Use of Geographically Weighted Regression Model for Exploring Spatial Patterns and Local Factors Behind NDVI-Precipitation Correlation. Kang, L., Di, L., Deng, M., Shao, Y., Yu, G., & Shrestha, R. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 7(11):4530–4538, November, 2014.
doi  abstract   bibtex   
Normalized Difference Vegetation Index (NDVI)-precipitation correlation has long been studied. In previous studies, the correlation was usually based on global regression model, which assumed such correlation be constant across the space. However, NDVI-precipitation correlation is spatially dependent and affected by local factors (e.g., soil background). In this paper, geographically weighted regression model is utilized to analyze the NDVI-precipitation correlation on three land use types (i.e., 1) grassland, 2) fallow/idle land, and 3) winter wheat land) within U.S. central great plain area. Results suggest that geographically weighted regression model has better performances than global regression models. Specifically, higher average R$^{\textrm{2}}$ (0.81) and lower proportion (9%) of residuals with spatial autocorrelation has been achieved under geographically weighted regression in comparison with lower average R$^{\textrm{2}}$ (0.68) and higher proportion (38%) of residual with spatial autocorrelation under global regression models. In addition, the spatially dependent correlation between NDVI and precipitation has been revealed with geographically weighted regression model. From the north to south, the increasing unit rate of NDVI's change with precipitation has been found through spatially varying regression slopes. Moreover, local factors affecting NDVI-precipitation correlation, such as soil permeability and thickness, have been identified through analyzing the local goodness of fitting under geographically weighted regression model. In summary, unveiled spatial patterns of NDVI-precipitation correlation provide another perspective for studying correlations between NDVI and climatic factors. This work should also be helpful to better understand crop responses to precipitation in agricultural management.
@article{kang_use_2014,
	title = {Use of {Geographically} {Weighted} {Regression} {Model} for {Exploring} {Spatial} {Patterns} and {Local} {Factors} {Behind} {NDVI}-{Precipitation} {Correlation}},
	volume = {7},
	issn = {1939-1404},
	doi = {10.1109/JSTARS.2014.2361128},
	abstract = {Normalized Difference Vegetation Index (NDVI)-precipitation correlation has long been studied. In previous studies, the correlation was usually based on global regression model, which assumed such correlation be constant across the space. However, NDVI-precipitation correlation is spatially dependent and affected by local factors (e.g., soil background). In this paper, geographically weighted regression model is utilized to analyze the NDVI-precipitation correlation on three land use types (i.e., 1) grassland, 2) fallow/idle land, and 3) winter wheat land) within U.S. central great plain area. Results suggest that geographically weighted regression model has better performances than global regression models. Specifically, higher average R$^{\textrm{2}}$ (0.81) and lower proportion (9\%) of residuals with spatial autocorrelation has been achieved under geographically weighted regression in comparison with lower average R$^{\textrm{2}}$ (0.68) and higher proportion (38\%) of residual with spatial autocorrelation under global regression models. In addition, the spatially dependent correlation between NDVI and precipitation has been revealed with geographically weighted regression model. From the north to south, the increasing unit rate of NDVI's change with precipitation has been found through spatially varying regression slopes. Moreover, local factors affecting NDVI-precipitation correlation, such as soil permeability and thickness, have been identified through analyzing the local goodness of fitting under geographically weighted regression model. In summary, unveiled spatial patterns of NDVI-precipitation correlation provide another perspective for studying correlations between NDVI and climatic factors. This work should also be helpful to better understand crop responses to precipitation in agricultural management.},
	number = {11},
	journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
	author = {Kang, L. and Di, L. and Deng, M. and Shao, Y. and Yu, G. and Shrestha, R.},
	month = nov,
	year = {2014},
	keywords = {agricultural management, Analytical models, atmospheric precipitation, climatic factor, climatology, Correlation, Estimation, fallow-idle land, Geographically weighted regression, geographically weighted regression model, global regression model, grassland, Indexes, land use, land use type, lower residual proportion, Meteorology, NDVI, NDVI change increasing unit rate, NDVI-precipitation correlation local factor, NDVI-precipitation correlation spatial pattern, normalized difference index-precipitation correlation, permeability, precipitation, precipitation crop response, soil, Soil, soil background, soil permeability, soil thickness, spatial autocorrelation, spatial nonstationarity, spatially varying regression slope, Standards, US central great plain area, vegetation, winter wheat land},
	pages = {4530--4538},
	file = {IEEE Xplore Abstract Record:/Volumes/mini-disk1/Google Drive/_lib/zotero/storage/EF84Y2UT/6930718.html:text/html;IEEE Xplore Full Text PDF:/Volumes/mini-disk1/Google Drive/_lib/zotero/storage/L74FCHEP/Kang et al. - 2014 - Use of Geographically Weighted Regression Model fo.pdf:application/pdf}
}

Downloads: 0