A soil intactness and sediment source prioritisation framework for catchment management using deep learning and LiDAR: Tasman District, New Zealand : A dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Precision Agriculture

Abstract

This dissertation develops a high-resolution, GIS-based Soil Intactness and Sediment Source Prioritisation framework for Tasman District (New Zealand) that translates remotely sensed surface condition and LiDAR-derived terrain connectivity into council-ready decision layers for catchment management. It first produces a 1 m, 7-class land cover map using DeepLab v3+ semantic segmentation as a proxy for surface protection and exposure, then integrates this with LiDAR terrain derivatives (including slope, flow accumulation, and TWI) and proximity-based disturbance indicators derived from roads and building footprints. A continuous Soil Intactness Index (SII) is constructed using an MCDA weighted linear combination of normalised factors and reclassified into five interpretive classes for operational communication, while a separate Delivery Potential index represents the likelihood that mobilised material is transferred to waterways based on contributing area, stream proximity, and slope. Sediment Source Priority is then defined by intersecting exposure with delivery (applying delivery only to bare land) and complemented with two actionable hotspot layers that isolate bare land on steep slopes and bare land near mapped flow paths. Zonal statistics across ten “Tasman Parts” reporting units support a regional ranking system, which identifies Moutere Inlet as the highest-priority region due to an extensive bare-land footprint (82.7%) and the strongest near-stream hotspot signal, while Upper Moutere (1,173 ha) and Ngatimoti (960 ha) contain the largest bare-and-steep exposure areas that imply a need for sustained, landscape-scale intervention; limitations highlighted for operational deployment include the need for field validation and sensitivity testing of MCDA parameters.