Presentation Information
[O12-P108]Detecting Emergent Floating Vegetation Using WARN: A Novel Remote Sensing Monitoring Network
*Karin Ito1, *Kanon Sugawara1, *Shixiao Jerry Liao1 (1. Makuhari Senior High School)
Keywords:
Alligator weed,Sentinel satellite data,Google Earth Engine (GEE)
1. Background and Objective
The purpose of this study is to evaluate the effectiveness of a monitoring approach using Google Earth Engine (GEE) and Sentinel satellite data for the early detection of invasive aquatic plants such as the Alternanthera philoxeroides (alligator weed). Invasive aquatic plants pose serious threats to ecosystems and contribute to water quality degradation. Due to their rapid proliferation, early detection and removal are essential.
Conventional field surveys and visual inspections are not well-suited for large-scale, continuous monitoring, highlighting the need for more efficient approaches. Sheffield et al., 2022 has shown that invasive aquatic plants can be detected using aerial imagery; however, while such imagery offers high spatial resolution, it is costly for frequent large-scale observation. Satellite-based approaches such as Shinohara & Kurita, 2025 have demonstrated potential but are not yet widely implemented for large-scale monitoring or automated alert systems. Therefore, this study aims to determine the spatial scale at which floating vegetation can be reliably detected using Sentinel-2 data and NDVI analysis.
2. Methods
Sentinel-2 optical imagery was used as the primary data source, with Sentinel-1 SAR data as supplementary information. The analysis compared June–September 2017, before the confirmation of alligator weed (baseline), with June–September 2018, when its presence was reported (target period). NDVI was used for vegetation detection, and MNDWI for water body extraction. A water mask was first applied to exclude terrestrial vegetation. Aquatic vegetation was then extracted using a combination of an NDVI threshold, adjacency to water bodies, and a minimum cluster size of three connected pixels, enabling detection of vegetation patches rather than isolated pixels.
Additionally, data from 2017 and 2018 were compared to identify newly emerged vegetation areas. A time-series analysis of NDVI from 2015 to 2019 was conducted to examine anomalous increases. Detection performance was evaluated using false positives(FP) and false negatives(FN), with the aim of identifying the minimum detectable infestation area while maintaining acceptable accuracy. The results were compared with publicly available data from Chiba Prefecture to validate the satellite-based detection.
3. Results
In 2018, aquatic vegetation increased markedly compared to 2017, with newly emerged patches showing high NDVI values across a wide area. The inclusion of the connectivity condition enabled patch-level detection and significantly reduced noise.
The NDVI time-series analysis also revealed a clear increase during the summer of 2018, consistent with the spatial expansion observed in satellite imagery. These findings suggest that, under certain conditions, satellite data alone can support automatic detection of aquatic plant formation and expansion.
4. Discussion and Future Challenges
The results indicate that restricting analysis to water bodies and combining NDVI with MNDWI effectively reduces false detections. Incorporating spatial conditions such as adjacency to water and minimum cluster size further improves detection stability.
However, several limitations remain, including the influence of water level fluctuations, turbidity, and suspended materials, as well as difficulty distinguishing alligator weed from other aquatic plant species. In addition, Sentinel-2’s 10 m spatial resolution may limit the detection of small, early-stage patches.
Future work should focus on testing the method in other regions, incorporating vegetation expansion models, utilizing higher-resolution data such as drone and commercial satellite imagery, and developing machine learning training datasets. Ultimately, this approach could be extended into an early warning system(WARN) with automated notification functions.
References
Sheffield, Kathryn J., et al. “Detection of Aquatic Alligator Weed (Alternanthera Philoxeroides) from Aerial Imagery Using Random Forest Classification.” Remote Sensing, vol. 14, no. 11, 2 June 2022, p. 2674, https://doi.org/10.3390/rs14112674.
