The effect of spatial resolution in remote sensing estimates of total evaporation in the uMgeni catchment.

Abstract

The estimation of total evaporation plays a vital role in water resources monitoring and management, especially in water-limited environments. In South Africa, the increasing water demand, due to population growth and economic development, threatens the long-term water supply. This, therefore, underscores the need to account for water by different consumers, for well-informed management, allocation and future planning. Currently, there are different methods (i.e. ground-based and remote sensing-based methods), which have been developed and implemented to quantify total evaporation at different spatial and temporal scales. However, previous studies have shown that ground-based methods are inadequate for understanding the spatial variations of total evaporation, within a heterogeneous landscape; they only represent a small area, when compared to remotely sensed methods. The advent of remote sensing therefore provides an invaluable opportunity for the spatial characterization of total evaporation at different spatial scales. This study is primarily aimed at estimating variations of total evaporation across a heterogeneous catchment in KwaZulu-Natal, South Africa, using remote sensing data. The first part provides an overview of total evaporation, its importance within the water balance and consequently in the management of water resources. It also covers various methods developed to estimate total evaporation, highlighting their applications, limitations, and finally, the need for further research. Secondly, the study determines the effect of sensor spatial resolution in estimating variations of total evaporation within a heterogeneous uMngeni Catchment. Total evaporation estimates were derived, using multispectral 30 m Landsat 8 and 1000 m MODIS, based on the Surface Energy Balance (SEBS) model. The results have shown that different sensors, with varying spatial resolutions, have different abilities in representing variations of total evaporation at catchment scale. It was found that Landsat-based estimates were significantly different (p < 0.05) from MODIS. The study finally estimates spatial variations of total evaporation from Landsat 8 and MODIS datasets for the uMngeni Catchment. It was found that the Landsat 8 dataset has greater potential for the detection of spatial variations of total evaporation, when compared to the

MODIS dataset. For instance, MODIS-based daily total evaporation estimates did not show any significant difference across different land cover types (One way ANOVA; F1.924 = 1.412, p= 0.186), when compared to the 30 m Landsat 8, which yielded significantly different estimates between different land cover types (One way ANOVA; F1.993= 5.185, p < 0.001). The validation results further indicate that Landsat-based estimates were more comparable to ground-based eddy covariance measurements (R2 = 0.72, with a RMSE of 32.34 mm per month (30.30% of the mean)). In contrast, MODIS performed poorly (R2 = 0.44), with a RMSE of 93.63 mm per month (87.74% of the mean). In addition, land cover-based estimates have shown that, not only does the land cover type have an effect on total evaporation, but also the land cover characteristics, such as areal extent and patchiness. Overall, findings from this study underscore the importance of the sensor type, especially spatial resolution, and land cover type characteristics, such as areal extent and patchiness, in accurately and reliably estimating total evaporation at a catchment scale. It is also evident from the study that the spatial and temporal variations in SEBS inputs (e.g., LAI, NDVI and FVC) and energy fluxes (e.g., Rn) calculated by SEBS for the two sensors can affect the spatial and temporal variations in total evaporation estimates. For instance, spatial variations in total evaporation reflected similar spatial variations in Rn. Areas with high NDVI, FVC and LAI (which denotes dense vegetation cover) tend to have higher total evaporation estimates, compared to areas with lower vegetation cover. In addition, the MODIS sensor at 1000 m spatial resolution showed lower estimates of SEBS inputs with less variability across the catchment. This resulted in lower total evaporation estimates, with less variability, compared to the 30 m Landsat 8. In addition, with regard to inputs derived from remote sensing, it was found that the spatial variations in total evaporation are not determined by individual variables (e.g., LST), but are influenced by a combination of many biophysical variables, such as LAI, FVC and NDVI. These findings lay a foundation for a better approach to estimate total evaporation using remote sensing for use in the management and allocation of water.