Gradual or abrupt? An algorithm to monitor urban vegetation dynamics in support of greening policies

Abstract

While greening becomes a more and more popular strategy to address multiple urban challenges and to enhance wellbeing and human-nature connectedness, there is an increasing need for usable methods and indicators to monitor its implementation. Earth observations produce a wealth of data on vegetation dynamics, but their use for monitoring urban greening policies is still limited. In this article, we develop and test an algorithm for the analysis of urban vegetation dynamics based on NDVI time series. Specifically, we focus on yearly greenest pixel composites that illustrate the maximum value of NDVI during the year (“greenness”): a key structural attribute to monitor urban ecosystems in the European Union. The algorithm is inspired by earlier examples of segmentation algorithms but fits the specific requirements of the targeted use in urban areas. It takes the series of NDVI values associated to each pixel, detects existing (multiple) break points, and quantifies related abrupt changes, as well as significant gradual changes that occurred during a selected period. We tested the algorithm on a 30-year Landsat series in Berlin and partially validated the output through a comparison with infrared ortophotos. The results reveal a net increase in NDVI between 1988 and 2017 in 84% of the pixels, with an average change over the whole city of + 0.096. Around 20% of the pixels show at least one abrupt change. Most abrupt changes (71.5%) were positive, but the negative ones had on average a greater absolute value (−0.170 vs +0.085). However, considering the cumulative impacts during the whole period, 97% of the total change is attributable to gradual changes. The validation proves that abrupt changes successfully capture variations in the extent of vegetation due to land cover changes (e.g., vegetation removal or new greening interventions), while gradual changes can be associated to vegetation growth or decline. We discuss the strengths and limitations of the proposed algorithm, and how the spatially- and temporally-explicit results can be a step forward in the interpretation of urban vegetation dynamics towards an effective monitoring of the impacts of local greening policies.