Center For Cultural And Technical Interchange Between East and West, Honolulu, US
This project responded to the solicitation for LCLUC studies in mountainous regions by examining forest dynamics in Nepal since the early 1990s. We topographically corrected Landsat time-series composites across Nepal before we began to analyze the Landsat data. After evaluating multiple topographic correction approaches and found that the effectiveness of a given correction approach varied across Landsat bands. For this reason, we used the Variable Empirical Coefficient Algorithm (VECA) for the Blue band and the Statistical-Empirical correction (S-E) for all other bands. We then used all Landsat 5, 7 and 8 surface reflected-corrected imagery from 1992-2016 that were accessible through Google Earth Engine. We used images taken during Nepal’s growing season--July to August--with a Scene Cloud Cover Score of less 80% to generate cloud free seasonal composite images. The C Function Mask (CFMask) algorithm was used to select mostly cloud-free imagery and was combined with the Simple Cloud Score algorithm to mask clouds and cloud shadows. Finally, reflectance values from Landsat 5, 7, and 8 were harmonized using the method outlined by Roy et al. (2016) to ensure consistent spectral values year-to-year, sensor to sensor. Our classification had an overall accuracy of 90% with 87% forest user’s accuracy. However, because we could not distinguish forests from plantations, orchards, trees growing on fallow land and other trees, we will henceforth refer to tree cover instead of forest cover. In 1992, we mapped 3.88 million ha of tree cover, which is equal to approximately 26.2% of Nepal’s land area. We mapped changes in tree cover every year until 2016. In that year trees cover 6.63 million ha, an amount equal to approximately 44.9% of Nepal’s land area. Hence, during this 24-year period tree cover almost doubled across Nepal.
We conducted RandomForest and multilevel regression analyses to relate socioeconomic and physiographic variables to tree cover change to identify the most significant predictors of tree cover loss or regrowth at the village development committee (VDC) level. We found that tree resurgence is positively associated higher incomes and higher education, tree cover is positively associated with percent of households in the VDC with modern houses (a proxy for wealth), and the percent of household members who are literate. We also found tree resurgence is positively rougher terrain and north facing slopes (both proxies for marginally suited agriculture), and accessibility, as measured by the time required to travel to the district headquarters (the longer it took to get to the district headquarters the more tree cover). Finally, we found that we found that being a member of a community-forest user group had a positive association with tree cover, and tree resurgence is positively associated with receiving remittance income from children who had migrated to work elsewhere in Nepal or abroad to work. In summary, this project produced a comprehensive set of maps and datasets at multiple scales that reveal the complex story of tree-cover change in Nepal over the last 25 years. We also produced an integrated database of biophysical and socioeconomic data at multiple scales that can be used to explore the drivers of tree-cover change as well as the policy implications of these changes. The project’s significance lies in its improved methods for mapping tree cover in mountainous regions, and its integration of methods and datasets for conducting a comprehensive, interdisciplinary assessment of tree dynamics in mountainous regions.