Multivariate integration of geospatial data in the assessment of fertility and soil quality in the cerrado biome

Abstract

Soil micronutrients deficiency, including Zinc (Zn), can restrict plant yield and reduce the nutritional quality of many agricultural products, directly impacting human health. Soil pH, available phosphorus (P), and clay content are considered antagonistic to Zn solubility. Hence, they should be taken into account for proper recommendations of sources and correct rates of Zn to be applied in agricultural areas to achieve sustainable agricultural production. Machine learning algorithms, coupled with powerful geospatial analysis platforms such as Google Earth Engine (GEE), can be used to develop strategies and refine agricultural operations. Thus, the objectives of this study were: a) to create spatial predictive models of sand, silt, and clay contents (chapter 1); b) to create spatial predictive models of Zn, pH, and P; and c) to integrate spatial information through rules extracted from geochemical and tacit knowledge of tropical soils under the Cerrado to map the risk of Zn deficiency (chapter 2). For this purpose, a novel legacy soil dataset comprising 32,272 samples of Zn, 42,163 of P, 42,417of pH, and 32,242 samples of clay, sand, and silt contents sampled from the soil surface (0-20 cm) of agricultural Cerrado soils were used. Available environmental covariates were pre-processed, and new environmental covariates were designed to account for contemporary anthropogenic effects on the soil properties. All products were rendered at 30m and 250m resolution. Finally, agricultural interpretation and cross-referencing of information were conducted to evaluate phytoavailability and the high, medium, and low risk of Zn deficiency in Cerrado soils. Spatial predictive models of clay, sand, silt and Zn, respectively, retrieved the most accurate performances with higher R2 (0.83, 0.87, 0.74, and 0.61), lower RMSE (Root mean squared error) (89.61, 104.41, 52.14, and 1.65) and MAE (Mean absolute error) (61.27, 68.77, 36.52, 1.18), and RPIQ values greater than 2.00 indicating excellent model performance. Worse but still consistent performances were obtained for P and pH, reaching R2 of 0.41 and 0.43, RMSE of 5.75 and 0.37, and MAE of 3.97 and 0.97, respectively. RPIQ ranged between 1.40 and 2.0 intervals, meaning reasonable models. Our study indicates that only 4% of the Cerrado biome (~7.7 million ha) is at “high” risk of Zn deficiency. At “medium” risk, considering Zn-Clay interactions encompasses 6% (~12 million ha). Zn-Pdryland achieved 43% (~88.5 million ha), and Zn-pH interactions resulted in 77% (~158 million ha). A total of 91% (~187 million ha) is at “low” risk of Zn deficiency considering Zn-Clay interactions; 89% (~182 million ha) considering Zn-Pirrigated; 55% (~118.8 million ha) considering Zn-Pdryland; and 19% (~38.8 million ha) considering Zn-pH. This work has highlighted the importance of addressing gaps in knowledge and proposing innovative solutions to promote sustainable agricultural practices and address the challenges of global malnutrition.