1. Introduction
Vegetation phenology is an important indicator to study the timings of the seasonal progression of plant activities through stages of dormancy, active growth, senescence, and back to dormancy [
1,
2]. Freely available satellite earth observation data such as the Moderate-Resolution Imaging Spectroradiometer (MODIS), Landsat-8, and Sentinel-2 provide wider coverage with varying spatial, temporal, and spectral resolutions to understand the seasonal vegetation dynamics from local to global scales [
3,
4,
5,
6]. MODIS vegetation indices products such as the normalized difference vegetation index (NDVI) [
7,
8] and the enhanced vegetation index (EVI) [
9] have been widely used for understanding the temporal behaviour of land surface vegetation phenology [
10,
11,
12]. Previous studies have used the potential of these time-series vegetation indices as a robust metric for estimating the photosynthetic activity, developmental status, and productivity of vegetation and retrieved land surface phenology metrics such as the beginning of onset, end of senescence, and length of growing season for different vegetation types [
13,
14,
15,
16]. However, validation of satellite-derived phenology metrics remains uncertain and challenging due to the limited availability of field observations at high temporal and spatial resolutions, mainly for remote locations or areas with difficult accessibility.
Digital repeat photography has become a reliable tool for a wide range of ecological applications due to its low cost, easy set up, temporally frequent observations, and high-resolution data in red, green, and blue (RGB) channels [
17,
18]. Digital cameras used for phenological observation, also called PhenoCam, have enabled the detection of leaf phenological events through the analysis of RGB colour changes over time [
19,
20]. PhenoCam provides several clear advantages over human observations of phenology because of the ability to collect automatically repeated images at high temporal resolution (daily or hour scale) and across wide spatial scales (from the individual to the landscape). These cameras become useful specifically for remote areas or where the accessibility of sites is prevented by harsh climatic conditions [
21,
22]. Direct phenological observations of canopy vegetation using PhenoCams filled the “gap of observations” between satellite and the traditional on-the-ground data [
17,
18]. In recent decades, imagery from PhenoCam has been used as an alternative to field observations to provide a more straightforward solution to monitor vegetation growth from canopy to landscape scales at hourly or daily temporal resolutions [
20,
23,
24,
25,
26]. Therefore, PhenoCam can be considered as a robust tool to evaluate and compare phenological metrics derived from satellite data [
14,
25,
27,
28].
Different PhenoCam indices, such as green chromatic coordinate (GCC), excess green index (ExG), normalized difference of the green and red bands (VIgreen), the red chromatic coordinate (RCC), or vegetation contrast index (VCI), can be derived using its red, green, and blue colour channels for vegetation phenology analysis [
14,
21,
25,
29]. Among these, GCC is the most widely used and reliable proxy to monitor the canopy phenology of coniferous species [
25,
26,
30]. To our knowledge, only one study is available comparing field observations of bud phenology in black spruce (
Picea mariana (Mill.) B.S.P.) with PhenoCam images. This study focused on a single site, thus lacking a global view of the performance of satellites in assessing vegetation indices across the latitudinal distribution of the boreal forest [
26]. Our study is unique in terms of the spatial amplitude of the monitoring, which is based on a PhenoCam network measuring phenology of the most important boreal species in North America.
The relationship between PhenoCam and satellite data raises concerns, in particular for the lack of infrared bands in various PhenoCams. To compute the vegetation index from digital images, PhenoCams use visible sections of the electromagnetic spectrum compared to the satellite-based vegetation indices that use infrared light. Some previous studies tried to establish links between PhenoCam and the traditional vegetation indices (e.g., NDVI) and compared phenophase transition dates such as the start, end, and length of the growing season for various tree species [
28,
31]. These studies demonstrated the presence of significant correlations, although satellite estimated an earlier start and later ending of the growing season compared with PhenoCams. Yet, the relationship between PhenoCam and remote-sensing-derived vegetation indices (e.g., NDVI and EVI) remains unknown for species of the boreal ecosystem, the largest biome in the world in terms of extension and importance. Studies are available from semiarid tropical forests of Brazil, temperate deciduous forests of eastern North America, grasslands, and mixed land cover types of North America [
14,
27,
32]. The results suggest that EVI is more performant than NDVI for assessing phenology. The question remains whether these results are true also at the higher latitudes of the Earth. The boreal biome experiences long winters with the soil covered by snow that can affect the temporal variation in vegetation greenness. The physical interaction between vegetation index, growth reactivation, and snow cover is not completely disentangled and is worthy of deeper studies [
14,
27,
32].
We present a phenological study on a boreal species (black spruce) that combines the broad extent of a satellite-based platform with the fine spatial and temporal scale observations of a digital camera (PhenoCam) across the entire latitudinal gradient of the closed boreal forest in Quebec, Canada. Our study links satellite-derived vegetation indices, collected at a resolution of 16 days, with PhenoCam data, obtained at a daily resolution, to improve landscape-scale phenology understanding of the boreal region
We expect to detect significant differences between RGB PhenoCam-based and satellite-based vegetation indices that use infrared light in plant phenology measurements. Therefore, the main objective of this paper was to compare the RGB PhenoCam-based (GCC) phenology with multispectral MODIS satellite-based vegetation indices (NDVI and EVI), collected in six black spruce stands of Quebec, Canada. While previous studies have linked in situ or satellite data with digital camera data, for different types of vegetation or tree species [
19,
28], our investigation focus on a unique network of PhenoCam data installed in black spruce stands dominated by the more important boreal species of North America.
5. Conclusions
Phenological observations in remote locations are scarce, and during recent decades, PhenoCam digital imagery has become popular, providing a more straightforward solution to understand plant phenology at both high spatial and temporal resolutions.
This study compared PhenoCam-based GCC at a daily temporal resolution with 16-day temporal resolution MODIS, NDVI, and EVI for six sites across the boreal forests of Quebec, Canada. Our study highlighted differences between the dynamics of canopy and stand-level phenology of black spruce measured from near-surface digital photography and satellite data, respectively, and therefore, the proposed approach has the potential to improve the relationship between landscapes captured by PhenoCam cameras and satellite sensors at different spatial resolutions [
29].
We found a close relationship of PhenoCam GCC with satellite-sensor-based vegetation indices. EVI performs better than NDVI, with an average absolute difference of 1 to 5 days in evergreen black spruce. These results rejected our initial hypothesis, which expected a significant difference between RGB-based GCC and infrared-based EVI and NDVI. Researchers can use this approach to understand the phenology of single or multiple species.
This study presents a unique approach to compute GCC-based phenological metrics despite missing PhenoCam images for some locations and incorporates the issue of field of view shifts of a camera caused by heavy winds in remote locations [
34,
52]. We provide evidence that EVI has a more reliable spectral vegetation index for estimating phenology in black spruce than NDVI. The time lag for the seasonal dynamics between EVI and PhenoCam was short and not significantly different for both onset and ending. Our study, based on years of data collection by a network of PhenoCams covering a large study area, builds a bridge between PhenoCam and satellite remote-sensing-based phenological observations. Our findings are useful for designing phenological investigations on wide regions, mainly in remote sites where field data collections at regular time intervals are prevented by the extreme weather and the limited accessibility.