Twenty-Five Years of Community Forestry: Mapping Forest Dynamics in the Middle Hills of Nepal
This project seeks to document forest cover change dynamics in mountainous Nepal in order to assess the influence of regional drivers such as community forest management. Though community forests occupy nearly 23% of Nepal's total forest area (1.2 million ha) and over 40% of Nepal's population is involved in community forestry, the spatially-explicit impacts of community forest-driven management on forest cover dynamics have never been documented. This shortcoming results in part from Nepal's extreme topographic relief that introduces variable solar illumination and shading on the country's forests, inhibiting repeat satellite monitoring of forest cover condition and extent. The first stage of our study directly addresses these challenges by developing a forest cover mapping approach that includes a rigorous evaluation of terrain correction approaches and a disturbance detection methodology (LandTrendr) that leverages the full Landsat time series, both of which were carried out within the Google Earth Engine analytical framework.
We evaluated three physical topographic illumination correction (TC) approaches and nine semi-empirical TC approaches that derive correction statistics from correlations of the illumination condition with various Landsat 5, 7, and 8 bands. The performance of different TC's if influenced by wavelength, land cover type, and sun angle, and we evaluated TC performance using eight separate measures related to the correlation of image bands with illumination conditions, the difference in forest reflectance before and after a given correction, and the comparison between sunlit and shaded slopes. We limited our analysis to forested areas and ranked TC's based on wavelength and their performance over different years to shows effects of variation in image composition and sun angle. Semi-empirical correction approaches like Statistical-Empirical, Variable Empirical Coefficient Algorithm, C-Correction, and most forms of the Minnaert correction performed best. The only physical TC that provided acceptable results was the Sun-Canopy-Sensor + C correction while Modified Minnaert, Cosine, Gamma, and C Huang Wei corrections performed poorly and were not used for assessing forest dynamics.
We applied the Statistical-Empirical correction algorithm on annual composites spanning 1990-2016, and input these data into a regionally-calibrated LandTrendr algorithm, which generated annual maps of forest cover loss and regeneration. A snapshot of our preliminary results southeast of the Kathmandu Valley is shown here, indicating a general expansion of forest cover over the study period that is common across the Middle Hills. Of note, these results capture long-term patterns of forest cover change regardless of illumination condition, and therefore include forest cover dynamics on regularly shaded slopes that have historically been excluded from bi-temporal change assessments. Our current efforts include validating forest cover change data, and merging remote sensing-based results with national census and field-collected interview data in community forests across Nepal's Middle Hills. Through this data synthesis, we will generate a nationally validated assessment of forest cover trends with a place-based understanding of the influence of community forestry as well as foreign labor migration and remittance income.
Comments
Post a Comment