Resumen

LOPEZ ARCADIA, Carlos Alberto  y  BONILLA MOHENO, Martha. Importance of probability sampling estimates for analyzing regional forest dynamics: an evaluation of Global Forest Change data in eastern Mexico. Invest. Geog [online]. 2024, n.113, e60790.  Epub 30-Jul-2024. ISSN 2448-7279.  https://doi.org/10.14350/rig.60790.

The Global Forest Change (GFC) database has facilitated access to soil dynamics analysis and monitoring at different scales. However, these data may contain spatialtemporal inaccuracies caused by landscape characteristics, agricultural management practices, and the spatial scale of the assessment. To reduce these inaccuracies, it is recommended to evaluate each case study using probabilistic sampling estimates. The number of studies that calculate forest cover loss in Mexico using the GFC database has increased; however, few studies have used sampling estimates for this purpose. This study assessed probabilistic sampling estimates in four coffee regions of eastern Mexico with complex climatic and topographic characteristics and numerous smallholders. We used the GFC to carry out probabilistic sampling, grouping the annual coverage loss pixels into three categories: No Loss (NP), Possible Loss (PP), and Loss (P). For each region, we randomly generated 200 sampling points for P and 100 for PP and NP. Using time series of maximum and minimum NDVI and EVI values and high-resolution images from Google Earth, we carried out a visual verification to identify the year with a loss of forest cover at each sampling point. We then used three procedures to calculate the loss: an estimate based on the GFC map, a sampling estimate using the P+PP categories, and a sampling estimate using the three categories (P+PP+NP). Additionally, we validated the temporal accuracy of the GFC database by comparing the year in which the loss was reported with the year in which the loss was observed in the assessment. We found that forest cover loss was correctly detected by the database more than 90 % of the time in all regions. However, the loss observed in PP and NP and not detected by the database was high (53-65 % and 23-26 %, respectively). We also found that the loss calculated in the sampling estimates was four (P+PP) to 70 (P+PP+NP) times greater than map-based estimates. Furthermore, we found a relatively low temporal accuracy (59.5-77.5 %), showing a tendency to report the loss of forest cover one or more years after it occurred. This indicates that the GFC database can significantly underestimate the loss of forest cover by conducting regional assessments on heterogeneous landscapes managed by smallholders. This study highlights the importance of conducting probabilistic sampling assessments to reduce spatio-temporal inaccuracies that could lead to erroneous inferences about patterns of forest cover loss.

Palabras llave : deforestation; sample estimation; complex landscapes; forest degradation; regional assessment.

        · resumen en Español     · texto en Español     · Español ( pdf )