Mapping leaf area index in subtropical upland ecosystems using RapidEye imagery and the randomForest algorithm

Canopy leaf area is the main factor for primary production, energy exchange, transpiration, and other physiological attributes related to ecosystem processes ([50], [31], [8], [65], [2]). The Leaf Area Index (LAI), defined as the projected leaf area per unit ground surface area, is frequently used to quantify the canopy leaf area. Thus, LAI is one of the key biophysical variables required by many process models describing the soil/plant/atmosphere system ([4], [39]). Because canopies and some of their characteristics can be directly observed from above, LAI is among the major variables of interest in remote sensing analyses ([40]), and its importance has led to considerable efforts to map its distribution over a variety of spatial and temporal scales ([16], [48], [74], [39], [63]). The majority of studies used (semi-) empirical relationships between LAI and combinations of spectral bands, namely vegetation indexes (VI), for LAI mapping ([4], [68]). In a review paper on the relationships between remotely-sensed VIs and canopy attributes, Glenn et al. ([29]) point out that VIs are often strongly related to light dependent physiological processes occurring in the upper canopy, but often exhibit only moderate relationships to detailed features of canopy architecture, such as LAI. VIs are generally regarded as important but have their limitations since they only utilize a fraction of the spectral information available in remote sensing data ([30]). Similarly, it has been pointed out that there is not enough evidence that spectral reflectance in the visible and near-infrared is sufficient to estimate LAI in forests, especially under close canopy situations ([40]). From their review, Glenn et al. ([29]) concluded that remote sensing models exclusively based on VIs to estimate LAI are, in particular, subject to error and uncertainty. Another fact to be considered is that LAI is a variable that cannot be directly measured in the field. Therefore, all efforts to correlate VIs derived from remote sensing interpretation to field observations are characterized by the lack of ground-level measurable parameters, as both information are results of modeling efforts. Additional information derived from remote sensing images with the potential to improve LAI predictions are texture features. Texture features quantify the spatial variability of pixel values within a neighborhood defined by a moving window, and thus, complement the spectral information with a spatial component ([17]). As the variation in texture is related to changes in the spatial distribution of vegetation ([70]), image texture can be linked to the spatial distribution of vegetation ([17]). The scope of this paper is to investigate the applicability of RapidEye imagery, which is optimized towards vegetation analyses, for LAI mapping along a disturbance gradient, ranging from heavily disturbed shrub-land to mature mountain rainforest. By incorporating image texture features into the analysis we aim at assessing the potential quality improvement of LAI maps and the reduction of uncertainties associated with LAI maps compared to those based on VIs solely ([29]). To predict LAI we use the Random Forest (RF) algorithm ([10]), an increasingly and widely used statistical modeling technique for predictive biological mapping ([51]). Field data come from forest inventory plots located in the uplands of Xishuangbanna, China. Xishuangbanna is located in the transition zone between tropical southeastern Asia and subtropical and temperate China, resulting in the region with the highest biodiversity in China ([73], [42]). This high biological diversity is threatened by the expansion of rubber (Hevea brasiliensis) plantations ([71], [42], [72]). In the time from 1976 to 2003 Xishuangbanna's estimated forest cover declined from 69% to less than 50% in conjunction with a decline of mean size of forest patches from 217 to 115 ha ([43]). The forest type most affected by the expansion of rubber plantations was tropical seasonal rain forest ([42]). For a better understanding of ecosystem dynamics and processes over large and remote areas, as faced in this region, LAI maps are of particularly high value ([14]).

---

• View source

---

DOI:

https://doi.org/10.3832/ifor0968-006

---

Altmetric score:

Dimensions Citation Count: