Open Collections

Monitoring miombo woodlands of Southern Africa with multi-source information in a model-based framework

University of British Columbia

Acquiring forest resources information for tropical developing countries is challenging due to financial and logistical constraints, yet this information is critical for enhancing management capability and engaging in initiatives such as Reducing Emissions from Deforestation and forest Degradation (REDD+). In this dissertation, I investigated innovative approaches to monitoring forest resources and deepened understanding of multi-source information (i.e., remote-sensing, environmental, and disturbance data) needs by examining methods using a model-based framework for assessment of a jurisdictional landscape in the miombo ecoregion of Zambia. I focused on percent canopy cover (CC) and above-ground biomass (AGB) within a National Forest Inventory (NFI) context because both are important for land management and REDD+. First, I compared multi-source information and four modeling methods to estimate CC and total forest area. Landsat outperformed RapidEye and a generalized additive model was most precise. Available soil water content (AWC), slope, distance to district capital, and texture of remotely sensed data were crucial predictor variables in improving estimates. Second, I used multi-source information and compared methods with and without predicted CC in three modeling methods to estimate total AGB. A nonlinear sigmoidal model was most precise when using predicted CC, AWC, pH, occurrence of late season fire, and the Normalized Difference Moisture Index as predictor variables. Third, I developed an innovative monitoring framework using time series classification and a stock-difference approach to estimate change in land cover and AGB over a 16-year annual time series of Landsat data. Forest/nonforest change trajectories were used to develop change classes relevant to underlying biotic and abiotic factors and provided ecologically meaningful context for land cover change and AGB change. Overall, predictor variables related to soil moisture, topography, shortwave infrared bands and texture of vegetation indices from remotely sensed data are vital to accurate models of CC and AGB. Genetic algorithms provide an opportune method for predictor variable selection across a diversity of modeling methods. Further, robust change estimates are feasible when using annual monitoring methods based on freely available, multi-source information. In conclusion, the model-based framework provides a precise, statistically sound, approach to estimating and mapping forest attributes within an NFI context.

Text

2017-02-22

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/60664

Forestry

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/