Shinohara, Kengo, and Hideharu Kurita. “Distribution of Invasive Alien Species Alternanthera Philoxeroides in the Hakken River, Chiba Prefecture Visualizing by Satellite Data .” Proceedings of the 74th JSIDRE Annual Conference. Paper No. 3-35(P), pp. 255–256.
The purpose of this study is to evaluate the effectiveness of a monitoring approach using Google Earth Engine (GEE) and Sentinel satellite data for the early detection of invasive aquatic plants such as the Alternanthera philoxeroides (alligator weed). Invasive aquatic plants pose serious threats to ecosystems and contribute to water quality degradation. Due to their rapid proliferation, early detection and removal are essential.
Conventional field surveys and visual inspections are not well-suited for large-scale, continuous monitoring, highlighting the need for more efficient approaches. Sheffield et al., 2022 has shown that invasive aquatic plants can be detected using aerial imagery; however, while such imagery offers high spatial resolution, it is costly for frequent large-scale observation. Satellite-based approaches such as Shinohara & Kurita, 2025 have demonstrated potential but are not yet widely implemented for large-scale monitoring or automated alert systems. Therefore, this study aims to determine the spatial scale at which floating vegetation can be reliably detected using Sentinel-2 data and NDVI analysis.
2. Methods
Sentinel-2 optical imagery was used as the primary data source, with Sentinel-1 SAR data as supplementary information. The analysis compared June–September 2017, before the confirmation of alligator weed (baseline), with June–September 2018, when its presence was reported (target period). NDVI was used for vegetation detection, and MNDWI for water body extraction. A water mask was first applied to exclude terrestrial vegetation. Aquatic vegetation was then extracted using a combination of an NDVI threshold, adjacency to water bodies, and a minimum cluster size of three connected pixels, enabling detection of vegetation patches rather than isolated pixels.
Additionally, data from 2017 and 2018 were compared to identify newly emerged vegetation areas. A time-series analysis of NDVI from 2015 to 2019 was conducted to examine anomalous increases. Detection performance was evaluated using false positives(FP) and false negatives(FN), with the aim of identifying the minimum detectable infestation area while maintaining acceptable accuracy. The results were compared with publicly available data from Chiba Prefecture to validate the satellite-based detection.
3. Results
In 2018, aquatic vegetation increased markedly compared to 2017, with newly emerged patches showing high NDVI values across a wide area. The inclusion of the connectivity condition enabled patch-level detection and significantly reduced noise.
The NDVI time-series analysis also revealed a clear increase during the summer of 2018, consistent with the spatial expansion observed in satellite imagery. These findings suggest that, under certain conditions, satellite data alone can support automatic detection of aquatic plant formation and expansion.
4. Discussion and Future Challenges
The results indicate that restricting analysis to water bodies and combining NDVI with MNDWI effectively reduces false detections. Incorporating spatial conditions such as adjacency to water and minimum cluster size further improves detection stability.
However, several limitations remain, including the influence of water level fluctuations, turbidity, and suspended materials, as well as difficulty distinguishing alligator weed from other aquatic plant species. In addition, Sentinel-2’s 10 m spatial resolution may limit the detection of small, early-stage patches.
Future work should focus on testing the method in other regions, incorporating vegetation expansion models, utilizing higher-resolution data such as drone and commercial satellite imagery, and developing machine learning training datasets. Ultimately, this approach could be extended into an early warning system(WARN) with automated notification functions.
References
Sheffield, Kathryn J., et al. “Detection of Aquatic Alligator Weed (Alternanthera Philoxeroides) from Aerial Imagery Using Random Forest Classification.” Remote Sensing, vol. 14, no. 11, 2 June 2022, p. 2674, https://doi.org/10.3390/rs14112674.
Shinohara, Kengo, and Hideharu Kurita. “Distribution of Invasive Alien Species Alternanthera Philoxeroides in the Hakken River, Chiba Prefecture Visualizing by Satellite Data .” Proceedings of the 74th JSIDRE Annual Conference. Paper No. 3-35(P), pp. 255–256.