Response of the regression tree model to high resolution remote sensing data for predicting percent tree cover in a Mediterranean ecosystem

Abstract

Percent tree cover is the percentage of the ground surface area covered by a vertical projection of the outermost perimeter of the plants. It is an important indicator to reveal the condition of forest systems and has a significant importance for ecosystem models as a main input. The aim of this study is to estimate the percent tree cover of various forest stands in a Mediterranean environment based on an empirical relationship between tree coverage and remotely sensed data in Goksu Watershed located at the Eastern Mediterranean coast of Turkey. A regression tree algorithm was used to simulate spatial fractions of Pinus nigra, Cedrus libani, Pinus brutia, Juniperus excelsa and Quercus cerris using multi-temporal LANDSAT TM/ETM data as predictor variables and land cover information. Two scenes of high resolution GeoEye-1 images were employed for training and testing the model. The predictor variables were incorporated in addition to biophysical variables estimated from the LANDSAT TM/ETM data. Additionally, normalised difference vegetation index (NDVI) was incorporated to LANDSAT TM/ETM band settings as a biophysical variable. Stepwise linear regression (SLR) was applied for selecting the relevant bands to employ in regression tree process. SLR-selected variables produced accurate results in the model with a high correlation coefficient of 0.80. The output values ranged from 0 to 100 %. The different tree species were mapped in 30 m resolution in respect to elevation. Percent tree cover map as a final output was derived using LANDSAT TM/ETM image over Goksu Watershed and the biophysical variables. The results were tested using high spatial resolution GeoEye-1 images. Thus, the combination of the RT algorithm and higher resolution data for percent tree cover mapping were tested and examined in a complex Mediterranean environment.

Understanding Forest Health with Remote Sensing-Part II—A Review of Approaches and Data Models

“…Semi-empirical approaches combine both empirical and physical modelling, e.g., by using the output from CR models to train neural networks to estimate biophysical parameters [235]. [239]; (AGB), [240,241] Ordinary least squares (height, density, DBH), [242] Reduced major axis (AGB), [243]; (LAI), [244] Canonical Correlation Analysis (forest structural conditions), [222] Redundancy Analysis (forest structural conditions), [245,246] Trend analysis (growth), [247] Non-parametric regression kNN (AGB, carbon), [248] CART (tree cover), [249]; (basal area, no. of trees) [250] RF (AGB) [243,251] SVM (height, density, DBH), [242] Physical Radiative transfer/canopy reflectance model Geometric-Optical (LAI), [252]; (AGB), [253]; (Chlorophyll), [254] Turbid-medium (LAI), [255] hybrid (allometry), [256] Computer simulation…”

Mapping Forest Health Using Spectral and Textural Information Extracted from SPOT-5 Satellite Images

“…The most frequently used approaches in forestry include regression and decision trees [102,103], artificial neutral networks [70,104] random forests [105][106][107], and support vectors [108,109]. The size of the training data set for machine learning greatly influences the stability and accuracy of the trained model [110].…”

Erosion Modelling In A Mediterranean Subcatchment Under Climate Change Scenarios Using Pan-European Soil Erosion Risk Assessment (PESERA)

“…The prevailing climate is characterized by Mediterranean with a mean annual precipitation of approximately 800 mm. The mean annual temperature is 19 °C (Donmez et al, 2015). The dominant soil types are Red-Brown Forest and Red Mediterranean Soils.…”