**4. Discussion**

#### *4.1. Remotely Sensed Estimation of LAI and TSA*

Our field investigation suggests that the LAI is highly correlated with TSA in early post-fire landscapes. Given that surveying TSA in the field is time-consuming and labor intensive, utilizing a DHP system can be a more efficient way to monitor post-fire tree sapling recovery. We found that all Landsat-derived spectral indices exhibited statistically significant correlations with both LAI and TSA, but the variance explained was usually less than 50%. EVI2 and NBR were the most predictive spectral indices for LAI and TSA respectively, but we found that NBR performed consistently well for simulating LAI and TSA in two Landsat analysis cases. This suggests that NBR may be the best choice for monitoring post-fire vegetation dynamics in our study area. A similar finding was also reported by Chen, et al. [75], who found that NBR is the most sensitive indicator of post-fire vegetation recovery in the Black Hills National Forest of the USA, suggesting that the value of NBR extends beyond our boreal forest study system. Previous studies have also shown that NBR is better than NDVI for quantifying burn severity [26,76], further supported by Kennedy, Yang and Cohen [54] and White, Wulder, Hermosilla, Coops and Hobart [53], who concluded that the NBR is a very useful indicator for detecting fire disturbance and monitoring post-fire vegetation trajectory.

Although Landsat-derived spectral indices exhibited significant correlations with the LAI and TSA, we found that the WorldView-2 (VHR) imagery performed even better. PPCT had a stronger explanatory power for LAI than for TSA because our DHP measurement of LAI consisted of canopy foliage of both surviving trees and tree saplings. The in-field upward canopy photograph results closely match PPCT derived from WorldView-2 imagery, further confirming the value of this technique. However, PPCT cannot wholly reflect TSA as there were surviving trees that exhibited disproportionately large canopy projection areas in the imagery when compared to tree

saplings. The low spectral separability of surviving trees and tree saplings limited the applicability of WorldView-2 imagery for monitoring TSA.

Image textures are commonly used as auxiliary inputs to provide spatial features that spectral information cannot characterize. Many previous studies have reported that including image textures as predictors can improve the estimation of forest structural parameters such as tree diameter [77], stand density [61] and crown diameter [78]. However, image textures did not exhibit strong explanatory power to either LAI or TSA in our study. One reason is that crown shadow is an important factor in representing spatial variation as reflected in the imagery [61]. In contrast to the coarse surface of mature forests, the post-fire landscape where dominated by tree sapling that did not represent very strong shadow effects (contrasts) in the imagery (Figure 3c). In addition, the value of textural features has no explicit relationship to spectral features and the same value of textural attributes may represent very different land surface properties. For example, post-fire landscapes usually exhibit high homogeneity, however similar homogeneity values may be corresponding to different land cover compositions.

#### *4.2. The Response of Forest Recovery to Spatial Controls*

We found that including surrogates of intra-specific competition (i.e., understory coverage) and density of surviving trees (i.e., shadow coverage) greatly improved model fit. According to partial dependence curves, we found a negative effect of understory (grass and shrub species) coverage on tree sapling regeneration (Figure 7a). Understory species usually act as pioneers in post-disturbance landscapes and rapidly occupy available spaces and resources. In addition, some understory species are resistant to wildfire and recover rapidly. By comparison, tree saplings may require longer timespans to establish; they will compete with understory species for available resources and may be limited by seed availability.

Tree sapling regeneration (estimated by LAI) is very sensitive to shadow coverage, suggesting that even a few surviving trees can strongly promote post-fire recovery (Figure 7b). The nearest distance to unburned area, which reflects the importance of seed inputs from out of the burned area, significantly impacted LAI recovery with areas near the edge of boundaries having a higher opportunity to receive seeds from unburned areas. Our results showed that edge effects extend for a distance of approximately 700 m for this fire (Figure 7h), which can partially explain why the core area exhibited poor tree recovery. Trees that survive fire are thus critical for supporting reestablishment of tree community post-burn [79,80]. Nonetheless, if the number of surviving trees was over a certain threshold, canopy closure was negatively affected because our study area was dominated by mature larch forests, which have wide tree spacing and relatively open canopies. Surviving trees may prevent the establishment of saplings, likely through resource competition.

