Biomass prediction of Typha latifolia on a paludiculture site by combining structural and spectral features from UAS data

Drained peatlands need to be rewetted to reduce carbon emissions. To realise this, sustainable land-use alternatives must be implemented after rewetting. Uncrewed Aerial Systems (UAS) can aid in monitoring crop growth and spatial heterogeneity in vegetation patterns, and hence contribute to improved management. We monitored Typha latifolia (‘Typha’) biomass for an 8.5 ha rewetted paludiculture site in north-east Germany using structural (digital surface model, DSM) and multispectral data (5 spectral bands and normalized difference vegetation index, NDVI) obtained by drone surveys in July, August and September 2021. We used in-situ harvests of Typha from 1 m2 square plots as training data. Biomass for validation plots was predicted from field measurements for the respective observation dates. The DSM’s and NDVI’s spatial resolution (2.76 cm) were resampled to 1 m and original values aggregated into different spatial metrics (i.e., percentiles of pixel height and NDVI values). Different regression models were separately tested for the different August DSM and NDVI metrics as explanatory variables from August data. A normalised Typha fraction cover mask from the multispectral data was used to exclude non-target species. To test the model for different phenological stages, we then applied the best performing model of August to July and September. The models were compared to non-destructive biomass predictions from linear relationships between field measurements. The combination of DSM or NDVI metrics with the Typha mask captured the heterogeneity of Typha biomass well (R2 = 0.65–0.71). Biomass overprediction for present non-target species was successfully excluded. DSM models outperformed NDVI models for dense Typha stands due to saturation of the NDVI at 500 g m-2 biomass. We were able to show biomass accumulation from July to August of up to 200 g m-2. The September model had the lowest performance (R2 = 0.6), due to a weakened height-biomass relation. Further, our model underestimated flower biomass. As single UAS surveys offer both structural and spectral information, UAS data will contribute to precise biomass and vegetation monitoring at high spatial resolution in upcoming rewetting efforts.