Determining suitable image resolutions for accurate supervised crop classification using remote sensing data

Abstract

Mapping the spatial distribution of crops has become a fundamental input for agricultural production monitoring using remote sensing. However, the multi-temporality that is often necessary to accurately identify crops and to monitor crop growth generally comes at the expense of coarser observation supports, and can lead to increasingly erroneous class allocations caused by mixed pixels. For a given application like crop classification, the spatial resolution requirement (e.g. in terms of a maximum tolerable pixel size) differs considerably over different landscapes. To analyse the spatial resolution requirements for accurate crop identification via image classification, this study builds upon and extends a conceptual framework established in a previous work¹. This framework allows defining quantitatively the spatial resolution requirements for crop monitoring based on simulating how agricultural landscapes, and more specifically the fields covered by a crop of interest, are seen by instruments with increasingly coarser resolving power. The concept of crop specific pixel purity, defined as the degree of homogeneity of the signal encoded in a pixel with respect to the target crop type, is used to analyse how mixed the pixels can be (as they become coarser), without undermining their capacity to describe the desired surface properties. In this case, this framework has been steered towards answering the question: “What is the spatial resolution requirement for crop identification via supervised image classification, in particular minimum and coarsest acceptable pixel sizes, and how do these requirements change over different landscapes?” The framework is applied over four contrasting agro-ecological landscapes in Middle Asia. Inputs to the experiment were eight multi-temporal images from the RapidEye sensor, the simulated pixel sizes range from 6.5 m to 396.5 m. Constraining parameters for crop identification were defined by setting thresholds for classification accuracy and uncertainty. Different types of crops display marked individuality regarding the pixel size requirements, depending on the spatial structures and cropping pattern in the sites. The coarsest acceptable pixel sizes and corresponding purities for the same type of crop were found to vary from site to site, and some crops could not be identified using pixels coarser than 200 m.