Environmental change detection and modeling using multitemporal and multiscore data.

摘要

Population growth and economic development has been stressing environmental carrying capacity. Ecoregion, net primary productivity (NPP), and cereal yield in cultivated land are fundamental parameters in assessing the natural system in supporting and sustaining standard of living and healthy populations. I have developed protocols to classify ecoregions, simulate NPP, and estimate cereal yield based on multisource and time series data from 1983 to 1994 in the study areas of mainland of China and the United States.; The data used in this study include monthly NOAA AVHRR data, monthly temperature and precipitation data, terrain data, and data of soil type, texture, depth, and slope. To assure the quality of the multisource data, a heuristic algorithm and a decomposition algorithm are developed to detect and correct spatial outliers and time series outliers.; Environmental problems are best addressed in the context of geographic areas defined by natural features. Kohonen's Self-Organizing Algorithm is applied to integrate multiple environmental factors and time series data to classify the study areas into 25 ecoregions. Each ecoregion has its unique natural conditions reflecting its potential in supporting and sustaining standard of living.; Water supply is a principal factor that controls primary production. Two third of the ecoregions in China are deficient in water resource, which constrains food supply. The trends of temperature change in the sampled points demonstrate different patterns that significantly correlate to latitudes. The primary productivity increases significantly in response to the increase of temperature in high latitude areas.; The CASA model is adapted to integrate the information of precipitation, soil type, and irrigation system in agricultural areas for estimating NPP. An NPP database is utilized to adjust the parameters by using the technique of least squares.; The protocol for cereal yield estimation in cultivated land based on the simulated NPP consists of three steps: (1) separate cropland from other classes using supervised neural network; (2) identify cropping types by comparing the time series curves of NDVI and the data of seeding, replanting, growing, and harvesting seasons; (3) estimate the yield using NPP, and harvest indices. Some experimental results are demonstrated.