Monitoring of defoliating insects

Detection of insect-induced defoliation with means of remote sensing is regarded to still be in its early stage and the spectral responses to damaged vegetation are not completely understood (Zhang et al. 2010; Wang et al. 2010). Further method development is needed before comprehensive adaptation at the operational level (Jepsen et al. 2009; Rullan-Silva et al. 2013). More challenging than the detection of needle loss is assessing the severity of

defoliation (Rullan-Silva et. al 2013). Although classification accuracies of 70%-80% for three severity classes have been observed (Rullan-Silva et al. 2013), it has to be kept in mind that especially low intensity defoliation is very difficult to monitor (Zhang et al. 2010).

Further, detection of defoliation within a sparse canopy cover is even more difficult (Dennison et al. 2009), such as in typical boreal pine forests.

Healthy and green vegetation results in a well-known pattern of spectral signature over the electromagnetic spectrum (Rullan-Silva et al. 2013). The pattern reveals the highest absorption and lowest reflectance in the visible range of the spectrum, followed by a plateau of highest reflectance in the NIR range. Within the range of visible light, a peak in the reflectance occurs in the green band (~0.54 µm), corresponding to the green color of a healthy crown (Rullan-Silva et al. 2013). It is suggested that the visual range is the most consistent indicator of plant stress shown in foliage (Carter 1993; Jensen 2005). A tree is regarded to suffer from stressors, when there is indication on change in the health condition in the foliage (Rullan-Silva et al. 2013). Under stress, reflectances of green and red are increased as the foliage turn into yellowish or chlorotic. Increase in reflectance in the NIR region seems to be consistent only at extreme stress levels (Rullan-Silva et al. 2013). Healthy foliage has a high reflectance of the NIR range. That is partly due to additional reflectance from the energy transmitted through the leaf and re-reflected by the leaves below (Jensen 2005). Hence, changes in the NIR region may be utilized in detecting defoliation (Rullan-Silva et al. 2013).

The region of 0.65-0.7 µm may be suitable for early detection of forest damage; the first indicators of stress are seen as increase in reflectance of the red edge (0.7 µm), sifting towards shorter wavelengths (Jensen 2005). Remote sensing sensor’s ability to record narrow sensitive ranges, such as hyperspectral data, may improve the detection accuracy (Carter 1993, 1996; Jensen 2005; Mutanga et al. 2009). Hyperspectral data may also be suitable to assess levels of chlorophyll absorption and photosynthetically active radiation related to, e.g., insect-induced defoliation (Jensen 2005). Shortwave infrared (SWIR) wavelengths have been useful in detecting insect-induced needle loss (Skakun et al. 2003; Wang et al. 2007;

Goodwin et al. 2008; Coops et al. 2010). There are two peaks in the SWIR reflectance in case of healthy vegetation (~1.6 µm and ~2.2 µm) located between atmospheric water absorption bands (Rullan-Silva et al. 2013). These ranges are reflecting water content of healthy foliage tissue, which is correlated with plant transpiration rates. With decreasing moisture, the infrared energy becomes scattered and the reflectance increases (Jensen 2005). Timing of data acquisition is also very important in case of defoliating insects. One key to a successful assessment of defoliation is the biological window referring to an optimal period for ‘visual expression of major forest pests and related damage’ (Wulder et al. 2004). The period vary depending on factors, such as host tree phenology, climate conditions, and natural enemies.

It is typically in accordance with the peak foliage period of the host (Rullan-Silva et al. 2013).

Further, this period is often short emphasizing the role of high temporal resolution. Typical cases are ephemeral outbreaks by defoliators especially in areas with frequent cloud cover (Hicke et al. 2012; Rullan-Silva et al. 2013).