Burn severity is often described as a legacy effect as high severity fires have a long-term impact on forest ecosystems through removing surface organic layers and creating canopy openings through killing nearly all large trees. Burn severity is well-reported to have adverse effects on forest restoration in boreal forests [81] and subalpine forest ecosystems [82]. Our results sugges<sup>t</sup> that high severity burns decrease tree sapling regeneration post-fire, consistent with previous findings in similar ecosystems. In our study wildfire exerted limited effects on tree recovery when dNBR was lower than 0.730 (Figure 7c), which is related to low severity (dNBR < 0.418) and moderate severity (dNBR < 0.942) levels according to published thresholds from Fang and Yang [26]. Surviving trees contributed a large part of canopy foliage at low and moderate burn severity areas and combining with tree sapling foliage to result in high post-fire LAI under low and moderate burn severities. We found that LAI recovery decreases dramatically with dNBR increasing, implying that moderate to high severity fires have profound effects on diminishing forest resilience. Severe fire behavior may cause crown fires in our study area [45], which will destroy areal seed banks in serotinous cones of coniferous trees [8]. At the same time, severe surface fires will consume fine fuels such as litters and soil organic materials, create exposure of soil and base rock, and destroy surface seed beds. Given the key roles of coniferous tree species on maintaining the unique functions of boreal forest communities [9,14,23], recovery of coniferous tree species has raised increasing concerns [10,21]. Coniferous tree species have relatively shorter seed dispersal distance (e.g., larch < 150 m, and Scotch pine ~ 500 m) than broadleaf trees (e.g., birch and aspen > 1000 m) due to heavier seed mass [83,84]. Many studies found that Great Xing'an boreal forests will experience severe fire activities that were characterized as more frequent and large burns with higher severity along with climatic changes and fuel accumulation [18,85,86]. It is not difficult to speculate that coniferous forests may experience more recovery limitations, which has been validated in boreal forests of Central Siberian [84] and North America [8,11]. Besides, very high severity fires may have also killed broadleaf tree species and understory species whose roots would otherwise provide a source of asexual reproduction even if above-ground portions are killed, creating opportunities for primary succession.

**Figure 7.** Partial dependence plots of a-h show influences of eight selected spatial controls (variable names see x-axis) on forest recovery in terms of leaf area index (LAI). Variable abbreviations were described in Table 3.

Topography is usually considered an environmental filter because it theoretically determines the drainage, thermal, solar radiation, surface roughness and other site factors which are usually exert considerable influence on tree species distribution and growth [87,88]. Total solar radiation during the growing season is typically the most important topographical factor and we found that LAI increased significantly when solar radiation exceeded 90,000 WH/m<sup>2</sup> (Figure 7d). At our study site, southern slopes and flat areas favored broad leaf tree species, such as white birch and aspen, that can regenerate through roots sprouting [21,47]. The TPI, TWI and elevation shared similar relative importance. Higher TPI indicates higher slope position (Figure 7e), and thereby sites that tend to have good drainage and open space to accept seed rain, especially low weight seeds of white birch and aspen dispersed via wind.

Our results sugges<sup>t</sup> that TWI is inversely related to LAI (Figure 7f) while elevation has a positive influence LAI (Figure 7g). TWI is known to have a strong direct relationship with soil moisture in boreal forests ecosystems [89] and our findings are consistent with Cai, Yang, Liu, Hu and Weisberg [21], who reported similar relationships between soil moisture, elevation and broad leaf tree sapling abundance through field investigation. However, while Cai, Yang, Liu, Hu and Weisberg [21] found moist soil valley bottoms favorable for larch saplings, our study finds them to be more limited in terms of tree recovery than more elevated locations. One explanation is that at our site permafrost is usually present in those moist valleys [90], and the seasonal thaw may prevent seed germination. In addition, moss (*Sphagnum cuspidatulum*) and grass (*Carex appendiculata*) is often thick in moist valley bottoms and can prevent larch seed from encountering the mineral soil necessary for their establishment.
