Indigenous forests cover 23.9% of New Zealand’s land area and provide highly valued ecosystem services, including climate regulation, habitat for native biota, regulation of soil erosion and recreation. Despite their importance, information on the number of tall trees and the tree height distribution across different forest classes is scarce. We present the first region-wide spatial inventory of tall trees (>30 m) based on airborne LiDAR (Light Detection and Ranging) measurements in New Zealand—covering the Greater Wellington region. This region has 159,000 ha of indigenous forest, primarily on steep mountainous land. We implement a high-performance tree mapping algorithm that uses local maxima in a canopy height model (CHM) as initial tree locations and accurately identifies the tree top positions by combining a raster-based tree crown delineation approach with information from the digital surface and terrain models. Our algorithm includes a check and correction for over-estimated heights of trees on very steep terrain such as on cliff edges. The number of tall trees (>30 m) occurring in indigenous forest in the Wellington Region is estimated to be 286,041 (±1%) and the number of giant trees (>40 m tall) is estimated to be 7340 (±1%). Stereo-analysis of aerial photographs was used to determine the accuracy of the automated tree mapping. The giant trees are mainly in the beech-broadleaved-podocarp and broadleaved-podocarp forests, with density being 0.04 and 0.12 (trees per hectare) respectively. The inventory of tall trees in the Wellington Region established here improves the characterization of indigenous forests for management and provides a useful baseline for long-term monitoring of forest conditions. Our tree top detection scheme provides a simple and fast method to accurately map overstory trees in flat as well as mountainous areas and can be directly applied to improve existing and build new tree inventories in regions where LiDAR data is available.

Repeat frequencies of optical remote sensing satellites have been increasing over the last 40 years, but there is still dependence on clear skies to acquire usable imagery. To increase the quality of data, composited mosaics of satellite imagery can be used. In this paper, we develop an automated method for clearing clouds and producing different types of composited mosaics suitable for use in cloud-affected countries, such as New Zealand. We improve the Tmask algorithm for cloud detection by using a parallax method to produce an initial cloud layer and by using an object-based cloud and shadow approach to remove false cloud detections. We develop several parametric scoring approaches for choosing best-pixel composites with minimal remaining cloud. The automated mosaicking approach produced Sentinel-2 mosaics of New Zealand for five successive summers, 2015/16 through 2019/20, with remaining cloud being less than 0.1%. Contributing satellite overpasses were typically of the order of 100. In comparison, manual methods for cloud clearing produced mosaics with 5% remaining cloud and from satellite overpasses typically of the order of 20. The improvements to cloud clearing enable the use of all possible Sentinel-2 imagery to produce automatic mosaics capable of regular land monitoring, at a reasonable cost.

Indigenous forests cover 24% of New Zealand and provide valuable ecosystem services. However, a national map of forest types, that is, physiognomic types, which would benefit conservation management, does not currently exist at an appropriate level of detail. While traditional forest classification approaches from remote sensing data are based on spectral information alone, the joint use of space-based optical imagery and structural information from synthetic aperture radar (SAR) and canopy metrics from air-borne Light Detection and Ranging (LiDAR) facilitates more detailed and accurate classifications of forest structure. We present a support vector machine (SVM) classification using data from the European Space Agency (ESA) Sentinel-1 and 2 missions, Advanced Land Orbiting Satellite (ALOS) PALSAR, and airborne LiDAR to produce a regional map of physiognomic types of indigenous forest. A five-fold cross-validation (repeated 100 times) of ground data showed that the highest classification accuracy of 80.5% is achieved for bands 2, 3, 4, 8, 11, and 12 from Sentinel-2, the ratio of bands VH (vertical transmit and horizontal receive) and VV (vertical transmit and vertical receive) from Sentinel-1, and mean canopy height and 97th percentile canopy height from LiDAR. The classification based on optical bands alone was 72.7% accurate and the addition of structural metrics from SAR and LiDAR increased accuracy by 7.4%. The classification accuracy is sufficient for many management applications for indigenous forest, including biodiversity management, carbon inventory, pest control, ungulate management, and disease management.

The southern beech (genus Fuscospora and Lophozonia) forest in New Zealand periodically has “mast” years, during which very large volumes of seeds are produced. This excessive seed production results in a population explosion of rodents and mustelids, which then puts pressure on native birds. To protect the birds, extra pest controls, costing in the order of NZD 20 million, are required in masting areas. To plan pest control and keep it cost-effective, it would be helpful to have a map of the masting areas. In this study, we developed a remote sensing method for the creation of a national beech flowering map. It used a temporal sequence of Sentinel-2 satellite imagery to determine areas in which a yellow index, which was based on red and green reflectance (red-green)/(red + green), was higher than normal in spring. The method was used to produce national maps of heavy beech flowering for the years 2017 to 2021. In 2018, which was a major beech masting year, of the 4.1 million ha of beech forest in New Zealand, 27.6% was observed to flower heavily. The overall classification accuracy of the map was 90.8%. The method is fully automated and could be used to help to identify areas of potentially excessive seed fall across the whole of New Zealand, several months in advance of when pest control would be required.

Indigenous forests cover 24% of New Zealand and provide valuable ecosystem services. However, a national map of forest types, that is, physiognomic types, which would benefit conservation management, does not currently exist at an appropriate level of detail. While traditional forest classification approaches from remote sensing data are based on spectral information alone, the joint use of space-based optical imagery and structural information from synthetic aperture radar (SAR) and canopy metrics from air-borne Light Detection and Ranging (LiDAR) facilitates more detailed and accurate classifications of forest structure. We present a support vector machine (SVM) classification using data from ESA’s Sentinel-1 and 2 missions, ALOS PALSAR, and airborne LiDAR to produce a regional map of physiognomic types of indigenous forest in New Zealand. A five-fold cross-validation of ground data showed that the highest classification accuracy of 80.9% is achieved for bands 2, 3, 4, 5, 8, 11, and 12 from Sentinel-2, the ratio of bands VH and VV from Sentinel-1, HH from PALSAR, and mean canopy height and 97th percentile canopy height from LiDAR. The classification based on the optical bands alone was 73.1% accurate and the addition of structural metrics from SAR and LiDAR increased accuracy by 7.8%. The classification accuracy is sufficient for many management applications for indigenous forest in New Zealand, including biodiversity management, carbon inventory, pest control, ungulate management, and disease management. National application of the method will be possible in several years, once national LiDAR coverage is achieved, and a national canopy height model is available.