Keywords
remote sensing, expert knowledge, Bayesian Networks, agriculture, land management
Start Date
16-9-2020 9:20 AM
End Date
16-9-2020 9:40 AM
Abstract
Increasingly available remote sensing products can provide valuable up-to-date information to land managers and decision-makers. For example, land-use monitoring can support the implementation of agricultural policies, such as direct payments, and provide timely information on risks such as erosion. However, local experts and stakeholders already have a wealth of knowledge about the land they manage, and integrating this knowledge can help improve the accuracy and usability of remote sensing products. We are developing a tool for the agricultural administration of the Swiss Cantons of Bern, Solothurn, and Fribourg, which aims to use remote sensing data to monitor agricultural land use and improve erosion risk detection. Using time series of Sentinel-2 images, a fairly accurate classification of crop types can be achieved. However, such a classification still contains some uncertainty, e.g. due to limited training data and natural variability in the reflectance signals. By including local expert knowledge on land-use decisions, related to topographical factors, traditional management practices and regulations, the uncertainties in land-use classifications can be reduced. Similarly, while it is challenging to detect small erosion events from satellite imagery alone, they can be a valuable source of information when combined with erosion risk models, meteorological data, and expert knowledge. We use spatial Bayesian Networks in order to integrate local expert knowledge and remote sensing products. This approach allows us to account for spatially heterogeneous uncertainties, which can help prioritize monitoring in the field and contribute to improved land management decisions.
Integrating remote sensing and expert knowledge for agricultural monitoring
Increasingly available remote sensing products can provide valuable up-to-date information to land managers and decision-makers. For example, land-use monitoring can support the implementation of agricultural policies, such as direct payments, and provide timely information on risks such as erosion. However, local experts and stakeholders already have a wealth of knowledge about the land they manage, and integrating this knowledge can help improve the accuracy and usability of remote sensing products. We are developing a tool for the agricultural administration of the Swiss Cantons of Bern, Solothurn, and Fribourg, which aims to use remote sensing data to monitor agricultural land use and improve erosion risk detection. Using time series of Sentinel-2 images, a fairly accurate classification of crop types can be achieved. However, such a classification still contains some uncertainty, e.g. due to limited training data and natural variability in the reflectance signals. By including local expert knowledge on land-use decisions, related to topographical factors, traditional management practices and regulations, the uncertainties in land-use classifications can be reduced. Similarly, while it is challenging to detect small erosion events from satellite imagery alone, they can be a valuable source of information when combined with erosion risk models, meteorological data, and expert knowledge. We use spatial Bayesian Networks in order to integrate local expert knowledge and remote sensing products. This approach allows us to account for spatially heterogeneous uncertainties, which can help prioritize monitoring in the field and contribute to improved land management decisions.
Stream and Session
false