Rapid Assessment and Dynamics of Carbon and Nitrogen Cycling in a Suburban Native Forest Ecosystem in Response to Prescribed Burning

Abstract

The risk of wildfire has been increased in Australian forests due to climate change, poor quality of foliage and low rates of herbivory. In response, prescribed burning (PB) is applied, under specified conditions, for reducing the risk of wildfire and protecting the built-up areas by reducing the accumulated forest fuels. The application of PB, however, has shown to affect carbon (C) and nitrogen (N) dynamics in the burned forest ecosystems. For example, concentrations of labile C and N increase after PB due to the improved C and N mineralisation and incorporation of ashes in soils. Repeated burning, however, results in the reduced soil C and N concentrations over the time after PB. Volatilization of C and N from the soil during burning is an important factor decreasing the soil C and N levels in the affected forests. The variations in soil C and N pools influence the plant ecophysiological status, litterfall mass production (LM) and litterfall quality. Although there are studies investigating the effects of PB on soil C and N dynamics in forest ecosystems, the information is still limited about the short- to long- term effects of PB on soil C and N cycling in subtropical urban forests. Hence, a 23- month study was conducted to understand the variations in C and N pools after PB in Toohey Suburban Forest, South-East Queensland, subtropical Australia. Since the measurements of C and N concentrations in soil and plant samples using mass spectroscopy are expensive and time-consuming, the possibility of using hyperspectral image analysis, as a cost-effective and rapid technology, for predicting C and N concentrations in soil and litterfall samples was also investigated. The objectives of this study, addressed in each chapter, were (1) understanding the short- (within the first 12 months) to medium-term (up to 23 months) variations in soil C and N pools after PB using frequent (monthly to every three months) sampling (Chapter 3); (2) investigating the short- to medium-term effects of PB as well as climatic factors on litterfall quantity and quality in the forest ecosystem (Chapter 3); (3) elucidating the mechanisms involved in the short- to medium-term changes in C and N cycling in the forest ecosystem subjected to PB (Chapter 3); and (4) investigating the possibility of using hyperspectral image analysis to predict TC, TN, δ13C and δ15N of the litterfall (Chapters 4) and soil (Chapter 5) as well as C functional group distributions in decomposing litterfall samples (Chapter 6). In Chapter 3, the dynamics of labile C and N pools under the influence of PB in Toohey Forest were investigated. Soil and litterfall samples were collected from two study sites (B0 and B1). The soil and litterfall samplings were commenced at the site B0 from month 1 to month 12 after PB and at the site B1 from month 5 to month 23 after PB. The soil and litterfall samples were collected monthly from month 1 to 6 and then every three months from month 6 to 12 after PB at the site B0. The samples were collected every three months at the site B1 (month 5 to 23 after PB). Soil labile C and N pools, soil C and N isotopic compositions (δ13C and δ15N), litterfall LM, TC, TN, δ13C and δ15N and tree leaf photosynthesis in dry and wet seasons in Eucalyptus Baileyana and Eucalyptus Planchoniana were determined. The gas-exchange measurements were conducted in dry (July 2015) and wet (February 2016) seasons at the site B0. The results indicated that soil labile C and N pools fluctuated within the first 12 months after PB which might be attributed to the incorporation of ashes in the soil and the re- establishment of soil microbial abundance and activity after PB. The highest values of soil δ13C were observed immediately after PB followed by a gradual decrease until end of the experiment. This high δ13C after PB might be due to the addition of ashes with high values of δ13C to the soil. The litterfall δ13C variations were mainly driven by photosynthetic capacity influenced by soil labile N variation after PB and climatic factors (temperature). There were seasonal and interspecies variations in photosynthesis (A1,400) and stomatal conductance (gs) with their higher values observed in the dry season, which might be attributed to the more favourable temperature in the dry season. The LM was also driven by climatic factors (rainfall) and N availability after PB. Overall, soil N availability and climatic factors (rainfall and temperature) were the main driving