Pringle, MJ; Denham, RJ; Devadas, R (2012). Identification of cropping activity in central and southern Queensland, Australia, with the aid of MODIS MOD13Q1 imagery. INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION, 19, 276-285.
Cropping activity has an importance that extends beyond farming communities, to governments, private industries, and to scientific research. We have developed a remote sensing-based method to detect arable cropping activity in central and southern Queensland. Australia, based on time series analysis of the NDVI layer of MODIS-Terra MOD13Q1 (250-m pixel) imagery. Local auto-regression was used to characterise phenological cycles in the NDVI time series. A random forest was then used to model three broad classes of agricultural vegetation (Grazing, Summer Cropping and Winter Cropping), as a function of phenological metrics and the local variance of the NDVI time series. The latter was found to be the most important distinguishing factor between the three classes. Pixel-by-pixel predictions of the random forest were obtained bi-annually for the study area over a 10-year period. Moderate agreement was seen between the predictions of the random forest and (independent) visual interpretation of Landsat imagery (Cohen's index of agreement, K-c, of 0.59). We then demonstrated how the random forest's predictions can be used to define the consistency of cropping activity at the spatial scale of an individual farm property; when compared with (independent) visual interpretation of Landsat imagery the agreement was also moderate (K-c = 0.68). In comparison with other crop-mapping approaches in the literature, our results have been achieved: (i) without restricting the method to annual NDVI time series; (ii) without assuming that the time series is regularly spaced and periodic; (iii) by considering only the 'greening-up' phase of the phenological cycles. Crown Copyright (c) 2012 Published by Elsevier B.V. All rights reserved.