Studying Land Cover and Water Availability in Southern Angola using Landsat and MODIS
Mantas, Vasco; Pereira, A.J.S.C.
IMAR, University of Coimbra, PORTUGAL

1. Introduction

The last decade of the 20th century was a period of unprecedented changes in the landscape of Angola. The end of the civil war (1975 - 2002) and the persistency of double-digit economic growth rates [1] led to a rapid change in land cover trends throughout the country.
Nonetheless, and because of the climate of political, social and economic instability that dominated the country throughout the second half of the last century, scientific research was brought to a halt up until now, limiting the collection of field data or the development of cartography.
The Kunene and Cuvelai basins are amongst the most important ecosystems of Southern Angola. Meaningful for both ecological and economic reasons, there is a growing interest in the study of these systems. The focus thus far has been directed at the Kunene, mostly for strategic reasons.
However some of the most interesting features of the region are the countless lagoons that are formed during the wet season. These are fundamental for agriculture, defining the traditional transhumance routes, travelled every year for many generations [1]. The lagoons are also structuring elements of the landscape, reducing the available space for human settlements and crop production. The availability of water and the distribution of the lagoons are thus fundamental aspects of the landscape.
In this study we set out to understand water availability and land cover changes in a sector of the Kunene and Cuvelai basins in the first decade of the 21st century.

2. Materials and Methods

In this study we analyzed a long time series of Landsat-7 ETM+ and Landsat-5 TM imagery to understand the seasonal cycles of the water level in the last decade and the evolution of land cover in the adjoining areas. Two batches of Landsat images were acquired. The first included all cloud free images (clouds covering under 20% of the study area) from January 2000 to December 2011. The second group included images acquired in April from 1989 to 1999. All images were converted to at-sensor reflectance using a customized tool developed on ERDAS Imagine ModelMaker using standard equations and calibration values [2]. Using high-resolution satellite imagery a batch of 10 images of the entire collection were used to train a Regression Tree Classifier [3] using Weka. The resulting rules were translated into an ArcGIS 10.1 model and the entire collection classified for the following classes: Water, Cropland and Natural Vegetation. It is important to notice that the local human settlements are built with traditional construction methods and principles. Each small group of huts is surrounded by a fenced area where crops are grown. The Cropland was considered to include the entire compound for simplification purposes. Since the dimension of the area reserved for the huts is very similar in all cases (and limited), the error is minimal.
The resulting classification maps were then compared with rainfall estimates (TOVAS 3B42) and modeled data (Soil Moisture from GLDAS).
Because of the limitations inherent to the use of Landsat data, a continuous field water-fraction model was built using MODIS Surface reflectance data (MOD09GA) with a spatial resolution of 500 meters. The model was constructed using the classified Landsat images as the training dataset. The training data was then exported to Weka where, through the use of a Regression Tree Model (M5P), the continuous water fraction model was constructed. The MODIS pixels were clustered by means of unsupervised classification prior to the development of the model. This was made to simplify the classification process and at the same time to guarantee that land cover changes, inducing the mobility of pixels between clusters would not prevent the correct water fraction estimation. Given the large number of cloud-free Landsat scenes in 2008 and 2009, this process was attempted for these years alone to test the viability of the concept.
But to model the surface water area of the lagoons, the MODIS water fraction model alone would not suffice. Yet, and given the extreme flatness of the ground, existing topographic maps or even SRTM data is unsuitable to accomplish this goal. Instead, a map of the water presence probability, calculated from 45 Landsat images of the wet season was used. The MODIS water-fraction model provided the area covered by water, which would then be distributed by the 30 m pixels that were more likely to be flooded given that percentage. The models were than compared against real Landsat data for the same date.

3. Results and Discussion

The project enabled the creation of a wealth of cartographic data that can now be used in the preparation of the field sampling campaigns, the analysis of which is discussed in detail in the presentation. Land Cover changes, including the expansion of the agricultural area (especially in certain areas where it is associated and simultaneously limited by the lagoons) are also addressed.
Amongst the findings is the confirmation of an increase in the total water area within the study area for equivalent periods in recent years. This is in line with indications provided by both the GLDAS soil moisture models and TRMM estimates, both suggesting an increase in water availability. For April acquisitions, for instance, the trend is clearly positive (y = 1.7021x - 3390.5, R2 = 0.4949) with a succession of years with a wider than average distribution of water pixels, especially in the second half of the last decade.
It was also possible to establish a strong correlation between the Water Area and GLDAS Soil Moisture (Level 4) values for 2008 and 2009 (abundant data available for the entire rainfall cycle). The MODIS water fraction model was also proved to be robust (average accuracy of 92%) enabling the modeling of the lagoons based on the Landsat data as aforementioned. The introduction of same-season Landsat scenes to adjust the water probability distribution reinforced the accuracy of the modeled lagoon areas.
The presentation also addresses the directions to be taken in subsequent projects, in order to understand in greater detail the water and land cover dynamics of this interesting but complex ecosystem.


1. Ministério da Agricultura e do Desenvolvimento Rural de Angola (2003) Diagnóstico Rural Rápido da Zona Agro-ecológica de Baixa Pluviosidade, Provincia do Cunene, 24 p.
2. Chander, G., Markham, B. L., Helder, D. L. (2009) Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors, Remote Sensing of Environment, 113, pp. 893 - 903.
3. Xian, G., Homer, C. (2010) Updating the 2001 National Land Cover Database Impervious Surface Products to 2006 using Landsat Imagery Change Detection Methods, Remote Sensing of Environment, 114, pp. 1676 - 1686.