Abstract:
As a nonrenewable resource, soils ought to be preserved to ensure a sustainable future. However, the pressures of a growing population have resulted in a higher demand for soil-related products leading to land degradation such as erosion, nutrient depletion, chemical contamination, acidification, and salinity.
Accurate and detailed spatial soil information are essential for sustainable land use and management as well as environmental modelling and risk assessment. Such information is increasingly required by governments and development partners to aid in improving land management. In that context, high resolution spatial information on soils can assist decision makers to better target areas for soil fertility interventions and implement knowledge-based policies that aim at increasing agricultural production and improving livelihoods especially for small-scale farmers. In this seminar, I will explore the application of digital soil mapping to map primary soil properties at high resolution as well as how these information could be integrated to mapping soil functions that can support end-user decisions. Findings revealed that machine learning techniques performed generally better than multiple linear regression, with ensemble models such as Random Forest providing in most cases the highest accuracy. Further, data pruning of a major reference soil groups could lead to an increase in prediction accuracy of minor soil groups by reducing error metrics due to the involvement of many surveyors. Finally, the spatial distribution of different soil fertility classes in a West African country such as Benin showed a preponderance of low fertility class with severe limitations for crop development along with some soils with high and average fertility levels which should be solely preserved for agriculture by policy makers.
Speaker: Dr. Ozias K. L. Hounkpatin
Affiliation: Aalto University, Finland
Time: 4:30 PM, Tuesday, Mar. 7, 2023
Venue: ZOOM 会议平台 会议 ID：312 430 8960 会议密码 PWD：666666
ZOOM
会议 ID：312 430 8960
会议密码 PWD：666666
