Temporally Generalizable Land Cover Classification: A Recurrent Convolutional Neural Network Unveils Major Coastal Change through Time

Patrick Clifton Gray,Justin T Ridge,Diego F Chamorro,David W Johnston,Hannah Rae Kerner,Emily A Ury

doi:10.3390/rs13193953

Patrick Clifton Gray, Justin T Ridge + Show 4 more

Open Access

https://doi.org/10.3390/rs13193953

Copy DOI

Abstract

The ability to accurately classify land cover in periods before appropriate training and validation data exist is a critical step towards understanding subtle long-term impacts of climate change. These trends cannot be properly understood and distinguished from individual disturbance events or decadal cycles using only a decade or less of data. Understanding these long-term changes in low lying coastal areas, home to a huge proportion of the global population, is of particular importance. Relatively simple deep learning models that extract representative spatiotemporal patterns can lead to major improvements in temporal generalizability. To provide insight into major changes in low lying coastal areas, our study (1) developed a recurrent convolutional neural network that incorporates spectral, spatial, and temporal contexts for predicting land cover class, (2) evaluated this model across time and space and compared this model to conventional Random Forest and Support Vector Machine methods as well as other deep learning approaches, and (3) applied this model to classify land cover across 20 years of Landsat 5 data in the low-lying coastal plain of North Carolina, USA. We observed striking changes related to sea level rise that support evidence on a smaller scale of agricultural land and forests transitioning into wetlands and “ghost forests”. This work demonstrates that recurrent convolutional neural networks should be considered when a model is needed that can generalize across time and that they can help uncover important trends necessary for understanding and responding to climate change in vulnerable coastal regions.

Highlights

Low lying coastal areas, home to a huge proportion of the global population, are locations where understanding long term changes in land cover is of particular societal importance
Model testing shows that all models are capable of high accuracy across all classes on the 2011 data they were trained on, but the RCNN (Table 3) had higher accuracy than other models when predicting on the 2000 data (Figure 5)
The RCNN model with only a single time step was considerably worse, which suggests that the RCNN is able to extract meaningful patterns out of the phenological data which may help it ignore ecological or sensor irregularities that only exist in a single time step

Summary

Introduction

Home to a huge proportion of the global population, are locations where understanding long term changes in land cover is of particular societal importance. These regions are affected by urban and agricultural expansion, stochastic natural events such as hurricanes, and chronic stressors such as sea level rise (SLR) [1]. Rates of conversion to ghost forests are not well understood, this is a widespread change of ecological and economic importance [2]. Forests are a crucial component of global carbon, nutrient, and hydrologic cycles [3]; they create vast wildlife habitat across coastal regions and provide important ecosystem services to communities across 4.0/).

Objectives

Methods

Results

Discussion

Conclusion