Fragment size, vegetation structure and physical environment control grassland functioning: a test based on artificial neural networks
Version of Record online: 11 DEC 2012
© 2012 International Association for Vegetation Science
Applied Vegetation Science
Volume 16, Issue 3, pages 426–437, July 2013
How to Cite
Herrera, L. P., Texeira, M., Paruelo, J.M. (2013), Fragment size, vegetation structure and physical environment control grassland functioning: a test based on artificial neural networks. Applied Vegetation Science, 16: 426–437. doi: 10.1111/avsc.12009
- Issue online: 4 JUN 2013
- Version of Record online: 11 DEC 2012
- Manuscript Accepted: 23 SEP 2012
- Manuscript Received: 20 JAN 2012
- Inter-American Institute for Global Change Research. Grant Number: CRN II 2031
- US National Science Foundation. Grant Number: GEO-0452325
- FONCYT. Grant Number: PICT 32399
- UBACYT. Grant Number: G006
- Enhanced vegetation index;
- Landscape structure;
- MODIS data;
- Neural networks;
- Paspalum quadrifarium ;
- Tall-tussock grassland
How do fragment-level characteristics affect remnant grassland functioning in a highly transformed landscape? Are artificial neural networks (ANNs) a better statistical tool to model variations in grassland functioning compared to linear regression models (LRMs)?
Tandilia Range, Southern Pampa, Buenos Aires Province, Argentina.
We characterized the dynamics of the vegetation functioning in 60 remnant grasslands using enhanced vegetation index (EVI) data provided by MODIS/Terra images from July 2000 to June 2005. First, we performed a principal components analysis (PCA) on the fragment mean monthly values of EVI in order to obtain synthetic measures (i.e. the PCA axes) of grassland functioning. Grassland fragments were also characterized by size, vegetation structure (abundance of the tall-tussock grass Paspalum quadrifarium) and physical environment (soil type – abundance of litholitic soils – elevation, aspect and slope). The relationship between grassland functioning and these explanatory variables was explored using linear regression models (LRMs) and artificial neural networks (ANNs).
The first and second PCA axes were related to the annual integral of EVI (EVI-I) and EVI seasonality (EVI-S), respectively; these explained jointly ca. 80% of total variability in mean EVI values. ANNs captured better than regression models the relationships among the proposed controls and the spatial variability of grassland functioning in Southern Pampa. Results showed that EVI-I variability was related to all independent variables except aspect. While fragment size, litholitic soils and slope were negatively related to EVI-I, the abundance of P. quadrifarium had a positive effect on the spectral index. Grasslands with high seasonality were large and had high slope and aspect, low abundance of P. quadrifarium and increased abundance of litholitic soils.
Our results showed that grassland functioning in Southern Pampa, as estimated by EVI, depends on fragment size, vegetation structure and physical factors (soil type, aspect and slope). Paspalum quadrifarium may have an important functional role in this grassland system.