Regression kriging to improve basal area and growing stock volume estimation based on remotely sensed data, terrain indices and forest inventory of black pine forests

Ferhat Bolat,Muammer Şenyurt,İlker Ercanlı,Alkan Günlü,Sinan Bulut

doi:10.33494/nzjfs502020x49x

Ferhat Bolat, Muammer Şenyurt + Show 3 more

Open Access

https://doi.org/10.33494/nzjfs502020x49x

Copy DOI

Abstract

Background: The use of satellite imagery to quantify forest metrics has become popular because of the high costs associated with the collection of data in the field.Methods: Multiple linear regression (MLR) and regression kriging (RK) techniques were used for the spatial interpolation of basal area (G) and growing stock volume (GSV) based on Landsat 8 and Sentinel-2. The performance of the models was tested using the repeated k-fold cross-validation method.Results: The prediction accuracy of G and GSV was strongly related to forest vegetation structure and spatial dependency. The nugget value of semivariograms suggested a moderately spatial dependence for both variables (nugget/sill ratio approx. 70%). Landsat 8 and Sentinel-2 based RK explained approximately 52% of the total variance in G and GSV. Root-mean-square errors were 7.84 m2 ha-1 and 49.68 m3 ha-1 for G and GSV, respectively.Conclusions: The diversity of stand structure particularly at the poorer sites was considered the principal factor decreasing the prediction quality of G and GSV by RK.

Highlights

Forest inventory studies are conducted to quantify forest attributes such as basal area (G), growing stock volume (GSV), biomass and carbon sequestration that are providing essential information for forest managers
We focused on the predictability of G and GSV using Multiple linear regression (MLR) and regression kriging (RK) based on Landsat 8, Sentinel-2 data, and terrain indices
EVI obtained from Landsat 8, NLINIRn2 obtained from Sentinel-2, elevation and STI were the best independent variables explaining G and GSV

Summary

Introduction

Forest inventory studies are conducted to quantify forest attributes such as basal area (G), growing stock volume (GSV), biomass and carbon sequestration that are providing essential information for forest managers. There remain a number of issues like unequal or fragmented forest distribution, differing tree species and age classes, which can lead to difficulty when trying to maximise the spatial variance explained when modeling forest metrics (Chirici et al 2008; Gebreslasie et al 2008; Ingram et al 2005; Lu et al 2004) Because of these drawbacks, and in order to estimate forest metrics at an acceptable level of confidence and at a fine level of detail, scientists have combined remotely sensed data and ground measurements using methods such as ordinary least square, machine learning, and geo-statistic methods in the last ten years (Mallinis et al 2004; Franco-Lopez et al 2001; Ingram et al 2005; Meng et al 2009; dos Reis et al 2018). The use of satellite imagery to quantify forest metrics has become popular because of the high costs associated with the collection of data in the field

Objectives

Methods

Results

Discussion

Conclusion