Evaluación de cobertura de suelos altoandinos mediante cloud computing de siete comunidades campesinas custodias de vicuñas de Ayacucho, periodo 2021
Cargando...
Código QR
Autores
Ochoa Pérez, Javier Ayrton
Correo de contacto
Resumen
Las comunidades andinas del centro y sur del Perú juegan un rol clave en la conservación de las vicuñas (Vicugna vicugna), especie protegida por la legislación nacional que depende de coberturas vegetales como el césped de puna y los bofedales (vegetación inundable) para su alimentación y acceso a fuentes de agua. Este estudio se enmarcó en siete comunidades de Ayacucho, ubicadas en la ecorregión puna caracterizado por escasa precipitación y bajas temperaturas, con énfasis en la clasificación de coberturas en ecosistemas altoandinos donde habitan vicuñas. Se evaluó el desempeño de 2 algoritmos de clasificación supervisada utilizando imágenes Sentinel-1 (S1) y Sentinel-2 (S2), la clasificación de tipos de cobertura utilizó puntos basados en tierra e imágenes, filtro para evitar la autocorrelación espacial, medidas de separabilidad como la distancia Jeffries-Matusita (JM), e incluyó Cloud Score+, procesador de evaluación de calidad de pixeles de S2, para una adecuada representación de clases de cobertura. El proceso de entrenamiento utilizó 73 variables que agrupan características del terreno (incluyendo vegetación, textura, topografía, nieve, agua, minerales, características de radar) divididos en 6 escenarios con la finalidad de evaluar la mejora de la precisión, la plataforma utilizada fue Google Earth Engine (GEE). Los resultados lograron identificar las coberturas presentes en áreas destinadas al manejo comunal de vicuñas total extensión 5.28 x104 ha distribuida en 9 clases pajonales (71.71%), no vegetación (11.69%), césped de puna (9.92%), matorrales arbustivos (4.37%), vegetación inundable (1.79%), cuerpos de agua (0.51%) mosaico agrícola (0.05%), cobertura arbórea (0.03%), nieve y glaciar (0.001%). El algoritmo Random Forest (RFo) alcanzó la mejor precisión general promedio (OA) de 93.6 % ± 0.009 y Kappa de 0,926 superando al algoritmo Support Vector Machine (SVM), que obtuvo una OA del 92.6 % ± 0.008 y Kappa de 0.915, para escenarios optimizados que incluyeron variables topográficas y textura.
Andean communities in central and southern Peru play a key role in the conservation of vicuñas (Vicugna vicugna), a species protected by national legislation that depends on vegetation cover such as puna grass and bofedales (flooded vegetation) for food and access to water sources. This study was carried out in seven communities of Ayacucho, located in the Puna ecoregion characterized by low rainfall and low temperatures, with an emphasis on land cover classification in high Andean ecosystems where vicuñas live. The performance of two supervised classification algorithms was evaluated using Sentinel-1 (S1) and Sentinel-2 (S2) images, the classification of land cover types using ground-based points and images, a filter to avoid spatial autocorrelation, separability measures such as the Jeffries Matusita (JM) distance, and Cloud Score+, a quality assessment processor to S2 pixels, was incorporated for an adequate representation of land cover classes. The training process used 73 variables that grouped terrain characteristics (including vegetation, texture, topography, snow, water, minerals, radar characteristics) divided into 6 scenarios to evaluate the improvement in accuracy, the platform used was Google Earth Engine (GEE). The results were able to identify the coverages present in areas destined for the communal management of vicuñas with a total area of 5.28 x104 hectares distributed in 9 classes: grasslands (71.71%), no vegetation (11.69%), puna grass (9.92%), shrubby scrub (4.37%), flooded vegetation (1.79%), water bodies (0.51%), croplands (0.05%), tree cover (0.03%), snow and glacier (0.001%). The Random Forest (RFo) algorithm achieved the best overall average accuracy (OA) of 93.6% ± 0.009 and Kappa of 0.926, outperforming the Support Vector Machine (SVM) algorithm, which obtained an OA of 92.6% ± 0.008 and Kappa of 0.915, for optimized scenarios that included topographic and texture variables.
Andean communities in central and southern Peru play a key role in the conservation of vicuñas (Vicugna vicugna), a species protected by national legislation that depends on vegetation cover such as puna grass and bofedales (flooded vegetation) for food and access to water sources. This study was carried out in seven communities of Ayacucho, located in the Puna ecoregion characterized by low rainfall and low temperatures, with an emphasis on land cover classification in high Andean ecosystems where vicuñas live. The performance of two supervised classification algorithms was evaluated using Sentinel-1 (S1) and Sentinel-2 (S2) images, the classification of land cover types using ground-based points and images, a filter to avoid spatial autocorrelation, separability measures such as the Jeffries Matusita (JM) distance, and Cloud Score+, a quality assessment processor to S2 pixels, was incorporated for an adequate representation of land cover classes. The training process used 73 variables that grouped terrain characteristics (including vegetation, texture, topography, snow, water, minerals, radar characteristics) divided into 6 scenarios to evaluate the improvement in accuracy, the platform used was Google Earth Engine (GEE). The results were able to identify the coverages present in areas destined for the communal management of vicuñas with a total area of 5.28 x104 hectares distributed in 9 classes: grasslands (71.71%), no vegetation (11.69%), puna grass (9.92%), shrubby scrub (4.37%), flooded vegetation (1.79%), water bodies (0.51%), croplands (0.05%), tree cover (0.03%), snow and glacier (0.001%). The Random Forest (RFo) algorithm achieved the best overall average accuracy (OA) of 93.6% ± 0.009 and Kappa of 0.926, outperforming the Support Vector Machine (SVM) algorithm, which obtained an OA of 92.6% ± 0.008 and Kappa of 0.915, for optimized scenarios that included topographic and texture variables.
Descripción
Universidad Nacional Agraria La Molina. Facultad de Ciencias. Departamento
Académico de Ingeniería Ambiental, Física y Meteorología
Palabras clave
Vicuña
Citación
Fecha
2026
Colecciones
Seleccionar año de consulta:
Licencia de uso

Excepto si se señala otra cosa, la licencia del ítem se describe como info:eu-repo/semantics/openAccess

