A Genetic Programming Approach to Estimate Vegetation Cover in the Context of Soil Erosion Assessment

This work describes a genetic programming (GP) approach that creates vegetation indices (VI’s) to automatically detect the sum of healthy, dry, and dead vegetation. Nowadays, it is acknowledged that VI’s are the most popular method for extracting vegetation information from satellite imagery. In particular, erosion models like the “Revised Universal Soil Loss Equation” (RUSLE) can use VI’s as input to measure the effects of the RUSLE soil cover factor (C). However, the results are generally incomplete, because most indices recognize only healthy vegetation. The aim of this study is to devise a novel approach for designing new VI’s that are better correlated with C, using field and satellite information. Our approach consists on stating the problem in terms of optimization through GP learning, building novel indices by iteratively recombining a set of numerical operators and spectral channels until the best composite operator is found. Experimental results illustrate the efficiency and reliability of our approach in contrast with traditional indices like those of the NDVI and SAVI family. This study provides evidence that similar problems related to soil erosion assessment could be analyzed with our proposed methodology. Introduction Soil erosion is a complex phenomenon that detaches and transports soil material through the action of an erosive agent. Soil erosion by flowing water on slopes is an important land degradation problem at a global scale, because it is PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Ap r i l 2011 363 Cesar Puente is with the Departamento de Geociencias Ambientales, EvoVisión Project,Centro de Investigación Cientifica y de Educación Superior de Ensenada (CICESE), Carretera Ensenada-Tijuana 3918 Zona Playitas, 22860, Ensenada, Baja California, México ([email protected]). Gustavo Olague is with EvoVisión Project, Centro de Investigación Cientifica y de Educación Superior de Ensenada (CICESE). Stephen V. Smith and Alejandro Hinojosa-Corona are with the Departamento de Geología, Centro de Investigación Cientifica y de Educación Superior de Ensenada (CICESE). Stephen H. Bullock is with the Departamento de Biología de la Conservación, Centro de Investigación Cientifica y de Educación Superior de Ensenada (CICESE). Miguel A. González-Botello is with the Maestría en manejo de ecosistemas de zonas áridas, Facultad de Ciencias, Universidad Autónoma de Baja Californa (UABC). Photogrammetric Engineering & Remote Sensing Vol. 77, No. 4, April 2011, pp. 363–376. 0099-1112/11/7704–0363/$3.00/0 © 2011 American Society for Photogrammetry and Remote Sensing A Genetic Programming Approach to Estimate Vegetation Cover in the Context of Soil Erosion Assessment Cesar Puente, Gustavo Olague, Stephen V. Smith, Stephen H. Bullock, Alejandro Hinojosa-Corona, and Miguel A. González-Botello strongly influenced by human activities such as agriculture and generates major environmental impacts and high economic costs. Most estimates of soil erosion, as undertaken by agricultural scientists, are at field scales (Brady and Weil, 2008). The aim of erosion research at regional scales is a general evaluation of the landscape and its susceptibility to soil erosion, taking into account only the main factors influencing the process. Therefore, the erosion assessment at regional scale is usually based on empirical models or expert evaluation. The most widely applied models are the Universal Soil Loss Equation (USLE) (Wischmeier and Smith, 1978), and its revision, the Revised Universal Soil Loss Equation (RUSLE) (Renard et al., 1997). USLE and RUSLE are statistically-based water erosion models related to six erosional factors: A R * K * L * S * C * P, (1) where R and K set the dimensions of A as the average of soil loss in Mg ha 1 yr 1. R is the rainfall-runoff factor and is measured as the product of total storm energy, E, and the maximum 30 minute intensity, I30, for all storms in a year, so its units are described in terms of MJ mm ha 1 h 1 yr 1. On the other hand, K represents the influence of soil properties related to soil texture and structure on soil loss during storm events; the units of K are Mg h MJ 1 mm 1. The remaining factors are dimensionless and serve to scale erosion relative to standard experimental conditions, which are described within the USLE and RUSLE manuals cited above. These scaling values range from 0 meaning no erosion, to numbers greater than 1, where erosion is more rapid than the experimental conditions. In this way, L represents the slope length factor, S is the slope steepness factor, C provides the ground cover factor, and P describes the conservation support practice factor. From the standpoint of soil conservation planning, the C factor is one of the most important parameters of RUSLE, because it measures the combined effect of all interrelated cover and management variables. Vegetation cover acts as a protective layer between the atmospheric elements and soil. For example, live or dead leaves and stems absorb most of the energy of raindrops and surface water to decrease the volume of rain reaching the soil surface (Asis and Omasa,