Accuracy Assessment Between In-Situ Observations and Remote Sensing Products.

It is well known that understanding and describing how land surfaces change over time is becoming more critical. Land cover and land use maps have proved especially useful for estimating the effects of various human activities upon ecosystems. From increases in the urbanization footprint to the extent of fires and forest disturbance, more recently with efforts being directed towards carbon sequestration and carbon stocks as well as modeling the potential impact that a change in behavior can generate, land use maps are part of our workflow.

Remote sensing and various classification methodologies have been the primary sources for studying these changes at different scales, both temporal and spatial, where surface observations are often used as ground truth to validate the final product's quality. Whether or not the true status of the system is being depicted, it is always the challenge. Consequently, the most crucial factor is the quality of the ground truth data; the closer it is to the true state, the better and more trustworthy the classification will be. This is especially true when finer resolutions are required, as in the case of small landholders in developing countries.

In this study, we compared a near real-time land use and land cover classification based on surface observations against field data collected in Senegal in 2018, 2019, and 2021. Although it is important to note the unbalanced nature of the dataset, all the classification metrics were quite low for most of the tested classes, suggesting a lack of agreement among both datasets. Despite this preliminary assessment, we acknowledge the significance of these maps and their contribution to the growth of society and science. However, we must be cautious of the shortcomings when using these layers, mainly when trying to make inferences in small landholders.