Soil texture and organic carbon mapping using surface temperature and reflectance spectra in Southeast Brazil

The generation of soil attributes maps is crucial for proper field management, especially in tropical countries where agriculture plays a major role in economy, such as Brazil. The Land Surface Temperature (LST) derived from thermal infrared wavelength can be related with soil properties, particularly texture. The availability of LST data on a regular basis, with good spatial resolution to cover extensive agricultural regions and also integrated with reflectance data enhances its use in soil attributes mapping. The objectives of this research were to verify the relationships between soil attributes texture and organic carbon (OC) with the remote sensing (RS) products LST and surface reflectance and perform maps of classes of such attributes based on these variables. The study area (473 ha) is located near the municipality of Barra Bonita, southeast of São Paulo State, Brazil. A regular sampling grid with 100 × 100 m2 and with sampling density of one sample per hectare was established on the study site, from where surface samples (0-0.2 m) were collected via auger and had their contents determined by wet chemistry analysis. Images from Landsat 5 were obtained for extraction of LST and reflectance. The algorithm Spectral Angle Mapper (SAM) was applied for soil attributes mapping using 55 toposequences samples and considering (a) three LST images; (b) only reflectance from one Landsat scene and (c) reflectance + LST from the same scene as in (b). Besides this, maps were performed using only one LST image with a histogram-based classification. The weighted kappa (kw) was calculated with validation samples and indicated the classification accuracy. Mapping of texture and OC through SAM algorithm produced better results than the classification of LST, regardless of the variables being considered. For both attributes, the best map was produced using reflectance + LST, with substantial agreement for texture and moderate for OC. The use of LST for mapping soil attributes concomitantly with reflectance spectra improved the mapping accuracy.