Spectroscopy-supported digital soil mapping

摘要: Global environmental changes have resulted in key ecosystem services that soils provide. It is necessary to up date soil information on regional and global scales ensure these continue be provided. As a result, Digital Soil Mapping (DSM) research priorities are among others, advancing methods for data collection analyses tailored towards large-scale mapping of properties. Scientifically, this thesis contributed the development methodologies, which aim optimally use remote proximal sensing (RS PS) DSM facilitate mapping. The main contributions work with respect latter (I) critical evaluation recent achievements identification knowledge gaps using RS PS data, (II) sparse RS-based sampling approach represent major variability at scale, (III) different state-of-the-art retrieve mineral from PS, (IV) improvement spatially explicit prediction models (V) integration geostatistical methods. A review existing literature about terrain was presented Chapter 2. Recent indicated large potential DSM. However, mapping, current will need extended beyond plot. Improvements may expected fields developing more quantitative methods, enhanced analysis improved transferability other areas. From findings, three interests were selected: strategies, retrieval continuous properties larger RS. Budgetary constraints, limited time available legacy restricted acquisition, 3. 15.000 km2 area located Northern Morocco served as test case. Here, sample collected constrained Latin Hypercube Sampling (cLHS) elevation data. proxy variability, alternative required supporting strategy. while minimizing acquisition efforts. This dataset representing variability. cLHS failed express spatial correlation; constraining LHS by distance criterion favoured over short distances. absence correlation sampled precludes additional predict Predicting thus modelled statistical relation between exhaustive predictor variables. For this, provided because strong spectral environment (Chapter 3 6). Concluding, considered cost efficient method acquiring resources further used derive characterize mineralogy In 4, influences complex scattering within mixture overlapping absorption features investigated. done comparing success PRISM’s MICA determining natural samples spectra. spectra developed linearly forward model reflectance spectra, fraction known constituents sample. accounted co-occurrence but eluded interaction components. found minerals could determined higher accuracy reflectance. less distinct or even absent, hampered classification routine. Nevertheless, grouping individual into categories significantly accuracy. These particularly useful scale studies, property parent material characterization formation. Characterizing described 6. Retrieval refined samples, such abundances, complex; estimating abundances requires accounts intimate mixture. can addressing non-linear 5). 5 showed mixtures estimated 2.1–2.4 µm wavelength region. First, behaviour parameterized exponential Gaussian optimization (EGO). Next, successfully predicted regression tree analysis, parameters inputs. Estimating prepared mixes calcite, kaolinite, montmorillonite dioctahedral mica field proved validity proposed method. necessity deconvolve EGO. Due nature simple representation few bands asymmetry saturation accurately Also, EGO profiles an important parameter samples. robustness handling omission during training phase tested replacing part quartz chlorite. content hardly affected. allowed than two advances has quantify wider set With science community inference final challenge Prediction especially difficulties relating variables having high correlation. 6 methodology scale-dependent observed Mineral predictions made X-ray diffraction PRISM. original lower performance those scaled Key same realized considering medium long-range Using Fixed Rank Kriging smoothing massive datasets ranges. resulting images resembled closely Further improvements multi-scale soil-landscape relationships mineralogy. maps agreement abundances. combination small sample, substantially improves feasibility quantitatively map Moreover, spectroscopic appeared sufficiently detailed Finally, modelling various thereby enhances perspective system inventorying monitoring earth’s resources. it demonstrated also essential source regional-scale Following findings thesis, concluded that: result integrated every step process, spectroscopy play role deliver manner. there issues resolved near future. Research involve operational tools properties, sensor integration, spatiotemporal allow working datasets. us future accurate comprehensive soils, and, ultimately, scale.