Validación de modelos predictivos de cambio de cubierta y uso del suelo en la península de Baja California, México

• Autores: Laura Chang-Martínez, Fernando A. Rosete Vergés, Juan Felipe Charre Medellin, Jean François Mas
• Localización: Investigaciones geográficas, ISSN 0188-4611, ISSN-e 2448-7279, Nº. 102, 2020
• Idioma: español
• Títulos paralelos:
  • Validation of predictive land use models in the peninsula of Baja California, Mexico
• Enlaces
  • Texto completo
• Resumen
  • español

    Los modelos predictivos de cambio de cubierta y uso del suelo (CCUS) son herramientas que permiten identificar cantidad o áreas susceptibles a los cambios, y prevenir condiciones de degradación ambiental. Existen diversos enfoques para realizar las simulaciones de CCUS, los modelos predictivos a evaluar en esta investigación utilizan un enfoque basado en patrones, que echan mano de datos de percepción remota, censos poblacionales, análisis estadísticos y conocimiento experto, lo que permite generar la parametrización de las transiciones de una categoría a otra y así generar mapas de cambio. A través de la validación de los modelos se pretende evaluar la exactitud de las predicciones, permitiendo identificar las mejores metodologías para generar modelos predictivos confiables. Como resultado de esto, el objetivo de esta investigación es conocer la capacidad predictiva de tres modelos de CCUS en la península de Baja California, México, generados en 2008. A través del método de actualización cartográfica, se otuvieron tres mapas de cubiertas y usos del suelo para de 2018. Esto se realizó con herramientas de percepción remota, sistemas de información geográficas, uso de software de análisis estadísticos (R) y detección de cambios (DINAMICA-EGO). Una vez obtenidos en 2018 los mapas de CCUS del año 2018, fue evaluada la fiabilidad de cada mapa. Y, finalmente, se evaluaron los modelos predictivos realizados. Los mapas de CCUS de 2018 presentaron una fiabilidad superior a 96% en las tres localidades. Las predicciones de los modelos de CCUS realizadas en el 2008 fueron muy cercanas a las observadas en el 2018 en dos de ellos, ya que en la localidad de Santo Domingo la asertividad fue de 77% y en San Quintín del 86%, mientras que en Tijuana fue solamente del 35%.  La metodología empleada es una propuesta que ayuda a conocer el grado de certidumbre de los modelos predictivos de CCUS y la generación de cartografía actualizada.

  • English

    Studying the CCUS helps to describe relationships between man and nature through a spatial interface, such as satellite images and Geographic Information Systems (GIS). This allows the monitoring of regional and global changes by looking at the distribution patterns of landscape covers and the effects on the available resources. This also makes it possible to use tools such as the Models of Change of Cover and Land Use (CCUS, in Spanish), which identify the quantity and/or areas susceptible to change, in addition to preventing environmental degradation conditions. These models mainly identify patterns of change by using remote perception data, population censuses, statistical analyses, and expert knowledge, allowing to generate parameterization of transitions between categories. Once the prospective model has been developed, the simulation or accuracy of the prediction can be evaluated, ideally through statistical methods produced to determine the accuracy of the projection of a model. Through the validation of CCUS models, we can establish the certainty of the changes that are foreseen. Therefore, we consider it important to find new methods to evaluate predictive CCUS models, and in this study, evaluates the accuracy of the predictions considering the present information to define the certainty of models that have already met the prediction time. Therefore, the objective of this work is to determine the predictive capacity of three CCUS models in the Baja California peninsula, Mexico, generated in 2008. The main input was the predictive models calculated for the period 1978-2003, produced from satellite images. Subsequently, through the mapping update method, three maps of land covers and uses were obtained for 2018, used to generate change maps for the period 2003-2018 and thus evaluate the assertiveness of the change surfaces projected by the predictive models. This was done with remote sensing tools, geographic information systems, statistical analysis software (R), and change detection (DINAMICA-EGO). The mapping update procedure required a total of four Landsat-8 scenes with a spatial resolution of 30; this method produces segments corresponding to units of the landscape units and avoids producing isolated pixels. Changes are detected through these segments based on atypical spectral responses compared to other objects in the image. This procedure combines digital image classification and visual interpretation processing to produce maps for different times based on existing mapping enhancement processes. Once the 2018 CCUS maps were obtained, the reliability of each 2018 map was evaluated, as well as the reliability of the user and producer of each map. Finally, the predictive models elaborated in 2008 were evaluated by observing the changes between the 2003 map and the updated 2018 map for each study site. Subsequently, matrices of cover changes between years 2003-2018 were elaborated in DINAMICA EGO, allowing us to determine the area (hectares) of change by category for each site and the total loss by category between years. From this, we estimated the area of xeric shrubland that changed, corresponding to the dependent variable in all predictive models. Finally, the current data of change for each model were compared with the prediction for each study site. The 2018 CCUS maps showed a reliability figure above 96% in the three locations. The predictions from CCUS models in 2008 were very close to those observed in 2018 in two cases; Santo Domingo showed an assertiveness of 77% (26 more hectares lost vs. prediction) and San Quintín, 86% (17 less hectares lost vs. prediction), while Tijuana showed an assertiveness of only 35% (13 less hectares lost vs. prediction). The methodology used is a proposal that helps to know the actual degree of certainty of the CCUS predictive models and the generation of updated cartography. This evaluation not only allows determining assertiveness but, under a more thorough and rigorous investigation, it will make it possible to identify the direct and indirect factors that led to change, thus contributing expert knowledge for future predictions.