Remote Sensing Vegetation Species using Multispectral Satellite Products

Multispectral satellite imagery has become an essential tool for analyzing vegetation, offering insights into vegetation health [1], water stress [2], and predicting species niches [3], among other applications. Here we will focus on identifying vegetation species [4, 5, 6, 7], which has significant implications for ecological management and restoration. Understanding species composition from satellite data allows for detecting invasive weeds, conducting habitat assessments, and monitoring remediation projects, without the requirement for site visits.

Multispectral satellite imagery can distinguish vegetation species based on their unique spectral signatures.Ground truth labels, obtained through field surveys or high-resolution aerial imagery, provide essential reference points for training and validating classification models. These labels correspond to specific plant species and are used to calibrate spectral response patterns. Corresponding multispectral satellite products are then obtained, examples including WorldView-3 (0.3m) of Sentinel-2 (10m).

By modeling the spectral response of different species across multiple wavelengths, we can distinguish vegetation types based on their unique spectral signatures over time. By considering spectral response over a time series, modelling can incorporate vegetation dynamics, and seasonal changes, which can improve classification accuracy. The study by Blickensdörfer et al. [7], albeit on discriminating crop species, provides a clear example of this modality (Fig.1 / Fig.2). Note that, in this case, the authors utilised both Sentinel-1 and Sentinel-2 products.

Fig.1. Time series vegetation ground truth labelling for 4 agricultural sites [7].

Fig.2. Spectral signatures for different agricultural species over time. Note how the spectral signatures vary over the growing period, demonstrating the advantage of time series analysis for species-level classification [7].

We will now briefly consider some studies, which demonstrate the utility of this modality.

Fig.3 ground truth, and remote sensed, Japanese knotweed using time series of Sentinel-2 products [8, 9].

Smerdu et al. reported successfully remote sensing the invasive Japanese knotweed using Sentinel-2 products [8, 9] (Fig.3). They noted the requirment for timeseries data, in order to bserve spectral (phenological) signatures of the plant species over time, and that the spatial resolution of the satellite product was limiting when it came to small stands.

A CSIRO study by Shendryk et al. utilsied WorldView-3 products to detect gamba grass with up to 90% accuracy (under optimal conditions) [10] (Fig.4). The detection of this invasive species allowed for more targeted management. They report that the wide range of spectral data allowed them to differentiate between gamba grass and native grass species. Modelling utilised support vector machine (SVM), and was single-pass (not timeseries).

Fig.4 Gamba grass detection using WorldView-3 modelling via SVM [10].

Lastly, Alonso et al. have completed a number of studies on the remote sensing of Eucalypt species, using multispectral products [5, 6]. The authors evaluated both Sentinel-2 and WorldView-3 products, utiling neural network [5] and random forest models [6]. They noted that the "highest accuracies were obtained when using...Sentinel-2 data in a multitemporal approach", particculary during the emergence of spring shoots. They also noted the benefits of using LiDAR. Worldview-3 had moderate discriminatory performance, which is intersting, and may indicate the importance of phenology (Fig.5).

Fig.5 Example Eucalyptus species classification using supervised random forest training on WorldView-3 products [6].

In conclusion, multispectral satellite imagery offers a powerful tool for ecological management at a landscape scale. Unlike simple vegetation cover assessments, multispectral data allows for fine-grained evaluation of habitat quality, distinguishing between native species, invasive weeds, and ecologically sensitive areas. Example applications include pre-construction planning, where habitat vulnerability must be assessed, or mine remediation, where monitoring ecosystem recovery is required. Once established, such remote monitoring may reduce the need for site visits, enhance decision-making, and support sustainable land management practices.

References

[1] Govender, M., Chetty, K., & Bulcock, H. (2007). A review of hyperspectral remote sensing and its application in vegetation and water resource studies. Water SA, 33(2), 145-151. https://doi.org/10.4314/wsa.v33i2.49149

[2] Thenkabail, P. S., Smith, R. B., & De Pauw, E. (2000). Hyperspectral vegetation indices and their relationships with agricultural crop characteristics. Remote Sensing of Environment, 71(2), 158-182. https://doi.org/10.1016/S0034-4257(99)00067-X

[3] Franklin, J. (2010). Moving beyond static species distribution models in support of conservation biogeography. Diversity and Distributions, 16(3), 321-330. https://doi.org/10.1111/j.1472-4642.2010.00641.x

[4] Carrión-Klier, C., Moity, N., Sevilla, C., Rueda, D. and Jäger, H., 2022. The importance of very-high-resolution imagery to map invasive plant species: Evidence from Galapagos. Land, 11(11), p.2026.

[5] Alonso, L., Rodríguez-Dorna, A., Picos, J., Costas, F. and Armesto, J., 2024. Automatic differentiation of Eucalyptus species through Sentinel-2 images, Worldview-3 images and LiDAR data. ISPRS Journal of Photogrammetry and Remote Sensing, 207, pp.264-281.

[6] Alonso Martinez, L., Picos Martín, J. and Armesto González, J., 2022. Mapping eucalyptus species using worldview 3 and random forest. ISPRS-International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences.

[7] Blickensdörfer, L., Schwieder, M., Pflugmacher, D., Nendel, C., Erasmi, S. and Hostert, P., 2022. Mapping of crop types and crop sequences with combined time series of Sentinel-1, Sentinel-2 and Landsat 8 data for Germany. Remote sensing of environment, 269, p.112831.

[8] The use of satellite time series for detection and mapping of invasive species. https://www.uia-initiative.eu/en/news/how-do-we-detect-invasive-nonnative-vegetation-satellite-images

[9] Smerdu, A., Kanjir, U. and Kokalj, Ž., 2020. Automatic detection of Japanese knotweed in urban areas from aerial and satellite data. Management of Biological Invasions, 11(4), p.661.

[10] Shendryk, Y., Rossiter-Rachor, N.A., Setterfield, S.A. and Levick, S.R., 2020. Leveraging high-resolution satellite imagery and gradient boosting for invasive weed mapping. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13, pp.4443-4450.