Most of the remote sensing studies on defoliating insects have utilized low to medium resolution data (Senf et al. 2017b). Medium-resolution Landsat and SPOT satellites have been the most widely utilized. Landsat data has been the most popular sensor (e.g., Luther et al. 1997; Radeloff et al. 1999; Franklin et al. 2003; Hall et al. 2003; Babst et al. 2010; Paritsis et al. 2011; Meigs et al. 2011; Olsson et al. 2012; Townsend et al. 2012; Thayn 2013;

Sangüesa-Barreda et al. 2014; Rullan-Silva et al. 2015; Senf et al. 2015). MODIS data have been utilized in, e.g., mapping defoliation by pine sawflies (Eklundh et al. 2009), gypsy moth (de Beurs and Townsend 2008; Spruce et al. 2011), and geometrid moths (Jepsen et al. 2009;

Olsson et al. 2016a). SPOT data have been utilized by, e.g., Muchoney and Haack (1994), Fraser and Latifovic (2005), and Gilichinsky et al. (2013). Kharuk et al. (2004, 2007, and 2009) have utilized NOAA AVHRR, MODIS, and SPOT VEGETATION data in monitoring Siberian silk moth (Dendrolimus superans sibiricus Tschetverikov). High-resolution satellite data has been so fat less utilized in monitoring insect defoliation. RapidEye images have been utilized by Adelabu et al. (2014) and Sentinel-2 based vegetation indices by (Hawryło et al.

2018).

Although aerial photographs are the most utilized data in forestry (Hall et al. 2003), they have not been as widely utilized as lower resolution satellite-based images in needle loss detection. Aerial digital photography has been used in some studies detecting insect-induced defoliation, such as by pine looper (Bupalus piniaria L.) (Långström et al. 2004) and common pine sawfly (Ilvesniemi 2009). Haara and Nevalainen (2002) classified non-specified Norway spruce (Picea abies L. Karst.) defoliation from aerial images. Aerial video data was used by Franklin et al. (1995) to detect defoliation by the western spruce budworm (Choristoneura occidentalis Freeman). Leckie and Ostaff (1988) tested use of 11 band multispectral scanner data in classification of spruce budworm induced defoliation. Kantola et al. (2010) combined aerial images with high pulse density LiDAR data to classify Scots pine (Pinus sylvestris L.) defoliation.

Even though use of LiDAR enables assessment of vegetation structure it has not been widely utilized so far in mapping of defoliation. Further, the features on LiDAR data that could be associated with insect defoliation are less investigated than the associated spectral traits. It has been observed, however, that various metrics calculated form LiDAR point clouds, such as canopy-based quantile metrics, can be linked to foliage biomass (e.g., Magnussen and Boudewyn 1998; Lim and Treitz 2004). Foliage biomass have been directly estimated from point clouds (Riaño et al. 2004), or full waveform data (Lefsky et al. 1999).

LiDAR have been utilized before in detection of defoliation by pine sawflies (Solberg et al.

2006, 2010; Kantola et al. 2010; Hanssen and Solberg 2007). Use of terrestrial laser scanning in classification of defoliation was tested by Huo et al. (2019). In contrast to LiDAR, active remote sensing SAR data seems to contribute only modestly to defoliation assessment (Rullan-Silva et al. 2013).

There is an increasing trend in comparing vegetation indices from different times to evaluate the changes by defoliators (Senf et al. 2017b). Multi-temporal satellite data derived vegetation indices have been utilized in, e.g., studies on defoliation by western spruce budworm (Meigs et al. 2015; Senf et al. 2015), Hungarian spruce scale (Physokermes inopinatus Danzig and Kozár) (Olsson et al. 2012), and pine processionary moth (Thaumetopoea pityocampa Denis and Schiffermüller) (Sangüesa-Barreda et al. 2014). The used vegetation indices utilized NIR or SWIR regions. Changes in NIR and SWIR enables detection of chlorosis and structural changes in the canopy, however, the relationship between the changes in spectral signature and insect defoliation is not as well understood as for bark beetles (Senf et al. 2015). Use of dense time-series may improve detection of defoliation with information on insect peak performance periods within a season (Fraser and Latifovic 2005). In case of broadleaved tree species, dense time-series may even compensate coarse spatial resolution (Senf et al. 2017b). Fusion of remote sensing data with different spatial and temporal resolution may be used to enhance the mapping of defoliation (Gärtner et al. 2016).