Sosnowsky's hogweed is one of the most widespread and dangerous invasive plant species in Russia. Combatting this weed is a top priority in recent environmental efforts. A promising method for identifying areas occupied by Sosnowsky's hogweed (SH) involves the use of Earth remote sensing data (ERS). Satellite images allow for the exploration of extensive territories in less time and often at lower costs compared to field surveys.

Detecting SH is most conveniently done using satellite imagery with meter-to-decimeter resolution, depending on the required precision. The most suitable data sources include Landsat-8, Sentinel-2, RapidEye, SPOT 6, 7, Triplesat, WorldView-2,3, Kanopus-V, and several others. Sentinel-2 imagery is well-suited for regional-level coverage studies. Its 10-meter spatial resolution allows for the detection of Sosnowsky's hogweed patches as small as 0.01 hectares. With a 5-day revisit frequency, it can cover large areas within relatively short periods (assuming clear weather conditions).

On satellite images, SH is distinguishable due to its enhanced reflectance in the green (550 nm) and near-infrared (830 nm) spectrum ranges. Figure 1 shows the spectral brightness curve of Sosnowsky's hogweed (obtained using a laboratory spectrometer) in different vegetation phases compared to typical herbaceous vegetation.

Рис. 1 Кривая спектральной яркости борщевика Сосновского

Fig. 1 Spectral brightness curve of Sosnovsky's burshevik

The graph illustrates that older SH leaves differ less from surrounding vegetation. Therefore, the most suitable time for identifying this weed on images is from May to July. Using images from different periods allows for the differentiation of SH from other plants with similar brightness characteristics during specific vegetation phases.

On natural color composite (RGB) satellite images, SH appears as a bright green color, standing out against other herbaceous vegetation (Fig. 2). Its high biomass provides a strong spectral response, allowing for the identification of even small areas with low projective coverage. Overall, this enables the detection of Sosnowsky's hogweed on satellite images with relatively high accuracy.

Рис. 2 Борщевик Сосновского на снимке Sentinel-2 (ярко-зеленый)

Fig. 2 Sosnovsky's borscht on Sentinel-2 image (bright green)

The following methods are most promising for identifying SH:

  • Use of spectral indices (HSI and NDVI);

  • Classification with supervised learning (Spectral Angle Mapper, Maximum Likelihood Classification, etc.).

The Hogweed Spectral Index (HSI) is specifically designed to distinguish Sosnowsky's hogweed from other plants and agricultural fields. When combined with the NDVI index for vegetation detection, the impact of anthropogenic objects can be minimized, enhancing SH detection accuracy. These indices ensure a high-quality outcome (Fig. 3). One advantage of this approach is that knowledge of SH distribution reference sites is not necessarily required.

Рис. 3 Ареалы борщевика Сосновского, выделенные с помощью индексов HSI и NDVI

Fig. 3 Sosnovsky's borer habitats identified using HSI and NDVI indices

Classification methods may provide slightly more accurate results, but they require knowledge of SH reference locations for classifier training. The spectral angle classification method (Spectral Angle Mapper) is particularly promising. This algorithm identifies pixels across the entire image that are most similar to the reference. The similarity measure is based on the angle in an n-dimensional space of spectral features - the smaller the angle, the more a pixel resembles the reference. For Sosnowsky's hogweed, the most accurate results are achieved with an angle of 0.02-0.03 radians. Figure 4 depicts SH areas identified using this method. Other supervised classification methods can also yield reliable results.

Рис. 4 Ареалы борщевика Сосновского, выделенные с помощью классификации Spectral Angle Mapper

Fig. 4 Sosnovsky's borer habitats identified by Spectral Angle Mapper classification

Satellite images are a reliable tool for the timely and accurate detection of Sosnowsky's hogweed distribution areas. The operational and comprehensive nature of ERS data is their primary advantage.

At present, the following Russian regions face particular challenges regarding the spread of Sosnowsky's hogweed:
Регионы, испытывающие особые проблемы по распространению борщевика Сосновского

Drawing on experience in monitoring and assessing SH distribution in Moscow, Moscow region, Smolensk, and Smolensk region, LLC "Innoter" is prepared to develop a comprehensive set of measures to combat Sosnowsky's hogweed.