deep learning generative model in order to establish the relationship between the environmental conditions and the view of landscapes from the Sentinel-2 satellite. We trained a conditional generative adversarial network to generate multispectral imagery given a set of climatic, terrain and anthropogenic predictors. The generated imagery of the landscapes share many characteristics with the real one. Results based on landscape patch metrics, indicative of landscape composition and structure, show that the proposed generative model creates landscapes that are more similar to the targets than the baseline models while overall reflectance and vegetation cover are predicted better. We demonstrate that for many purposes the generated landscapes behave as real with immediate application for global change studies. We envision the application of machine learning as a tool to forecast the effects of climate change on the spatial features of landscapes, while we assess its limitations and breaking points.

AtmoDist: Self-supervised Representation Learning for Atmospheric Dynamics

“…super-resolution, which does not require any labels and has been considered in a variety of previous works, e.g. (Groenke et al, 2020;Requena-Mesa et al, 2019;Stengel et al, 2020). In particular, we build on the recent work by Stengel et al (2020) that provides a state-of-the-art GAN-based downscaling technique and, to facilitate a direct comparison, employ Before we turn to downscaling, we begin, however, with an intrinsic evaluation of the AtmoDist metric using the average distance between atmospheric states with a fixed temporal separation Δ𝑡.…”

EarthNet2021: A large-scale dataset and challenge for Earth surface forecasting as a guided video prediction task

“…unguided, [17,18,22,25,31,36,40,43,60]). The used models inherit many characteristics known the be useful in modeling Earth surface phenomena: short-long term memory effects [35,55], short-long range spatial relationships [53], as well as, the ability to generate stochastic predictions [2,23,37], ideal to generate ensemble forecasts of Earth surface for effective uncertainty management [72]. Guided video prediction is the setting where on top of past frames models have access to future information.…”

“…This is the setting of EarthNet2021. Some past works resemble the strongly guided setting, however either the future information are derived from the frames themselves, making the approaches not suitable for prediction [67] or they use the dense spatial information but discard the temporal component [42,53].…”

Physically-Consistent Generative Adversarial Networks for Coastal Flood Visualization

“…Generative adversarial networks (GANs) have been used to generate highly photorealistic imagery of faces (Isola et al, 2017;Wang et al, 2018c), animals (Zhu et al, 2017a;Brock et al, 2018), or even street-level flood imagery (Schmidt et al, 2019). Recent works, have adapted GANs to generate satellite imagery (Requena-Mesa et al, 2019;Frühstück et al, 2019;Mohandoss et al, 2020;Singh & Komodakis, 2018;Audebert et al, 2018). Synthetic satellite imagery, however, needs to be trustworthy (Barredo Arrieta et al, 2020).…”