Banca de DEFESA: GABRIELA DA ROCHA SALDANHA

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : GABRIELA DA ROCHA SALDANHA
DATE: 31/08/2026
TIME: 13:30
LOCAL: videoconferência
TITLE:

Digital Mapping of Soil Classes and Attributes of the Upper Taquari Watershed Based on Legacy Data and Prediction

KEY WORDS:

Machine learning; Fragile soils; Soil erosion

PAGES: 165
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUMMARY:

The Upper Taquari Watershed (UTW) has been suffering from intense erosion processes since the 1970s, as a result of the replacement of native vegetation by agricultural and livestock activities, carried out in a disorderly manner, disregarding soil limitations, without adequate management and conservation practices. In this context, the study was divided into three chapters. The first aimed to spatially predict sand, clay, and organic carbon contents using different machine learning (ML) models, combining soil data and environmental covariates. The second aimed to evaluate different ML models using the delineation of irregular polygons around control sample points (ROICs) method as a sample balancing strategy in the spatial prediction of Soil Mapping Units (SMUs). The third aimed to estimate the mean annual soil loss using the Revised Universal Soil Loss Equation (RUSLE), identifying the spatial distribution of soil loss potential in the watershed. For the modeling procedures, a database containing 329 soil profiles was used in the first chapter, and 326 soil profiles for the second and third chapters, respectively. As covariates, terrain data (SAGA GIS software), optical spectral data (Landsat OLI-8), and radar spectral data (Sentinel-1) were used for the first chapter. In the second chapter, in addition to those mentioned, predicted maps of clay and organic carbon at different depths (0–20; 20–60; 60–80; 80–100; 100–150 cm) were added. The covariates were subjected to Spearman's correlation |r| > 0.95 and subsequently to the Recursive Feature Elimination (RFE) algorithm to select the best covariates. The models were run using functions from the caret package. As results of the first chapter, a predominance of high sand content in the area was found, evidencing the fragility of the soils in the area, where sandy texture predominates. Among the models, Random Forest was the best predictor for sand and carbon, while Cubist was the best for clay. Regarding covariates, terrain attributes were the best predictors of the mentioned properties. According to the predicted data, the municipalities of Figueirão, Camapuã, Alcinópolis, and Alto Araguaia presented sand contents of 849.63, 839.06, 823.07 g kg⁻¹, and 806.93 g kg⁻¹, respectively, while Alto Taquari and Costa Rica had higher clay contents, equal to 438.74 and 205.94 g kg⁻¹, respectively. The municipality of Alto Taquari presented the highest carbon content, followed by Coxim, São Gabriel do Oeste, and Costa Rica, with values of 1.32, 0.99, 0.98, and 0.97%, respectively. In the second chapter, it was found that the predominant SMU in the basin was AR-dy (Dystric Arenosols/Neossolos Quartzarênicos). In the municipalities of Alcinópolis, Alto Araguaia, Pedro Gomes, and Figueirão, a predominance of the AR-dy (Dystric Arenosols/Neossolos Quartzarênicos) SMU was observed, while in the municipalities of Alto Taquari, Costa Rica, and São Gabriel do Oeste, FR-ro (Rhodic Ferralsols/Latossolos Vermelhos) were predominant. In Coxim and Camapuã, the predominance was of the AC-ha (Haplic Acrisols/Argissolos Vermelho-Amarelos) SMU, while in Rio Verde de Mato Grosso, it was CA-dy (Dystric Cambisols/Cambissolos Háplicos). Even with low occurrences, the AR-gl (Gleyic Arenosols/Neossolos Quartzarênicos Hidromórficos) and FR-rh-ar (Rhodic Ferralsols (Arenic)/Latossolos Vermelhos psamíticos) SMUs were found with greater extent in Coxim and Pedro Gomes, respectively. In the third chapter, for the application of RUSLE, the following factors were formulated: rainfall erosivity (R), soil erodibility (K), topographic (LS), cover management (C), and support practices (P). Historical data from the Agência Nacional de Águas (ANA), an equation considering data from the 326 soil samples, a raster representing the relationship between slope length and gradient (SAGA GIS software), a calculation based on the NDVI image, and a unit value adopted to represent support practices were used, respectively. As a result, it was observed that there was a predominance of a high erosion degree (100–200 t ha⁻¹ year⁻¹). The municipalities with a low erosion degree were Alto Taquari, Coxim, and São Gabriel do Oeste, while the municipalities with the highest estimated mean annual loss were Alcinópolis, Alto Araguaia, and Costa Rica. These results can be related to the texture, soil classes, and climatic and terrain characteristics of the area. It is concluded that the soils of the UTW, due to their texture, soil classes, and susceptibility to erosion, require conservation management in order to achieve environmental balance, aiming to reduce the erosion processes reported over time, as well as to optimize agricultural and livestock activities.

COMMITTEE MEMBERS:
Externo à Instituição - NÍCOLAS AUGUSTO ROSIN - EMBRAPA
Presidente - 2223668 - HELENA SARAIVA KOENOW PINHEIRO
Interno - 1060711 - MARCOS GERVASIO PEREIRA
Interno - 2136627 - NIVALDO SCHULTZ
Externo à Instituição - Silvio Barge Bhering - EMBRAPA