Comprehensive and robust estimations of forest disturbance over time are lacking in many parts of the world. In recent years, Landsat satellite data has been used extensively to retrospectively map forest disturbance across several decades. Human interpretation of Landsat time series, in association with ancillary data, has emerged as an effective and affordable way to collect reference data, which is often used to calibrate and validate disturbance models. If the reference data is collected using a robust probability sampling protocol, estimates of disturbance can be derived directly from the reference dataset using simple calculations. In this paper, we present the results of a reference data project in New South Wales, Australia, undertaken to characterise three decades of canopy disturbance across 13 Mha of predominantly native eucalypt forests. Our results indicate that disturbance is a frequent occurrence, with 70% of the forest disturbed during the last 33 years and around 35% experiencing two or more disturbances. The most common disturbance agent was wildfire, however, most wildfire disturbance occurred in few years, with over 4 Mha burnt in 2019–20 alone. The second most common disturbance agent was drought, which together with fire, was strongly influenced by low rainfall in the preceding years. Forest recovery was often quite rapid following disturbance, particularly in the event of increased rainfall. In this study, we demonstrate the value of collecting information in a systematic and comprehensive framework, which can be used to address numerous monitoring questions regarding forest canopy disturbance.

世界许多地方都缺乏对森林干扰随时间变化的全面而可靠的估计。近年来，Landsat 卫星数据已被广泛用于追溯数十年来的森林干扰图。Landsat 时间序列的人工解释与辅助数据相结合，已成为收集参考数据的有效且负担得起的方式，通常用于校准和验证干扰模型。如果参考数据是使用稳健的概率抽样协议收集的，则可以使用简单的计算直接从参考数据集得出扰动的估计值。在本文中，我们展示了澳大利亚新南威尔士州的一个参考数据项目的结果，致力于表征 13 Mha 主要原生桉树林中三年的冠层干扰。我们的结果表明干扰是经常发生的，在过去 33 年中，70% 的森林受到干扰，大约 35% 经历了两次或更多次干扰。最常见的干扰因素是野火，然而，大多数野火干扰发生在几年内，仅 2019-20 年就有超过 4 Mha 被烧毁。第二个最常见的干扰因素是干旱，它与火灾一起受到前几年低降雨量的强烈影响。受到干扰后，森林恢复通常很快，特别是在降雨量增加的情况下。在这项研究中，我们展示了在系统和全面的框架中收集信息的价值，