Assessment of NDVI and NDWI spectral indices using MODIS time series analysis and development of a new spectral index based on MODIS shortwave infrared bands * Alicia Palacios-Orueta a,b , Shruti Khanna b , Javier Litago c,b , Michael L. Whiting b and Susan L. Ustin b a E.T.S.I. Montes. Universidad Politécnica de Madrid, email: alicia.palacios@upm.es b Calspace, University of California, Davis c E.T.S.I. Agrónomos. Universidad Politécnica de Madrid ABSTRACT We put forward a new spectral index, Shortwave Angle Normalized Index (SANI), based on the NIR and SWIR MODIS bands. The new index parameterizes the general shape of this part of the spectrum by measuring the angle at SWIR1 and the normalized index between NIR and SWIR2. Preliminary results show that it performs well in tracking moisture and discriminating between soil, vegetation and dry vegetation. We use Time Series Analysis to explore the temporal evolution of NDVI, NDWI and SANI and climatic data for the years 2000 to 2005. Our analyses show that SANI is synchronized with precipitation in grasslands but not in irrigated cropland where irrigation is a major source of moisture. NDVI does not follow precipitation closely in either of the two regions. SANI also shows an overall negative trend, which corresponds to the overall positive trend in precipitation levels from 2000 to 2005. Thus, this index seems to be a powerful tool for uncovering subtle sources of variability, inter- annual trends in environmental variables and dynamic relationships between soil and plant variables. 1 INTRODUCTION In the last few decades, climate change has begun to affect the responses of natural and cultivated vegetation. Although ecosystem response to the climatic trends can be subtle and difficult to detect, changes can have a strong impact on ecosystem health at medium to longer time periods. It is crucial to assess such impacts in order to design effective management strategies. Long-term ecosystem variability can change abruptly due to land use or disturbance or slowly and cumulatively, as in the case of climate change. On the other hand, non-permanent inter- annual variability occurs with annual climate anomalies [1] or land management, e.g., crop rotation and it is important to be able to distinguish between small but significant trends versus natural inter-annual variability. Time series data acquired through remote sensing instruments can provide information about ecosystem dynamics at medium (decadal) time scales and at a frequency that makes it possible to study both abrupt and gradual change in response to short and longer term variability. Such information will allow assessment of multivariate relationships between climate and remote sensing variables as well as among those variables. Spectral indices are one of the most common techniques used for analyzing remote sensing data [2]. Indices are based on combinations of a small number of bands that enhance specific spectral properties. In vegetated environments spectral indices focus on emphasizing vegetation characteristics. When working with multispectral data only a few bands can be used, therefore indices are usually indicators of general characteristics such as plant cover, greenness, or amount of exposed soil. The most universal index, NDVI (Normalized Difference Vegetation Index) has been used extensively to monitor ecosystems; in both the spatial and temporal domains [3] because it is proven to be a good indicator of ecosystem parameters like biomass, LAI and FPAR [4] among others. While NDVI derived indices are based on plant pigment absorption there are other indices which try to discriminate vegetation parameters such as water content and the amount of non-photosynthetic vegetation. The NDWI, (Normalized Difference Water Index) is a good indicator of soil and vegetation water content [5]. There are also several indices that characterize non- photosynthetic vegetation, (NPV) [6]. Plants function within a constrained range of biochemical variation. For example, a healthy plant will have high pigment content and also high water and nitrogen contents. Thus indices used to measure these different characteristics are frequently physiologically correlated. Most indices are also constructed using red (R) and/or near-infrared (NIR) bands, which means that they are mathematically correlated. Presented at the 1 st International Conference of Remote Sensing and Geoinformation Processing, Trier (Germany), Sep 7-9 2005