Modelling Ecosystem Response to Flooding: a Remote Sensing Approach

Modelling the impact of river regulation on large floodplain wetland ecosystems is essential for long-term management of these systems. Understanding the response to hydrological events is critical to developing conceptual models, while appropriate data is required to calibrate, test and validate models. In the Gwydir wetlands, NSW, Australia, satellite-derived normalised difference vegetation index (NDVI) has been used to assess flood response. This paper firstly compares the NASA Moderate Resolution Imaging Spectroradiometer (MODIS) with NOAA Advanced Very High Resolution Radiometer (AVHRR) NDVI for use in long-term temporal profile analysis and then explores the use of these data to identify the flood response from wetland ecosystems. Terra MODIS 16-day maximum value composite (MVC) NDVI data at a pixel resolution of 250 m(2) is stacked temporally for the period 29 September 2000 to 26 June 2005 for the study area. The NOAA AVHRR High Resolution Picture Transmission (HRPT) dataset (1 km(2)) was extracted for the period 21 March 1992 to 26 June 2005 and an MVC algorithm applied to correspond to the MODIS 16-day periods. Both MODIS and AVHRR data are used to determine the flood response using two sites, an internationally recognised wetland site and an adjacent native grassland site. Rainfall and inflow data were extracted from Bureau of Meteorology and NSW Department of Natural Resources databases respectively at daily resolution and the cumulative total for each 16-day antecedent period calculated. For both the wetland and grassland site, the AVHRR NDVI value was lower overall than for MODIS NDVI for the 109 composite periods analysed. A simple linear regression model explained over 80% of the variation between the AVHRR and MODIS data for both sites, although the intercept was higher for the wetland site. Mean NDVI was significantly higher in the wetland site compared to the grassland site but both sites can reach similar peak values following large rainfall or flood events. The NDVI time-series has significant auto-correlation at lags of 1 to 4 (64 days). Cross-correlation between NDVI, rainfall and inflow was generally significant at lags up to 80 days. Events were extracted from the AVHRR 16-day time-series where peak wetland NDVI exceeded 0.45. The mean NDVI for the event, initial NDVI, antecedent rainfall and inflows (80 days) were calculated for each event. Multiple regression analysis indicated that pre-event NDVI and antecedent inflows accounted for over 79% of the event NDVI for the wetland site. Neither rainfall nor inflows were significant for grasslands. This exploratory analysis of events indicates that the modelling of NDVI response is possible for small wetland sites using the AVHRR satellite data and can be compared to higher resolution MODIS data to provide a level of confidence when scaling from field sites to MODIS to AVHRR. Further investigation of modelling approaches using time-series may strengthen the analysis. More research is required to provide confidence in the sensitivity of models and data to small differences in event NDVI that separated highly productive wetlands from normal seasonal greening responses. This work must also be coupled with field monitoring to validate the NDVI results and determine the target event NDVI response required to maintain wetland vegetation communities in a healthy state. Despite some limitations and the need for further analysis, the results of this paper show promise for application to water management. These types of models can be used to estimate the water requirements to achieve a pre-determined NDVI response under a variety of antecedent conditions (as indicated by pre-event NDVI).