Context

Open data policies and accessible computation platforms allow efficient extraction of information from remote sensing data for landscape research. Landscape ecology is strongly influenced by remote sensing, and the value of fine resolution temporal information for characterising landscapes is under-explored.

Objectives

We highlighted the importance of temporal information extracted from remote sensing data gathered over a period of time for landscape research. A case study approach was used to show how time-series information can benefit the mapping of land cover and landscape elements in a heterogeneous landscape dominated by agricultural land use.

Methods

We constructed four composite images of the study area, each incorporating different levels of temporal information. The images either represent a single date or summarise temporal information into single values as the median of spectral bands or vegetation indices. Random forest and k-means clustering methods were used to classify the images.

Results

The overall accuracy of the landscape classifications ranged between 0.3 to 0.8, increasing substantially when including temporal information, for mapping both land cover and small landscape elements. Using temporal information and a RF-based classification it was generally possible to map crop and forest types. The size of landscape elements was overestimated, although the clustering model predicted elements close to their true size and complexity.

Conclusions

The approach highlights the importance of temporal resolution for landscape ecology research. The easy-to-implement methodology offers an opportunity for landscape ecologists to increase the accuracy of landscape mapping and identify ecologically important landscape elements that might otherwise be missed.

中文翻译：

---

覆盖全球的精细时间分辨率卫星传感器：景观生态学家的机会

语境

开放数据政策和可访问的计算平台允许从遥感数据中有效提取信息以进行景观研究。景观生态学受到遥感的强烈影响，用于表征景观的高分辨率时间信息的价值尚未得到充分探索。

目标

我们强调了从一段时间内收集的遥感数据中提取的时间信息对景观研究的重要性。案例研究方法被用来展示时间序列信息如何有利于在以农业土地利用为主的异质景观中绘制土地覆盖和景观元素。

方法

我们构建了研究区域的四个合成图像，每个图像都包含不同级别的时间信息。这些图像要么代表单个日期，要么将时间信息汇总为单个值，作为光谱带或植被指数的中值。使用随机森林和k均值聚类方法对图像进行分类。

结果

景观分类的总体准确度在 0.3 到 0.8 之间，当包括时间信息时，用于绘制土地覆盖和小景观元素的地图精度大大提高。使用时间信息和基于 RF 的分类，通常可以绘制作物和森林类型的地图。尽管聚类模型预测的元素接近其真实大小和复杂性，但景观元素的大小被高估了。

结论

该方法强调了时间分辨率对景观生态学研究的重要性。这种易于实施的方法为景观生态学家提供了一个机会，可以提高景观制图的准确性，并识别可能会被遗漏的具有生态重要性的景观元素。