Riihimäki, Henri
(Helsingfors universitet, 2013)
The ongoing global change challenges us to examine the key factors of rapidly changing northern ecosystems. One of the most important factors in these environments is living vegetation biomass, also known as phytomass. This thesis examines above ground phytomass in an artic-alpine environment, located in northwesternmost Finland and Ráisduottarháldi –area, Norway.
The most important aim of the study was to produce a best possible estimate of the phytomass in the study area. Typically, phytomass modelling in artic-alpine areas has been done by using linear regression models having spectral vegetation index (SVI), usually NDVI, as an explanatory variable. Goodness of the model is typically assessed by coefficient of determination (R2). This thesis expands this approach and tests different SVI's alongside NDVI. Bias, root mean square error (RMSE), and correlation of observed and predicted phytomasses are used in addition. The effect of sample size is also briefly tested.
Factors affecting phytomass, such as topography, were also examined. Topographic variables, such as the topographic wetness index (TWI), slope, potential yearly radiation and curvature were derived from digital elevation model and used as a predictors. Rock and soil variables were also used, but the quality of the data was found poor. In addition to linear regression models (LM), generalized linear models (GLM) and variation partition were used to find out wether the simple SVI-models can be improved by adding topographic factors into the models. Boosted regression trees (BRT) were utilized to find out the importance of individual effects of topographic factors to phytomass.
NDVI was found to be the best SVI to predict phytomass (R2 61,6 %, RMSE 593,5 g/m2). However, the model was slightly biased (–4,3 %), although not statistically significantly. Forest areas cause significant deviaton to the data, which might explain why the explanatory power of the NDVI model is lower compared to other similar studies carried in pure arctic environments. Based on variation partition, the NDVI-model cannot be improved by topograhic nor soil or rock variables. In BRT-models, elevation was found to be the single most important variable explaining phytomass. The relative importance of elevation in the phytomass model was 72,8 %. Potential radiation (11,3 %) and calcium contet of parent material (11,4 %) were also important. TWI also had a slight effect, as its relative importance was 4,9 %. Curvature was not a significant factor in the models.
Based on the linear regression model (NDVI), phytomass varied between 0– 6790 g/m2: Mean phytomass of the study area was 687 g/m2. Most of the phytomass is located in trees and other vascular plants located in low elevations (< 600 m a.s.l.). Phytomass decreases rapidly above treeline, which is typically located around 600–700 m a.s.l. at the study area. Only 0,1 % of the total phytomass is located above 1000 m a.s.l. South- and southwestern slopes have higher phytomasses compared to average (c. +17–21 %), which is caused by higher thermal radiation. Phytomass estimates of the study are well in line with other similar studies.
Model uncertainties were assessed carefully in comparison to many other studies. The results imply that this kind of approach is needed as the model results varied a lot. Sample and sample size had a significant effect to results and therefore need to be addressed in future studies more carefully. The number of observations was high in this study compared to almost any other similar studies, but it has to be noted that the study designs differ. Clearly, there is a need of more extensive research on the uncertainties of phytomass estimates.