Construction of prediction model for water retention of forest ecosystem in alpine region based on vegetation spectral features

Abstract

The water retention service of the forest ecosystem has ecological functions such as adjusting the climate and maintaining the ecological water balance. The Qinghai-Tibet Plateau is an alpine region. Due to its high altitude and harsh environment, it is difficult to manually observe the water retention in the field, and it is impossible to better evaluate the water retention function. In order to better obtain the water retention in the alpine region, hyperspectral technology is introduced and applied to the acquisition of surface vegetation information, and the water retention in a specific area is obtained by constructing a model. In this study, the Bayi District of Nyingchi Prefecture was used as the research area. The main tree species in the study area are Picea likiangensis var. linzhiensis(PLVL), Quercus aquifolioides(QA), Pinus densata(PD) and Rhododendron nivale(RN). In actual situations, it is not easy to directly obtain water retention information, so a model can be found to quantitatively express the relationship between leaf spectrum and water retention. Then based on the leaf spectrum to invert the water retention. In order to study the quantitative relationship between different vegetation and water retention, each type of vegetation collects leaf samples and water retention data at 30 sampling points. Use ASD Fildsoec Handheld spectrometer to obtain hyperspectral data. Seven band indexes of red edge, green peak, NDVI, NDWI, EVI, WBI and NDPI were selected, and the relationship between vegetation band index and water conservation was fitted through many kinds of regression models. Comparing the fitting results, construct water retention prediction model. The interception of vegetation canopy, litter water holding capacity and soil water content are obtained through experiments. The sum of the three represents the water retention capacity of vegetation. The reflectance spectra of the four types of vegetation leaves all show similar regularities, and the difference in the visible light band is not obvious. The near-infrared to mid-infrared bands show four distinct water absorption bands, with the highest reflectivity in the red to near-infrared bands (700 nm-1400 nm). The reflectance of the four types of vegetation varies across different spectral bands, with the reflectance levels exhibiting the characteristic order of QA > PD > PLVL ≈ RN. Comparing the fitting results of different regression models with seven waveband parameters, the R² of the four types of vegetation are higher in the regression models of EVI and NDPI, and reach a significant level. According to the regression model corresponding to each kind of vegetation, the water retention prediction model is composed, and the simulation accuracy is tested by R² and RMSE. The overall simulation accuracy R² is greater than 0.7 and the RMSE is basically less than 10 t·hm⁻², indicating that the forecasting model has a good forecasting effect and the model can effectively estimate the water retention of the forest ecosystem.