An Evaluation of Radarsat-1 and ASTER Data for Mapping Veredas (Palm Swamps)

Philippe Maillard, Thiago Alencar-Silva, David A. ClausiSensors 2008 · DOI

TL;DR

This paper evaluates the effectiveness of Radarsat-1 (SAR) and ASTER (optical) satellite data for mapping "veredas" (Brazilian palm swamps), crucial wetlands in the Cerrado biome. They found that Radarsat-1, when processed with a Markov Random Field segmentation within a hydrographic buffer, can effectively delineate veredas, especially during the wet season with a low incidence angle. However, for detailed vegetation type classification within veredas, ASTER's VNIR bands outperformed SAR data.

What Problem Does This Paper Address?

Veredas are vital wetland ecosystems in the Brazilian savanna (Cerrado), serving as critical water sources during the dry season. Despite their ecological importance, their extent and condition are largely unknown, making their protection challenging. Traditional mapping methods are time-consuming, and existing satellite data often struggle to accurately delineate these narrow, dynamic features. This research aims to assess the suitability of fine-resolution Radarsat-1 and ASTER satellite data for both mapping the extent of veredas and classifying their internal vegetation types.

Methods

The study employed two main approaches:

  • Unsupervised Vereda Delineation (Radarsat-1): They utilized the MAMSEG (Modified Adaptive Markov Random Fields Segmentation) algorithm, a region-based segmentation method that leverages spatial context. To improve performance and reduce confusion, processing was restricted to a buffer zone around the hydrographic network, as veredas are known to follow these networks. This approach aimed to automatically delineate veredas without requiring ground truth data for the initial segmentation.
  • Supervised Vegetation Type Classification (ASTER + Radarsat-1): Traditional supervised classification methods were used to characterize the vegetation types within the veredas. This involved evaluating different combinations of ASTER bands (VNIR and SWIR) and Radarsat-1 scenes. A "knock-out" approach was used for feature selection to identify the most informative bands and scenes.

Datasets

  • Radarsat-1: Four SAR images were acquired from the Canadian Space Agency (CSA) in "standard beam" mode. These were taken during two distinct periods (end of wet season - April, and end of dry season - September) and at two different incidence angles (S2: 24°-31°, S6: 41°-46°).
  • ASTER: One ASTER image from the Terra satellite, acquired on August 21, 2006, was used. Specifically, the VNIR bands (green, red, near infrared) and some SWIR bands (2, 3, and 5) were utilized.
  • Ground Truth Data: Field campaigns involved 24 transects across veredas to collect vegetation type information. This data was used for training and validation of the classification models.

Key Insights and Ideas

  • Hydrographic Buffering is Key for SAR Delineation: Restricting SAR processing to a buffer zone around the hydrographic network significantly improved the accuracy of vereda delineation, reducing class confusion and highlighting the importance of this a priori knowledge.
  • Wet Season and Low Incidence Angle Favor SAR Delineation: Radarsat-1 images acquired during the wet season (April) with a lower incidence angle (S2) consistently yielded better results for vereda delineation, likely due to higher soil moisture and more pronounced backscattering from saturated vegetation.
  • Optical Data Excels at Internal Classification: While SAR is good for broad delineation, ASTER's VNIR bands alone provided superior accuracy for classifying the different vegetation types within the veredas, suggesting that optical data captures finer spectral nuances crucial for this task.
  • SAR Alone Struggles with Fine Vegetation Types: Using only Radarsat-1 data, even with multiple dates and incidence angles, resulted in low classification accuracy for distinguishing vegetation types within veredas, highlighting the limitations of SAR for this specific application.
  • Vereda Width and Location Matter: Narrower veredas, especially near headwaters where soils are drier, were more difficult to detect with SAR. This is attributed to reduced dielectric contrast and volumetric scattering.

Key Results

  • Vereda Delineation (Radarsat-1):
    • The MAMSEG algorithm with hydrographic buffering achieved overall accuracies between 62% and 69% for delineating veredas.
  • Excluding grassland from the "vereda" class improved accuracy, with the April S2 image yielding the best results (69.2% overall).
  • The riparian forest portion of veredas was generally well captured (around 74.2% of riparian forest classified as vereda), even with lower accuracy SAR data.
  • Vegetation Type Classification:
    • Using only ASTER VNIR bands achieved the highest overall success rate of 78.8%.
  • Combining VNIR with April Radarsat-1 scenes did not improve accuracy.
  • Using only the four Radarsat-1 scenes resulted in a low overall success rate of 50.8%.

Limitations and Future Work

  • Resolution Limitations: The 12.5-15m resolution of the satellite data was a limitation for capturing very narrow vereda features (often less than 20m wide), leading to gaps and difficulties in delineation. Higher resolution data (e.g., Ikonos, Quickbird) might be necessary for more precise mapping at a higher cost.
  • SAR Artifacts: The SAR-based delineation sometimes captured savanna areas that were not veredas, particularly in the dry season images, indicating a lack of clear contrast between veredas and other vegetation.
  • Future Work: The authors suggest that while SAR can provide a rough evaluation of vereda extent, fine-resolution optical data like ASTER is better suited for characterizing their types. More precise solutions would likely involve higher-resolution data. The study also provides insights into the relationship between SAR backscattering and vereda types, which could inform future SAR applications.

Summary updated

Abstract

Veredas (palm swamps) are wetland complexes associated with the Brazilian savanna (cerrado) that often represent the only available source of water for the ecosystem during the dry months. Their extent and condition are mainly unknown and their cartography is an essential issue for their protection. This research article evaluates some of the fine resolution satellite data both in the radar (Radarsat-1) and optical domain (ASTER) for the delineation and characterization of veredas. Two separate approaches are evaluated. First, given the known potential of Radarsat-1 images for wetland inventories, the automatic delineation of veredas is tested using only Radarsat-1 data and a Markov random fields region-based segmentation. In this case, to increase performance, processing is limited to a buffer zone around the river network. Then, characterization of their type is attempted using traditional classification methods of ASTER optical data combined with Radarsat-1 data. The automatic classification of Radarsat data yielded results with an overall accuracy between 62 and 69%, that proved reliable enough for delineating wide and very humid veredas. Scenes from the wet season and with a smaller angle of incidence systematically yielded better results. For the classification of the main vegetation types, better results (overall success of 78.8%) were obtained by using only the visible and near infrared (VNIR) bands of the ASTER image. Radarsat data did not bring any improvement to these classification results. In fact, when using solely the Radarsat data from two different angle of incidence and two different dates, the classification results were low (50.8%) but remained powerful for delineating the permanently moist riparian forest portion of the veredas with an accuracy better than 75% in most cases. These results are considered good given the width of some types often less than 50 m wide compared with the resolution of the images (12.5 - 15 m). Comparing the classification results with the Radarsat-generated delineation allows an understanding of the relation between synthetic aperture radar (SAR) backscattering and vegetation types of the veredas.
BibTeX
@inproceedings{maillard2008evaluation,
  author = {Philippe Maillard and Thiago Alencar-Silva and David A. Clausi},
  title = {{An Evaluation of Radarsat-1 and ASTER Data for Mapping Veredas (Palm Swamps)}},
  booktitle = {Sensors},
  year = {2008},
  doi = {10.3390/s8096055},
}
Figure