Classifying Inundation in a Tropical Wetlands Complex with GNSS-R

Nereida Rodriguez-Alvarez,Kyle C Mcdonald,Katherine Jensen,Erika Podest

doi:10.3390/rs11091053

Abstract

The use of global navigation satellite system reflectometry (GNSS-R) measurements for classification of inundated wetlands is presented. With the launch of NASA’s Cyclone Global Navigation Satellite System (CYGNSS) mission, space-borne GNSS-R measurements have become available over ocean and land. CYGNSS covers latitudes between ±38°, providing measurements over tropical ecosystems and benefiting new studies of wetland inundation dynamics. The GNSS-R signal over inundated wetlands is driven mainly by coherent scattering associated with the presence of surface water, producing strong forward scattering and a distinctive bistatic scattering signature. This paper presents a methodology used to classify inundation in tropical wetlands using observables derived from GNSS-R measurements and ancillary data. The methodology employs a multiple decision tree randomized (MDTR) algorithm for classification and wetland inundation maps derived from the phased-array L-band synthetic aperture radar (PALSAR-2) as reference for training and validation. The development of an innovative GNSS-R wetland classification methodology is aimed to advance mapping of global wetland distribution and dynamics, which is critical for improved estimates of natural methane production. The results obtained in this manuscript demonstrate the ability of GNSS-R signals to detect inundation under dense vegetation over the Pacaya-Samiria Natural Reserve, a tropical wetland complex located in the Peruvian Amazon. Classification results report an accuracy of 69% for regions of inundated vegetation, 87% for open water regions, and 99% for non-inundated areas. Misclassification of inundated vegetation, primarily as non-inundated area, is likely related to the combination of two factors: partial inundation within the GNSS-R scattering area, and signal attenuation from dense overstory vegetation, resulting in a low signal.

Highlights

Wetlands cover a small portion of Earth’s ice-free land surface [1], yet they have major impacts on global biogeochemistry, hydrology, and biological diversity
We first determine the global navigation satellite system reflectometry (GNSS-R) observables and ancillary datasets associated with improving our inundation classification
Training and validation data were obtained from PALSAR-2-based reference inundation maps that were developed and evaluated with ground measurements collected during a field campaign in the region [10]

Summary

Introduction

Wetlands cover a small portion (less than 5%) of Earth’s ice-free land surface [1], yet they have major impacts on global biogeochemistry, hydrology, and biological diversity They are the largest natural source of atmospheric methane, with their extent and inundation variability playing a large role in ecosystem dynamics, contributing roughly 20–40% of global methane emissions [2]. One of the main limitations in large-scale mapping of wetland ecosystems from Earth-orbiting remote sensing observations, especially in the tropics, is the almost constant presence of clouds Optical instruments such as the Moderate Resolution Imaging Spectroradiometer (MODIS) have been used to map wetlands’ extent and dynamics at spatial resolutions of 500 m–1 km, and with temporal repeats of up to 8 days [11,12,13]. The lower the microwave frequency (the longer the wavelength), the greater the penetration through the medium (vegetation) and the better the capability for mapping surface inundation beneath vegetation, which is especially difficult under dense forests

Methods

Results

Discussion

Conclusion