One of the main challenges in land degradation assessment is that a rigorous and systematic approach to addressing its complex dynamics is still missing. The development and application of operative tools at regional and global scales remain a challenge. Land degradation is usually defined as a long-term decline in ecosystem function and productivity. Due to its temporal and spatial resolution as well as data availability, the use of time series of spectral vegetation indexes obtained from satellite sensors has become frequent in recent studies in this field. Slope of linear trends of the normalized difference vegetation index is usually considered an accurate indicator and is widely used as a proxy for land degradation. Yet this method is built on a number of simplifying conceptual and methodological assumptions that prevent capturing more complex dynamics, such as cyclic or periodic behaviors. Our aim was to examine the limitations associated with using linear normalized difference vegetation index trends as proxies for land degradation by comparing outcomes with an alternative methodological procedure based on wavelet autoregressive methods. We explored these issues in 5 case studies from Africa and South America. We observed that trend explained a marginal portion of total temporal variability, whereas monotonic functions, such as linear trends, were unable to capture dynamics that were non-unidirectional, resulting in misinterpretation of actual trends. Wavelet autoregressive method results were encouraging as a step towards the application of more accurate methods to provide sound scientific information of land degradation and restoration.

中文翻译：

---

阶段或制度？重新审视 NDVI 趋势作为土地退化的代理

土地退化评估的主要挑战之一是仍然缺乏一种严格和系统的方法来解决其复杂的动态问题。在区域和全球范围内开发和应用操作工具仍然是一个挑战。土地退化通常被定义为生态系统功能和生产力的长期下降。由于其时间和空间分辨率以及数据的可用性，从卫星传感器获得的光谱植被指数时间序列的使用在该领域的近期研究中变得频繁。归一化差异植被指数的线性趋势斜率通常被认为是一个准确的指标，并被广泛用作土地退化的代理。然而，这种方法建立在许多简化的概念和方法论假设之上，这些假设阻止捕获更复杂的动态，例如循环或周期性行为。我们的目的是通过将结果与基于小波自回归方法的替代方法程序进行比较，来检查与使用线性归一化差异植被指数趋势作为土地退化代理相关的局限性。我们在来自非洲和南美洲的 5 个案例研究中探讨了这些问题。我们观察到趋势解释了总时间可变性的边际部分，而单调函数，如线性趋势，无法捕捉非单向的动态，导致对实际趋势的误解。