Abstract. Environmental and human benefits of Urban Green Spaces (UGSs) have been known for a long time. However, the definition of a reasonable greening strategy still remains challenging due to the lack of sufficient baseline information as well as a lack of consensus what constitutes a UGS. Therefore, accurate identification of the existing green spaces in cities is crucial for developing UGS inventories for urban planning and resource management activities. In this paper we explore the potential of freely available highest resolution multi-spectral remote sensing imagery to identify large homogeneous as well small heterogeneous UGSs. The approach of using a Random Forest classification on Sentinel-2 imagery is shown to be useful to identify various UGSs with a 97 % accuracy. Freely available data and a relatively straightforward implementation of the proposed approach makes it a valuable tool for decision and policy makers.

The development and application of Intelligent Transportation Systems (ITSs) leads to a growing demand of traffic data. Floating Car Observers (FCOs) contribute by providing information about ambient traffic of vehicles while driving. We present an approach to implement an FCO that uses particulate matter sensors for obtaining road abrasion from cars driving ahead of a test vehicle. Using Random Forest (RF), we predict presence and absence of ambient traffic in the vicinity of test vehicle with particulate matter readings (𝑃𝑀 01 , 𝑃𝑀 2.5 , 𝑃𝑀 10 ) as predictor variables. Results show that RF reaches prediction accuracy ranging from 86 to 99 percent for different train/test split options when analysing individual trajectories as well as 88 to 91 percent accuracy when analysing all trajectories combined. We face limitations mainly when merging single trajectories, due to different initial ambient particulate matter values. We conclude that presence and absence of ambient traffic are predictable using Random Forest with road abrasion values as predictor variables. Further, rainfall events (that may cause wash-off effects on roads) do not significantly change the accuracy of our classification. Optimisation of the model and the need of testing more diverse weather and road conditions remain open tasks for future research. CCS CONCEPTS• Hardware → Sensors and actuators; Sensor applications and deployments; Sensor devices and platforms; • Computing methodologies → Machine learning.

Calcareous alvar grasslands are one of the most species-rich habitats in Estonia. Land-use change and cessation of traditional agricultural practices have led to a decrease of the area of these valuable grasslands during the past century. Therefore, their conservation and restoration are becoming increasingly important. Efforts to restore these habitats have already been made in recent years. Land suitability analysis for potential restoration sites, using the machine learning technique Random Forest (RF), was performed for the first time in this study, which aimed to assess the use of RF for a suitability analysis of alvar grassland. RF predicted 610.91 km 2 of areas suitable for restoring alvar grasslands or for creating alvarlike habitats in Estonia. These areas include all existing alvar areas as well an additional 140.91 km 2 suitable for establishing new habitat similar to calcareous alvar grasslands. We discuss suitability analysis to help with restoration planning and find it to be a reasonable and efficient tool that has potential to provide relevant information. The quality of the prediction could be improved by including additional data relevant for alvar grasslands, such as soil depth, but such data was unfortunately unavailable.

Abstract. Microplastics (MP), until now mostly studied in aquatic ecosystems, are also largely polluting terrestrial ecosystems, especially soil systems. Overall, there is a lack of robust and fast methods to identify, separate and eliminate MPs from soils. This paper is a first attempt to use 2D shape descriptors and Random Forest Machine Learning method in order to discriminate soil and MP particles. The results of this study demonstrate promising potential of the Machine Learning approach and shape descriptors in this relatively new scientific field of determining MPs in soils.

Abstract. Urban Green Spaces (UGSs) are recognized as crucial parts of the human-nature ecosystem in densely populated urban centers. Even though they have been intensively studied, an ultimate list of all types of UGSs in Europe still does not exist. This challenges decision making on whether an area should be considered an UGS or belong to another land-use class. Furthermore, the means of precise identification of UGSs are dependent, among others, on their type and semantics. Therefore, in this paper, we investigate forests as UGSs and automatically identify them using their distinct characteristics from Sentinel-2 images as well as descriptive properties derived from them, i.e., vegetation indices and texture metrics.We enrich these properties with forest relevant features such as minimum vegetation height and homogeneity. To assess the reliability of the proposed workflow, we test our approach in two German cities and compare the results with existing governmental land use data sets. With the implemented approach we precisely identify over 90% of the existing forests in the study areas. The main restriction of the approach is the transferability of the thresholds of predictor variables such as homogeneity and dissimilarity.