By Marc Mayes

African dry tropical forests are critical for biogeochemical cycles, water conservation and provide natural resources for over 100 million people who directly depend on these forests for their livelihoods. The region is a hotspot for land use change, with one of the highest global deforestation rates, yet remains understudied. Remote sensing offers promise for studying drivers and impacts of forest change in the African dry tropics. However, common approaches for identifying forests and mapping change rely on indicators of landscape greenness (MODIS EVI) and have struggled to identify forest change reliably in dry tropical landscapes.

We experimented with novel remote sensing approaches to study forest changes in a rapidly developing part of western Tanzania – part of the 2.7 million km2 Miombo Woodlands, home to the largest contiguous area of dry tropical forests in the world. We addressed the questions, “How, and when, are forests most distinguishable from non-forest areas?” and “How have trajectories of forest loss and gain changed between the 1990s-2000s and 2000s-2010s?”  For the first question, we studied data from MODIS, Landsat and Quickbird satellites across validation points collected at forest and non-forest field sites since 2012. The challenge was to find a way to differentiate forest cover in a highly patchy landscape where forest and non-forest greenness phenology is confounded – nearly all agriculture is rain-fed and 85% of forest species are deciduous, losing their leaves in the dry season just after harvests. We used spectral mixture analysis (SMA), a physical modelling approach that partitions satellite pixel reflectance into known land cover features (e.g. green vegetation, soil, non-photosynthetic vegetation such as wood, senesced leaves or grasses), to study how these land cover features varied across forest and non-forest sites. We found that proportions of substrate exposure and non-photosynthetic vegetation were more important than green vegetation in differentiating non-forest from forest sites.  Identifying forest cover by measures of non-green vegetation features differs from methods based on greenness measures commonly used in wetter regions, and illustrates the need for remote sensing approaches for dry tropical forests tuned to these ecosystems’ unique characteristics rather than global-scale assumptions about how to “see forests through the trees.”

We then applied our SMA-based method across a time series of 25 years of Landsat data to quantify net and gross patterns of forest loss and gain in western Tanzania. Given highly variable inter-annual forest productivity and sub-canopy burns that are prevalent throughout the landscape, we used a “Back to the Future” approach to identify areas that showed sustained signs of change for multiple years. Using this approach, we found our study area experienced a net loss of 15% of its 1995 forest cover by 2011. However, forest regrowth comprised a significant proportion of gross forest changes – over a 16-yr period total forest area gains were 43% of gross forest loss areas. Gross forest loss and gain activity showed distinct patterns of change that suggest different drivers across administrative districts. For example, we observed larger loss and regrowth areas near roads and urban areas, where there is better access to markets. The change patterns we detected can inform collection of novel ground data to identify drivers of change (e.g. fuel wood, clearing for agriculture). Our team has used these maps of forest change – particularly regrowth – to identify a chronosequence of 18 sites over which to study carbon and nitrogen cycling changes as forest stands regrow following cultivation.

Next steps in our studies of forest change in the Miombo include questions spanning coupled human-natural systems research, biogeochemistry and land management interests. We plan to scale up our approach to map change over larger regions of the Tanzanian Miombo and estimate carbon, nitrogen cycling and hydrological cycling implications of forest change. We also have made the data available to in-country collaborators to improve monitoring of forest changes in national forest areas that contribute various ecosystem services, including protection of water quality in dam catchments. Our research demonstrates an approach to remote sensing of land cover changes that is tuned to specific challenges of tropical dry forests, and capable of identifying gross forest changes in complex landscapes such as the Miombo in sub-Saharan Africa.  Beyond net forest changes, identifying gross forest loss and gain dynamics adds important granularity for studying biogeochemical, ecological and socioeconomic drivers of forest loss and regrowth.  We also look forward to chances for collaborating with other researchers studying ecosystem changes in other dry tropical forests in Africa or around the world – whether for testing if non-green landscape features can identify forest change elsewhere, or contributing data on how carbon, nitrogen and water cycles change during forest regrowth.

Citation:  Mayes, M. T., J. F. Mustard, and J. M. Melillo. 2015. Forest cover change in Miombo Woodlands: modeling land cover of African dry tropical forests with linear spectral mixture analysis. Remote Sensing of Environment 165:203–215.

Photo credit: Marc Mayes. Left: Mature Miombo forest. Right: Site naturally regenerating after 16 years of fallow.

If you have a recent publication you would like featured on the PARTNERS blog please contact trevor.caughlin (at) gmail.com

Please follow and like us:
error