Arctic Vegetation Mapping Using Unsupervised Training Datasets and Convolutional Neural Networks

Abstract

Land cover datasets are essential for modeling and analysis of Arctic ecosystem structure and function and for understanding land–atmosphere interactions at high spatial resolutions. However, most Arctic land cover products are generated at a coarse resolution, often limited due to cloud cover, polar darkness, and poor availability of high-resolution imagery. A multi-sensor remote sensing-based deep learning approach was developed for generating high-resolution (5 m) vegetation maps for the western Alaskan Arctic on the Seward Peninsula, Alaska. The fusion of hyperspectral, multispectral, and terrain datasets was performed using unsupervised and supervised classification techniques over a ∼343 km2 area, and a high-resolution (5 m) vegetation classification map was generated. An unsupervised technique was developed to classify high-dimensional remote sensing datasets into cohesive clusters. We employed a quantitative method to add supervision to the unlabeled clusters, producing a fully labeled vegetation map. We then developed convolutional neural networks (CNNs) using the multi-sensor fusion datasets to map vegetation distributions using the original classes and the classes produced by the unsupervised classification method. To validate the resulting CNN maps, vegetation observations were collected at 30 field plots during the summer of 2016, and the resulting vegetation products developed were evaluated against them for accuracy. Our analysis indicates the CNN models based on the labels produced by the unsupervised classification method provided the most accurate mapping of vegetation types, increasing the validation score (i.e., precision) from 0.53 to 0.83 when evaluated against field vegetation observations.

Mapping Vegetation at Species Level with High-Resolution Multispectral and Lidar Data Over a Large Spatial Area: A Case Study with Kudzu

“…For example, Dorigo et al [17] integrated multiple sources of RS data to map an invasive plant by collecting samples for all vegetation classes, and Nguyen et al [18] used RS data and OBIA to map multiple plant species by collecting samples for all primary vegetation classes. Using coarsely or imperfectly labeled samples extracted from open source platforms [20] or land cover maps [21] is a valuable method to facilitate classifications when accurate samples are limited [20][21][22][23]. Langford et al [21] used a coarse land-cover map together with K-means clustering to generate training samples for the arctic vegetation mapping.…”

“…Using coarsely or imperfectly labeled samples extracted from open source platforms [20] or land cover maps [21] is a valuable method to facilitate classifications when accurate samples are limited [20][21][22][23]. Langford et al [21] used a coarse land-cover map together with K-means clustering to generate training samples for the arctic vegetation mapping. Maggiori et al [20] first used imperfect data to train neural networks, and then refined the model with small amounts of accurately labeled samples to detect urban buildings.…”

Mapping Vegetation at Species Level with High-Resolution Multispectral and Lidar Data over Large Spatial Area: A Case Study with Kudzu

“…For examples, Dorigo et al [17] integrated multiple source of RS data to map an invasive plant by collecting samples for all vegetation classes, and Nguyen et al [18] used RS data and OBIA to map multiple plant species by collecting samples for all primary vegetation classes. Using coarsely or imperfectly labeled samples extracted from open source platform [20] or land cover maps [21] is a valuable method to facilitate classification when accurately labeled samples are limited [20][21][22][23].…”

“…Langford et al [21] used a coarse land cover map together with K-means clustering to generate training samples for vegetation mapping, and Maggiori et al [20] first used imperfect data to train neural networks and then refined the model with small amount of accurately labeled data to detect Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 31 December 2019 doi:10.20944/preprints201912.0418.v1 urban buildings. In our proposed workflow, we designed an innovative use of coarsely labeled samples for mapping target vegetation species to decrease sampling effort for non-target vegetation classes.…”

“…Nowadays deep learning models, especially convolutional neural networks (CNNs), have been increasingly applied for image classification with RS data due to their ability of self-extracting features through convolutional layers [20,21,32]. However, CNNs is computationally redundant and require large samples to train the model, which can be a serious challenge for mapping vegetation at species level with high resolution data over large area.…”