Vineyard identification in the tulum valley, argentina using RADARSAT imagery

摘要

Two different vineyard planting and management practices, known as parral and vina, are to be found in the San Juan region of Argentina. RADARSAT fine-beam 3, descending images recorded in January, August and September are used to separate vineyards from pasture fields and natural areas, to differentiate between parral and vina, and to separate new, maturing, productive and abandoned parral vineyards. The calibrated and geometrically corrected RADARSAT images were analyzed using the Jeffries-Matusita signature-separability test on both the unfiltered and 3 * 3 gamma-filtered data. In addition, K-nearest-neighbour per-field classification was performed. Results of the Jeffries-Matusita test indicate that perfect signature separability is obtained using the filtered data. However, a classification accuracy of only 68% was obtained using per-field mean values, even though pasture (100%), natural vegetation (100%) and mature parral (93%) were well-identified.