Evaluation of low-cost Earth observations to scale-up national forest monitoring in Miombo Woodlands of Malawi
Kadzuwa, Henry Harry
This study explored the extent that low-cost Earth Observations (EO) data could effectively be combined with in-situ tree-level measurements to support national estimates of Above Ground Biomass (AGB) and Carbon (C) in Malawi’s Miombo Woodlands. The specific objectives were to; (i) investigate the effectiveness of low-cost optical UAV orthomosaics in geo-locating individual trees and estimating AGB and C, (ii) scale-up the AGB estimates using the canopy height model derived from the UAV imagery, and crown diameter measurements; and (iii) compare results from (ii), ALOS-PALSAR-2, Sentinel1, ESA CCI Biomass Map datasets, and Sentinel 2 vis/NIR/SWIR band combination datasets in mapping biomass. Data were acquired in 2019 from 13 plots over Ntchisi Forest in 3-fold, vis-a-vis; (i) individual tree measurements from 0.1ha ground-based (gb) plots, (ii) 3-7cm pixel resolution optical airborne imagery from 50ha plots, and (iii) SAR backscatter and Vis/NIR/SWIR bands imagery. Results demonstrate a strong correlational relationship (R2 = 0.7, RMSE = 11tCha-1) between gb AGB and gb fractional cover percent (FC %), more importantly (R2 = 0.7) between gb AGB and UAV-based FC. Similarly, another set of high correlation (R2 = 0.9, RMSE = 7tCha-1; R2 = 0.8, RMSE = 8tCha-1; and R2 = 0.7) was observed between the gb AGB and EO-based AGB from; (i) ALOS-PALSAR-2, (ii) ESA-CCI-Biomass Map, and (iii) S1-C-band, respectively. Under the measurement conditions, these findings reveal that; (i) FC is more indicative of AGB and C pattern than CHM, (ii) the UAV can collect optical data of very high resolution (3-7cm resolution with ±13m horizontal geolocation error), and (iii) provides the cost-effective means of bridging the ground datasets to the wall-to-wall satellite EO data (£7 ha-1 compared to £30 ha-1, per person, provided by the gb system). The overall better performance of the SAR backscatter (R2 = 0.7 to 0.9) establishes the suitability of the SAR backscatter to infer the Miombo AGB and fractional cover with high accuracy. However, the following factors compromised the accuracy for both the SAR and optical measurements; leaf-off and seasonality (fire, aridness), topography (steep slopes of 18-74%), and sensing angle. Inversely, the weak to moderate correlation observed between the gb height and UAV FC % measurements (R2 = 0.4 to 0.7) are attributable to the underestimation systematic error that UAV height datasets are associated with. The visual lacunarity analysis on S2-Vis/NIR/SWIR composite band and SAR backscatter measurements demonstrated robust, consistent and homogenous spatial crown patterns exhibited particularly by the leaf-on tree canopies along riverine tree belts and cohorts. These results reveal the potential of vis/NIR/SWIR band combination in determining the effect of fire, rock outcrops and bare land/soil common in these woodlands. Coarsening the EO imagery to ≥50m pixel resolution compromised the accuracy of the estimations, hence <50m resolution is the ideal scale for these Miombo. Careful consideration of the aforementioned factors and incorporation of FC parameter in during estimation of AGB and C will go a long way in not only enhancing the accuracy of the measurements, but also in bolstering Malawi’s NFMS standards to yield carbon off-set payments under the global REDD+ mechanism.