Forest Cover Monitoring
Also access:
➤ Paper with methodology and mapping results
➤ Download the shapefiles: Request through the email florestasc@furb.br or make your registration in our maps platform
Methodology of Forest Coverage and Land Use Mapping
Database
The MonitoraSC employs time series of images and data captured (in all seasons) by orbital optical sensors and RADAR. Most images come from NASA and ESA: MODIS, Landsat series, Sentinel-1, and Sentinel-2, including greater resolution images, such as RapidEye. Topographical data (e.g., altitude, slope, and radiation exposure) extracted from digital elevation models, as well as climate data (e.g., precipitation and temperature) are used to improve the accuracy of land use classification.
Reference data about general land cover, also called "ground truth", are very important in studies involving RS. These data are necessary to train pixel classifiers and to assess the accuracy of the generated products. The IFFSC's field data are a valuable database about forest resources and their spatial configuration; it also may stand as reliable "ground truth" information.
Procedures
The methodological procedures adopted by MonitoraSC are supported by three main product categories: SR products, auxiliary products, and field information.
The set of procedures, illustrated in the flowchart below, was carried out independently in the 42 subdivisions (scenes) of the Santa Catarina territory, aiming to contemplate local and regional idiosyncrasies.
Satellite images were classified using the Random Forest algorithm, resulting in a map composed of 12 thematic classes, with a minimum mapped area of 0.5 hectares and a global accuracy of 95%.
FLOWCHART – MonitoraSC
Methodology of Restinga Mapping
MonitoraSC performed two restinga mappings:
1) Potential restinga areas – areas potentially (or originally) covered by different restinga formations;
2) Remaining restinga areas – current restinga remnants located in potential areas.
Potential restinga areas
Potential areas were established by crossing five attributes: altimetry (NASA – SRTM v.3), accumulation model (IBGE, 2004, 2009), soil classification (EMBRAPA, 2004), relief (CPRM, 2016), and hydrogeological subdomain (CPRM, 2007). The areas were classified into three levels of probability of restinga occurrence: very high, high, and transition. The classification was based on the recognition of the most likely characteristics of each attribute in restinga areas and the information contained in each polygon from crossing the five attributes.
Remaining restinga areas
Restinga remnants were mapped using the same procedures applied to other land use classes, as shown in the flowchart above. The Random Forest algorithm was trained and subsequently validated with real terrain samples from the various restinga formations identified in spatial images with great resolution (Santa Catarina, 2012; Google Earth platform), as well as with in loco photographs taken by the IFFSC. and Google Street View tool. In this mapping, a single thematic class called “Restinga” was mapped.
