Current estimates of forest degradation are associated with large uncertainties. However, recent advancements in the availability of remote sensing data (e.g., the free data policies of the Landsat and Sentinel Programs) and cloud computing platforms (e.g., Google Earth Engine (GEE)) provide new opportunities for monitoring forest degradation. Several recent studies focus on monitoring forest degradation in the tropics, particularly the Amazon, but there are less studies of temperate forest degradation. Compared to the Amazon, temperate forests have more seasonality, which complicates satellite-based monitoring. Here, we present an approach, Continuous Change Detection and Classification - Spectral Mixture Analysis (CCDC-SMA), that combines time series analysis and spectral mixture analysis running on GEE for monitoring abrupt and gradual forest degradation in temperate regions. We used this approach to monitor forest degradation and deforestation from 1987 to 2019 in the country of Georgia. Reference conditions were observed at sample locations selected under stratified random sampling for area estimation and accuracy assessment. The overall accuracy of our map was 91%. The user's accuracy and producer's accuracy of the forest degradation class were 69% and 83%, respectively. The sampling-based area estimate with 95% confidence intervals of forest degradation was 3541 ± 556 km² (11% of the forest area in 1987), which was significantly larger than the area estimate of deforestation, 158 ± 98 km². Our approach successfully mapped forest degradation and estimated the area of forest degradation in Georgia with small uncertainty, which earlier studies failed to estimate.

目前对森林退化的估计存在很大的不确定性。然而，遥感数据可用性（例如，Landsat 和 Sentinel 计划的免费数据政策）和云计算平台（例如，谷歌地球引擎 (GEE)）的最新进展为监测森林退化提供了新的机会。最近的几项研究侧重于监测热带地区的森林退化，特别是亚马逊地区，但对温带森林退化的研究较少。与亚马逊相比，温带森林具有更多的季节性，这使基于卫星的监测变得复杂。在这里，我们提出了一种方法，连续变化检测和分类 - 光谱混合分析 (CCDC-SMA)，它结合了在 GEE 上运行的时间序列分析和光谱混合分析，用于监测温带地区的突然和逐渐森林退化。我们使用这种方法来监测 1987 年至 2019 年格鲁吉亚国家的森林退化和森林砍伐。在分层随机抽样下选择的样本位置观察参考条件，以进行面积估计和准确性评估。我们地图的整体准确率为 91%。森林退化等级的用户准确率和生产者准确率分别为 69% 和 83%。森林退化的 95% 置信区间的基于抽样的面积估计为 3541 ± 556 公里 在分层随机抽样下选择的样本位置观察参考条件，以进行面积估计和准确性评估。我们地图的整体准确率为 91%。森林退化等级的用户准确率和生产者准确率分别为 69% 和 83%。森林退化的 95% 置信区间的基于抽样的面积估计为 3541 ± 556 公里 在分层随机抽样下选择的样本位置观察参考条件，以进行面积估计和准确性评估。我们地图的整体准确率为 91%。森林退化等级的用户准确率和生产者准确率分别为 69% 和 83%。森林退化的 95% 置信区间的基于抽样的面积估计为 3541 ± 556 公里²（1987 年森林面积的 11%），明显大于森林砍伐面积估计值 158 ± 98 km ²。我们的方法成功地绘制了森林退化图，并以较小的不确定性估计了乔治亚州的森林退化面积，而早期的研究未能估计这一点。