Classification of floodplain wetland types based on quad-pol ALOS/PALSAR-2 and SAOCOM polarimetric attributes

MORANDEIRA, NATALIA SOLEDAD; FRANCO, MARIANO; ROJAS BARRIOS, ADRIANA; GRINGS, FRANCISCO MATÍAS

Tokio

Encuentro; FY2023 The Joint PI Meeting of JAXA Earth Observation Missions; 2023

Japan Aerospace Exploration Agency (JAXA)

The work aims to exploit L-Band Full Polarimetric data capabilities to monitor wetland landscapes and to discriminate and characterize wetland types. Our motivation is to promote the inclusion of SAR and PolSAR information in the recently created National Wetland Inventory (Argentina) program. The case study is the Lower Paraná River floodplain, a freshwater wetland landscape with forests, marshes and shallow lakes. L-Band Quad-pol Fine-Beam/StripMap imagery was used. A total of 11 scenes from ALOS/PALSAR-1 (year 2010, mid-wet period), ALOS/PALSAR-2 (2015, wet period), SAOCOM-1A and -1B (2021-2022, dry period) were available. All the acquisitions had an Ascending orbit direction and similar frequency (1.2 GHz to 1.275 GHz) but differed in their incidence angle (steeper for ALOS/PALSAR acquisitions than for SAOCOM). The scenes were processed, coregistered, warped, and tiled. Polarimetric attributes were extracted, including the backscattering of the single polarizations, polarimetric indexes, Cloude-Pottier components, Freeman-Durden components, co-polarization descriptors, pedestal height, conformity factor, and geometric intensity. Random Forest classifications were conducted to discriminate wetland types. The model with the best performance included both ALOS/PALSAR and SAOCOM scenes from different hydrological conditions. VV backscattering and the geometric intensity were selected as important features for several observations. L-Band polarimetric information extracted from ALOS/PALSAR-2 and SAOCOM imagery can help to discriminate wetland cover types. A multitemporal approach is adequate for dynamic wetland landscapes: classifications based on polarimetric features increase their accuracy when scenes from multiple hydrological and climatic conditions are included. [Página web del Encuentro y programa: https://pco-prime.com/pi_joint_ws2023/ ; expositora Morandeira Natalia; proyecto ER3A2N503 PI Morandeira]