Ivan Ho Yin Kwong

Secondary succession is one of the major processes in forest habitat restoration across degraded landscapes globally, especially in tropical regions. Hong Kong, situated on the northern fringes of the Asian tropics, has undergone near-complete clearing of its original forests due to human activities in history, and most of its current vegetation was formed by regenerations in recent decades. Understanding the dynamics of vegetation changes over time involves various biotic, abiotic, and anthropogenic factors related to different ecological processes. Remote sensing imagery, with the ability to discern habitat patterns across spatial and temporal scales, provides an effective tool for addressing this requirement. In particular, the Landsat satellite mission has provided continuous earth observation data since 1972 and has been widely used in time-series analyses of habitat transformations.This study leveraged all available Landsat imagery to examine the coverage of six habitat classes in the forest-regenerating landscape of Hong Kong from 1973 to 2022. A multi-temporal classification workflow was developed, which combined cross-calibration of Landsat sensors, random forest classification, decision-level fusion after classification, and temporal smoothing. An overall accuracy of 90.1% was achieved when assessed using various office- and field-collected data, with accuracy exceeding 86% and 88% when individual classes and mapping periods were considered respectively. Based on the multi-temporal habitat maps produced from the classification workflow, survival analysis was used to examine the time required for successional changes, and correlation analysis was used to associate the transition time with various natural and anthropogenic factors.The results indicate that (i) a single classification model could be developed using all images acquired by multiple Landsat sensors across years, including the earliest Landsat 1–5 MSS data, which is crucial in extending the temporal baseline and adding a decade of habitat information. (ii) Incorporating more images in the classification model enhanced overall accuracy, with the highest accuracy achieved when all available images were included. Classification performances for earlier years and transitional classes showed higher vulnerability to the reduced proportion of input images. (iii) The natural landscape in Hong Kong gradually transformed from being grassland-dominated in the 1970s to woodland-dominated in the 2010s. Grasslands took a median time of 21 years to become shrublands and another 29 years to become woodlands, but the first quartiles of 7 and 10 years respectively indicate a high spatial variability. Hill fire was the most important factor positively correlated with the transition time (restricting forest succession), while increasing proximity to seed sources and protected area designation produced the highest negative correlations (accelerating the process).This study demonstrates the value of connecting the Landsat time series with human impacts and management practices to produce spatially explicit ecological insights. The experience of forest regeneration in Hong Kong, formed by both conservation interventions and natural succession, will benefit the increasing interest in forest protection and restoration in the wider tropical region.

Intertidal mudflats are productive habitats and monitoring their morphodynamics is critical to the sustainable management of their ecological values. The mudflat in Deep Bay, situated between Hong Kong and Shenzhen (China) and designated as two adjacent Ramsar Sites, serves as an internationally important foraging ground for waterbirds on the East Asian-Australasian Flyway. Compared to traditional ground surveys, satellite remote sensing provides a low-cost and efficient tool to reconstruct intertidal topography. This study leveraged long-term records from Landsat satellites and near-daily observations from PlanetScope satellites to estimate the topography of intertidal mudflats in Deep Bay in multiple periods in 1991–2020 and 2020–2021 respectively. The waterline method was used to delineate contours from the satellite images by combining image thresholds at different tide heights and sea level information from nearby tide gauge records, which were then interpolated to generate intertidal digital elevation models. The estimated topography showed high accuracies when validated with in situ measurements of the mudflat surface. Comparison among periods indicated that both accretion and erosion occurred at different locations with average rates of change from -1.4 to 2.8 cm per year. Sedimentation was evident near the mangal regions which flattened the slope and led to a gradual seaward extension of mudflats at mean sea level, while slope steepening tended to occur at lower levels along the river channels. Such topographic changes, combined with seaward expansions of the intertidal mangal, have caused temporal fluctuations in mudflat areas exposed at different tide heights. The results can be applied to studies on coastal environmental processes and management plans of the two Ramsar Sites. Dense time-series observations from PlanetScope satellites can facilitate more frequent sampling within the tidal cycle and monitoring of intertidal topographic changes in future.

Early naturalists such as Humboldt observed that changes in topography and anthropogenic disturbances influenced vegetation structure and the composition of animal communities. This holistic view of community assembly continues to shape conservation and restoration strategies in an era of landscape degradation and biodiversity loss. Today, remote sensing affords ecologists the tools for obtaining rapid and precise measures of topography, disturbance history and vegetation structure. Nonetheless, the capacity of such measures to predict the structure of diverse and functionally important insect communities has not been fully explored. We sampled ground‐dwelling ant assemblages with pitfall traps along a successional gradient (15 grasslands, 21 shrublands and 44 forests) in subtropical Asia, and measured the taxonomic (TD) and functional diversity (FD). We used airborne Light Detection and Ranging (LiDAR) and aerial photography—to measure topography, anthropogenic‐fire history and vegetation structure at each site. Using structural equation models, we tested the hypothesis that vegetation structure mediated the effects of topography and anthropogenic‐fire history on ant assemblage TD and FD, with stronger effects on the latter. We found that low elevation and anthropogenic‐fire history promoted ant TD, and by mediating vegetation structure, these factors further controlled ant FD. Specifically, assemblages of ant species occupying more similar niches—as indicated by their lower FD—were found in secondary forests that had more structurally homogeneous vegetation. These sites also had low insolation and high water moisture content, and were not recently burned as revealed by LiDAR‐derived metrics and aerial images. Furthermore, remotely sensed vegetation structures were closely associated with individual ant traits, such as body size and eye length, which reflect species' preferences for habitat structure. Synthesis. Our study uncovers the interactive effects of topography, disturbance history and vegetation structure in determining the TD and FD of ant assemblages in subtropical landscapes. Moreover, it demonstrates that remote sensed data can be leveraged to efficiently elucidate the complex effects of environmental change and disturbances on vegetation structure and consequently insect biodiversity, representing ecological proxies to refine ground investigation plans and support appropriate conservation and restoration measures for degraded landscapes.

