Assessment of Landsat Based Deep-Learning Membership Analysis for Development of from–to Change Time Series in the Prairie Region of Canada from 1984 to 2018

Abstract

The prairie region of Canada is a dynamically changing landscape in relation to past and present anthropogenic activities and recent climate change. Improving our understanding of the rate, timing, and distribution of landscape change is needed to determine the impact on wildlife populations and biodiversity, ultimately leading to better-informed management regarding requirements for habitat amount and its connectedness. In this research, we assessed the viability of an approach to detect from–to class changes designed to be scalable to the prairie region with the capacity for local refinement. It employed a deep-learning convolution neural network to model general land covers and examined class memberships to identify land-cover conversions. For this implementation, eight land-cover categories were derived from the Agriculture and Agri-Food Canada Annual Space-Based Crop Inventory. Change was assessed in three study areas that contained different mixes of grassland, pasture, and forest cover. Results showed that the deep-learning method produced the highest accuracy across all classes relative to an implementation of random forest that included some first-order texture measures. Overall accuracy was 4% greater with the deep-learning classifier and class accuracies were more balanced. Evaluation of change accuracy suggested good performance for many conversions such as grassland to crop, forest to crop, water to dryland covers, and most bare/developed-related changes. Changes involving pasture with grassland or cropland were more difficult to detect due to spectral confusion among classes. Similarly, conversion to forests in some cases was poorly detected due to gradual and subtle change characteristics combined with confusion between forest, shrub, and croplands. The proposed framework involved several processing steps that can be explored to enhance the thematic content and accuracy for large regional implementation. Evaluation for understanding connectivity in natural land covers and related declines in species at risk is planned for future research.

Exploration of convolutional neural network architectures for large region map automation

“…Also, the multisize/scale ResNet ensemble architecture was invented by Refs. 39 and 40 to perform a single-pixel prediction based on an input image cropped to a specific size.…”

Exploration of Convolutional Neural Network Architectures for Large Region Map Automation

“…In 2017, Sharma presented a patch-based CNN, 45 which performed effectively on medium-resolution satellite images. Also, Multi-Size/Scale ResNet Ensemble (MSRE) architecture was invented by 46,47 to perform a single-pixel prediction based on an input image cropped to a specific size.…”

Remote Sensing of Land Change: A Multifaceted Perspective

“…A feature space of more than a few dozens to even hundreds of dimensions could be created from the electromagnetic radiation (EMR) that is recorded at different wavelengths, the texture of the spectral bands, and the intra-annual/inter-annual temporal trajectory from the time series observations, which could be further used to determine the land cover based on image classification (Gómez et al, 2016;Pouliot and Latifovic, 2016) or to estimate the biophysical/ biochemical parameters based on machine learning or regression from empirical models (Garbulsky et al, 2011;Lin et al, 2020;Verrelst et al, 2015). Recently, the deep-learning-based approaches, particularly Convolutional Neural Network (CNN), have shown better performance in land cover classification compared to the traditional machinelearning-based methods (Kussul et al, 2017;Liu et al, 2021b;Pouliot et al, 2021), and are capable of incorporating the spatial domain of the remote sensing data by automatically extracting a suitable representation of the remote sensing data through a hierarchy of spatial filters at different sizes, which avoids the feature creation and selection processes that most traditional machine learning methods require in advance for preparation of the classification predictors (Molinier et al, 2021).…”

“…However, if the forest harvest is reducing forest cover from 90% to 5%, then the land cover will be likely changed from forest to barren or grass, and we usually call this land cover conversion, which is corresponding to more substantial land changes that cause land cover transitions from one to another. In the remote sensing community, huge efforts have been given to land cover conversions (Chowdhury et al, 2021;Colditz et al, 2014;Homer et al, 2015Homer et al, , 2020Pouliot et al, 2021, but fewer studies were targeted on the land cover modification, which often occur at a spatial scale similar to or even larger than land cover conversion (Asner et al, 2005;Qin et al, 2021;Rigge et al, 2019). Detecting land cover modification is inherently difficult in remote sensing, as the subtle spectral change signal may be at a change magnitude similar to other background noise.…”