Continuous change detection and classiﬁcation of land cover using all available Landsat data
A new algorithm for Continuous Change Detection and Classiﬁcation (CCDC) of land cover using all available Landsat data is developed. It is capable of detecting many kinds of land cover change continuously as new images are collected and providing land cover maps for any given time. A two-step cloud, cloud shadow, and snow masking algorithm is used for eliminating “noisy” observations.A time series model that has components of seasonality, trend, and break estimates surface reﬂectance and brightness temperature. The time series model is updated dynamically with newly acquired observations. Due to the differences in spectral response for various kinds of land cover change, the CCDC algorithmuses a threshold derived fromall seven Landsat bands. When the difference between observed and predicted images exceeds a threshold three consecutive times, a pixel is iden-tiﬁed as land surface change. Land cover classiﬁcation is done after change detection. Coefﬁcients from the time series models and the Root Mean Square Error (RMSE) from model estimation are used as input to the Random Forest Classiﬁer (RFC).We applied the CCDC algorithm to one Landsat scene in New England (WRS Path 12 and Row 31). All available (a total of 519) Landsat images acquired between 1982 and 2011 were used. A random stratiﬁed sample design was used for assessing the change detection accuracy, with 250 pixels selected within areas of persistent land cover and 250 pixels selected within areas of change identiﬁed by the CCDC algorithm.The accuracy assessment shows that CCDC results were accurate for detecting land surface change, with producer#39;s accuracy of 98% and user#39;s accuracies of 86% in the spatial domain and temporal accuracy of 80%. Land cover reference data were used as the basis for assessing the accuracy of the land cover classiﬁcation. Theland cover map with 16 categories resulting from the CCDC algorithm had an overall accuracy of 90%.
Mapping and monitoring land cover have been widely recognized as an important scientiﬁc goal(Anderson, 1976; Foody, 2002; Friedl et al.,2002; Hansen, Defries, Townshend, amp; Sohlberg, 2000; Homer, Huang,Yang, Wylie, amp; Coan, 2004; Loveland et al., 2000; Wulder et al., 2008).Land cover inﬂuences the energy balance, carbon budget, and hydrolog-ical cycle as many different physical characteristics change as a function of land cover, such as albedo, emissivity, roughness, photosynthetic capacity, and transpiration. Land cover change can be natural or anthro-pogenic, but with human activity inch surface has been modiﬁed signiﬁcantly in recent years by various kinds of land cover change. Knowledge of land cover and land cover change is necessary for modeling the climate and biogeochemistry of the Earth system and for many kinds of management purposes. Satellite images have long been used to assess the Earth surface because of repeated synoptic collection of consistent measurements (Lambin amp; Strahler, 1994).
1.1.Monitoring land cover change with remote sensing
Images from the Landsat series of satellites are one of the most im-portant sources of data for studying different kinds of land cover change,such as deforestation, agriculture expansion and intensiﬁcation, urban growth, and wetland loss (Coppin amp; Bauer, 1996; Galford et al., 2008;Jensen, Rutchey, Koch, amp; Narumalani, 1995; Seto et al., 2002; Woodcock,Macomber, Pax-Lenney, amp; Cohen, 2001), due to their long record of con-tinuous measurement, spatial resolution, and near nadir observations(Pﬂugmacher, Cohen, amp; Kennedy, 2012; Woodcock amp; Strahler, 1987;
Wulder et al., 2008). One of the drawbacks of Landsat data is the rela-tively low temporal frequency. For each Landsat sensor, overpasses ofthe same location occur every 16 days, and data at this temporal frequency are only commonwithin the United Stateswhere the sensors are turned on for every overpass. For other parts of the world, the fre-quency of data collection is generally less, depending on many factors such as cloud cover predictions and For other parts of the world, the fre-quency of data collection is generally less, depending on many factors such as cloud cover predictions and availability of international ground stations of international ground stations (Arvidson, Goward, Gasch, amp;Williams, 2006). Even for images that are collected, clouds reduce the amount of usable data (Zhang,Rossow, Lacis, Oinas, amp; Mishchenko, 2004). Therefore, most change detection algorithms using Landsat have used two dates of Landsat im-ages (Collins amp;Woodcock, 1996; Coppin, Jonckheere, Nackaerts, Muys, amp; Lambin, 2004; Healey, Cohen, Yang, amp; Krankina, 2005; Masek et al.,2008; Singh, 1989). Though these kinds of algorithms are relatively sim-ple to implement, they are not always applicable. It may take a few years to ﬁnd an ideal pair of Landsat images that are free of clouds,cloud shadows, and snow(hereafter referred to as “clear”) and acquired at the same time of year.