Interrelationships among landscapes, NDVI, and stream water quality in the United States Central Plains.

摘要

Landscape pattern metrics (LPMs) and the normalized difference vegetation index (NDVI) were analyzed for regional watershed monitoring in Kansas, Nebraska, and Missouri. Principal components analysis (PCA) was performed on 27 landscape pattern metrics derived from a Kansas land cover data base at three spatial resolutions: 30 m, 100 m, and 1 km. The PCA reduced the number of variables to five components that explained 81%–89% of the variation in the original datasets. The components were interpreted as landscape texture, patch shape/size, cropland/grassland class-specific metrics, patch interspersion, and a nearest neighbor attribute. The same interpreted components appeared at each resolution level, supporting the use of the same subset of metrics for landscape monitoring in the region at different resolutions. The most important metrics included the Modified Simpson's Diversity index, the area-weighted mean patch fractal dimension, the interspersion and juxtaposition index, and the largest patch index for grasslands.; NDVI and vegetation phenological metrics (VPMs) were related to water quality Two hundred ninety stream sites in Nebraska, Kansas, and Missouri were sampled between May and September in 1994 or 1995 for conductivity, turbidity, total phosphorus, nitrate-nitrite nitrogen, the Index of Biotic Integrity, and a habitat index Statistically significant relationships were found between NDVI or VPMs and the stream data (r values up to 0.8, α = 0.05). NDVI and/or VPMs were more highly correlated to stream conditions than were land use/land cover proportions. Early spring NDVI or VPMs associated with the date of onset of greenness were most frequently correlated with the stream data, suggesting the possibility of estimating summer stream conditions with spring NDVI data.; Relationships between stream conditions and the LPMs were generally not statistically significant. Several problems with using LPMs were noted: small watersheds having only one or two patches; collinearity with LULC data; and inconsistent results due to the importance of factors other than landscape pattern in determining water quality. A comparison of regression models using a combination of LPMs and LULC versus NDVI metrics alone showed that NDVI had greater promise for monitoring landscapes for regional scale water quality monitoring in this part of the country.