Use of crop sequences for data-mining of remotely sensed time series across multiple scales: opportunities for scaling up research on agricultural dynamics

Farming activities are rapidly evolving thanks to technological improvements, even though global statistics indicate stagnation or a collapse in total yields for major crops. Further improvements require therefore agronomists to enhance the ways they address the farming use of land and water. Considering the limitedness of these two resources farming would definitely benefit from a smarter spatial design of cropping and mixed crop-livestock systems. Moreover, researchers both from agronomy and from interdisciplinary approaches, such as land change science and eco-agriculture, call for stronger integration of farming features in the research on land management systems. Accordingly, it is crucial to scale up the research on farming systems from plot/farm level to the landscape level so as to build farming system design upon an improved understanding of the land patterns determined by the interactions between farming practices and natural resources. This implies addressing in a spatially explicit way how farmers are choosing what to cultivate and the way to manage it, hence dealing with farming systems from a landscape agronomy perspective. Such an approach, however, largely depends on the availability of data over large areas and for long periods and on the methods to tackle them. Our aim hereby is to present a data-mining method to handle land cover sequences. In particular, we will discuss how segmenting a landscape by using the observed land cover sequences can help identify flexible land units and their potential for cross-scale farming system studies.