Estimation of Woody and Herbaceous Leaf Area Index in Sub‐Saharan Africa Using MODIS Data

Abstract

Savannas are widespread global biomes covering ~20% of terrestrial ecosystems on all continents except Antarctica. These ecosystems play a critical role in regulating terrestrial carbon cycle, ecosystem productivity, and the hydrological cycle and contribute to human livelihoods and biodiversity conservation. Despite the importance of savannas in ecosystem processes and human well‐being, the presence of mixed woody and herbaceous components at scales much finer than most medium‐ and coarse‐resolution satellite imagery poses significant challenges to their effective representation in remote sensing and modeling of vegetation dynamics. Although previous studies have attempted to separate woody and herbaceous components, the focus on greenness indices and fractional cover provides little insight into spatiotemporal variability in woody and herbaceous vegetation structure, in particular, leaf area index (LAI). This paper presents a method to partition 1 km spatial resolution Moderate Resolution Imaging Spectroradiometer (MODIS) aggregate green leaf area index (LAIA) from 2003 to 2015, into separate woody (LAIW) and herbaceous (LAIH) constituents in both drought seasonal savannas and moist tropical forests of Sub‐Saharan Africa (SSA). In our analysis, we use an allometric relationship describing the variation in peak within‐canopy woody LAI of dominant tree species (LAIWpinc) across gradients in mean annual precipitation, coupled with independent estimates of woody canopy cover (τw), to constrain seasonally changing LAIW. We present the LAI partitioning approach and highlight the broad spatial and temporal patterns of woody and herbaceous LAI across SSA. The long‐term average 8 day phenologies of woody and herbaceous LAI (averaged across 2003–2015) are available for evaluation, research, and application purposes.

Fire in sub‐Saharan Africa: The fuel, cure and connectivity hypothesis

“…The average annual maximum green LAI is used in this analysis as a proxy for fuel load, based on the logic that, in tropical savannas burning senescent leaves rather than wood, there is very little carry‐over of leaf biomass between years (i.e., it is generally eaten, burned or decomposes at annual time‐scales), thus peak leaf area during the rainy season will be closely correlated with the amount of leaf biomass that senesces and cures to constitute fuel during the subsequent dry season. The MODIS aggregate (total) leaf area index (Myneni, Knyazikhin, & Park, ) and a recently derived woody–herbaceous partitioned LAI product (Kahiu & Hanan, ) were used as proxies for fuel load, analogous to the IFP approach (total biomass) and the FCC approach (herbaceous only), respectively.…”

“…The partitioned LAI products are generated using the 8‐day LAI A at a spatial resolution of 1 km for the time period 2003–2015. As detailed by Kahiu and Hanan (), LAI A is separated into woody (LAI W ) and herbaceous (LAI H ) constituents in SSA, using independent tree cover estimates and an allometric relationship between mean annual precipitation and seasonal LAI maxima for dominant woody species in SSA. From the partitioned product, we computed the per pixel annual maximum LAI, which was then averaged for the 13 years of our study (Figure b,c).…”

“…At the time of this analysis, the data were accessed through fuoco.geog.umd.edu. We used summaries derived from GFED4 burnt area product for the years 2003–2015; see section 2.1.1 in the main text. Burn frequency derived from MODIS collection 6 burned area monthly product described in section 2.1.1 in the main text can be found here: https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table/mcd64a1_v006 Aggregate and partitioned LAI data (section 2.1.2 in main text) are described by Kahiu and Hanan (). Data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.v5s0j Rainfall data used for computing DSL are described and can be accessed at http://chg.geog.ucsb.edu/data/chirps/ Human population density data were from Landscan, developed by U.S. Oak Ridge National Laboratory, which we are not at liberty to share owing to the data use privacy policy; see their website for further details: http://web.ornl.gov/sci/landscan/landscan_data_avail.shtml…”

“…The LAI A was preprocessed to fill missing data and reduce noise in the time series caused by atmospheric contamination, sensor and solar geometry issues (Chen et al, 2004). The preprocessing steps are detailed further by Kahiu and Hanan (2017). The preprocessed LAI A was used to derive annual average maximum LAI (LAI Amax ; Figure 4a).…”

“…The satellite fire products include the MODerate Resolution Imaging Spectroradiometer (MODIS) active fire detection product (MCD64A1) collection 6 (C6), commonly referred to as burnt area, and the Global Fire Emissions Database Version 4 (GFED4; without small fires) burnt fraction as complementary estimates of the fire return frequency and annual average percentage burned area, respectively. The partitioned leaf area index (LAI) time series developed by Kahiu and Hanan () are used to derive estimates of the average annual maximum LAI as proxies for fuel load. The probability that fuel has time to dry sufficiently to burn (Krawchuk & Moritz, ) is indexed using estimates of DSL.…”