Continuous monitoring of coastal water qualities is critical for water resource management and marine ecosystem sustainability. While remote sensing data such as Sentinel-2 satellite imagery routinely provide high-resolution observations for time-series analysis, the cloud-based Google Earth Engine (GEE) platform supports simple image retrieval and large-scale processing. Using coastal waters of Hong Kong as the study area, this study utilized GEE to (i) query and pre-process all Sentinel-2 observations that coincided with in situ measurements; (ii) extract the spectra to develop empirical models for water quality parameters using artificial neural networks; and (iii) visualize the results using spatial distribution maps, time-series charts and an online application. The modeling workflow was applied to 22 water quality parameters and the results suggested the potential to predict the levels of several nutrients and inorganic constituents. In-depth analyses were conducted for chlorophyll-a, suspended solids and turbidity which produced high correlations between the predicted and observed values when validated with an independent dataset. The selected input variables followed spectral characteristics of the optical constituents. The results were considered more robust compared to previous works in the same region due to the automatic extraction of all available images and larger number of observations from different years and months. Besides visualizing long-term spatial and temporal variabilities through distribution maps and time-series charts, potential anomalies in the monitoring period including algal bloom could also be captured using the models developed from historical data. An online application was created to allow novice users to explore and analyze water quality trends with a simple web interface. The integrated use of remotely-sensed images, in situ measurements and cloud computing can offer new opportunities for implementing effective monitoring programs and understanding water quality dynamics. Although the obtained levels of accuracies were below the desired standard, the end-to-end cloud computing workflow demonstrated in this study should be further investigated considering the cost and computational efficiency for timely information delivery.

Identification and mapping of various habitats with sufficient spatial details are essential to support environmental planning and management. Considering the complexity of diverse habitat types in a heterogeneous landscape, a context-dependent mapping framework is expected to be superior to traditional classification techniques. With the aim to produce a territory-wide habitat map in Hong Kong, a three-stage mapping procedure was developed to identify 21 habitats by combining very-high-resolution satellite images, geographic information system (GIS) layers and knowledge-based modification rules. In stage 1, several classification methods were tested to produce initial results with 11 classes from a WorldView-2/3 image mosaic using a combination of spectral, textural, topographic and geometric variables. In stage 2, modification rules were applied to refine the classification results based on contextual properties and ancillary data layers. Evaluation of the classified maps showed that the highest overall accuracy was obtained from pixel-based random forest classification (84.0%) and the implementation of modification rules led to an average 8.8% increase in the accuracy. In stage 3, the classification scheme was expanded to all 21 habitats through the adoption of additional rules. The resulting habitat map achieved >80% accuracy for most of the evaluated classes and >70% accuracy for the mixed habitats when validated using field-collected points. The proposed mapping framework was able to utilize different information sources in a systematic and controllable workflow. While transitional mixed habitats were mapped using class membership probabilities and a soft classification method, the identification of other habitats benefited from the hybrid use of remote-sensing classification and ancillary data. Adaptive implementation of classification procedures, development of appropriate rules and combination with spatial data are recommended when producing an integrated and accurate map.

Accurately identified trees can serve as a basis of estimating forest variables through the individual tree-based approach. Increasing richness of remote sensing data also provides opportunities to explore the potential uses of various types of data sources. This study adopted three widely used remote sensing data, including airborne light detection and ranging (LiDAR), unmanned aerial vehicle (UAV) photography and traditional digital aerial photos (DAP), and aimed to investigate their potentials on estimating tree heights and extracting individual tree information in four forested sites in Hong Kong with different tree compositions. Image-based point clouds were generated through photogrammetry. Local maxima and region growing methods were adopted to identify treetops and delineate tree crowns, respectively, with different fixed and variable window size settings. Tree heights obtained from remote sensing datasets resulted in correlation coefficients (r) = 0.58–0.94 and root-mean-square errors (RMSE) = 1.33–3.78 m compared to field-measured values and similar levels of correspondences among the datasets. Point cloud characteristics were highlighted through point-based and profile-based analysis. The highest F-scores for treetop detections in each site ranged from 0.53 to 0.69 with variations caused by different window sizes and data sources. Matched rates of reference trees were positively correlated (r = 0.19–0.49) with geometric properties including diameter at breast height (DBH), tree height, crown area, and distance to the nearest neighbour. No single remote sensing dataset was clearly superior in all methodologies in this study, but unique properties were demonstrated in terms of both data acquisitions and analysis. Knowledge and testing on both characteristics of study areas and data sources were important when deciding the best window size parameters. Heterogeneity of forest environment could be a major factor hindering the delineation performance with further influences on plot-level difference and tree-level detectability.