Synergy between digital soil mapping and crop modeling: influence of soil data on sugarcane attainable yield

• Autor(a) principal: Santos, Natasha Valadares dos
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/11/11140/tde-21052021-084653/
• topic: DSSAT; DSSAT; Grid yield forecast; Previsão de produtividade em grade; Qualidade de dados de solo; Soil data quality; Spatial soil variability; Variabilidade espacial de solos

Models of crop production play a key role in food security, predicting future agriculture challenges and supporting the establishment of public policies and sustainable management practices. However, due to the lack of reliable information, especially in developing countries, they have presented limited performance and restrictions for spatially explicit analyses. Thus, the objective of this study was to evaluate the DSM (Digital Soil Mapping) as an alternative to fill the gap of soil data. Our study site is in Southwest of Brazil in a 4,815 km2 area heterogeneous in geology and soil classes. The study were conducted with the following framework: (i) We used a soil survey data, containing 1,125 collected points with auger and 27 profiles and applied equal-spline equations to standardized the soil dataset into depth; (ii) A machine learning (ML) algorithm were used to predict soil attributes and their uncertainties (iii) Pedotransfer functions were performed to obtain soil hydrological properties (iv) DSSAT-Canegro was simulated in a 250m grid to sugarcane planted in October with harvest completing 12 months (v) We compared three levels of soil data source: a soil map (SM) (1:100,000 scale), SoilGrids (SG) and the map of attributes (MA) derived from our ML. Clay was the attribute that obtained the best performance to surface and subsurface (R2=0.70 and 0.59, RMSE= 88.87 and 141 g kg-1) and low uncertainty (40 and 110%). In depth the attributes were reduced in their content and increased uncertainty. Therefore, the MA to be the most reliable source of data, being the one that most resembles field data, presents the best index of agreement (d= 0.8) and confidence coefficient (c=0.74). In addition, a 250m grid allowed the evaluation of the spatial variability of the attainable yield of sugarcane at a regional level. Nitisols achieved higher productivity and shallow soils did not exceed 100 t ha-1 Thus, this work showed the applicability of digital mapping for application in crop modeling. This methodology can be replicated for decision-making at a regional level and also to improve management strategies for agriculture.