factors of C and N cycling in the studied forest after PB. In Chapter 4, a visible to near-infrared (400–1700 nm) hyperspectral imaging system was employed to capture the hyperspectral images from the 462 litterfall samples. The spectra extracted from the images were correlated with their corresponding values of litterfall TC, TN, δ13C and δ15N using partial least-square regression (PLSR) models. The wavelengths with higher β coefficient were selected as the most important wavelengths and were used to develop the final models. The final models were, then, tested using an external validation set (90 samples). The results showed that the data of litterfall TC and δ13C could not be fitted to the PLSR model, possibly due to small variations observed in the litterfall TC and δ13C data and low resolution of the cameras (~10 nm). The models, however, were successfully developed to predict litterfall TN (R2 = 0.76; RMSE = 0.50 %) and δ15N (R2 = 0.72; RMSE = 0.98 %). The external validation R2t of the prediction was 0.64 and 0.67, and the RMSEt was 0.53 % and 1.19 ‰, for litterfall TN and δ15N respectively. The ratio of performance to deviation (RPD) of the predictions was 1.48 and 1.53, respectively for litterfall TN and δ15N. The obtained RPDs show that the models were reliable for the prediction of TN and δ15N in the new forest leaf litterfall samples. In Chapters 5 and 6, a hyperspectral camera with higher resolution (~ 1.3 nm) was used in the spectral region of 400–1000 nm to capture hyperspectral images from 120 soil samples and 118 decomposing litterfall samples. In Chapter 5, PLSR models were used to correlate the spectra extracted from the images with their corresponding values of TC, TN, δ13C and δ15N in the soil samples. The wavelengths with higher β coefficient were selected as the informative wavelengths and were used to develop the final models. The developed models provided acceptable predictions with high R2 and low RMSE for soil 93 TC (R2 = 0.82; RMSE = 1.08 %), TN (R2 = 0.87; RMSE = 0.02 %), δ13C (R2 = 0.82; RMSE = 0.27 ‰) and δ15N (R2 = 0.90; RMSE = 0.29 ‰). The prediction abilities of the final models were then evaluated using the spectra of the external test set (24 samples). The models provided acceptable predictions with high R2t and RPD of test set for soil TC 97 (R2t = 0.76; RPD = 2.02), TN (R2t = 0.86; RPD = 2.08), δ13C (R2 = 0.80; RPD = 2.00) and δ15N (R2t = 0.81; RPD = 1.94). The results indicated that the laboratory-based hyperspectral image analysis has the potential to estimate and predict soil TC, TN, δ13C and δ15N. In Chapter 6, the potential of hyperspectral image analysis for predicting C functional group distributions in decomposing litterfall samples was investigated. Particle swarm optimisation (PSO) technique was used to select the most important wavelengths before developing the models. Then PLSR and artificial neural network (ANN) models were used to correlate the most important wavelengths with their corresponding values of C functional group distributions, determined using 13C-NMR spectroscopy, in the calibration data sets (86 samples). The developed models were then validated using the external test sets (21 samples). The results showed high accuracies of prediction of alkyl- 109 C (R2t = 0.72; RPD = 1.86), O,N-alkyl-C (R2t = 0.73; RPD = 1.82), di-O-alkyl-C1 (R2 = 110 0.71; RPD = 1.78), di-O-alkyl-C2 (R2 = 0.74; RPD = 1.71), aryl-C1 (R2t = 0.76; RPD = 1.90), aryl-C2 (R2 = 0.75; RPD = 1.76) and carboxyl derivatives (R2t = 0.63; RPD = 1.43) in the external test sets using PLSR models. The ANN models also successfully predicted 113 O,N-alkyl-C (R2t = 0.62; RPD = 1.54), di-O-alkyl-C1 (R2t = 0.68; RPD = 1.76), di-O-alkyl- 114 C2 (R2t = 0.69; RPD = 1.52), aryl-C1 (R2 = 0.82; RPD = 2.10) and aryl-C2 (R2t = 0.67; RPD = 1.72) in the external test sets. With the exception of aryl-C1, the results indicated more reliable predictions of the C functional group distributions using the PLSR model compared to those of ANN models given the limited amount of training data. Neither the PLSR nor the ANN model was successful in the prediction of the carbohydrate-C and O- aryl-C. Overall, the soil labile C and N and δ13C, were influenced by the incorporation of ashes in the soil after PB. The litterfall quality was mainly influenced by N availability and climatic (seasonal) factors. The hyperspectral image analysis in combination with the PLSR modelling could successfully predict litterfall TN and δ15N, soil TC, TN, δ13C and δ15N and C functional group distributions in the decomposing litterfall samples in Toohey Forest ecosystem under PB.