Forests are the planet’s main CO2 filtering agent as well as important economical, environmental and social assets. Climate change is exerting an increased stress, resulting in a need for improved research methodologies to study their health, composition or evolution. Traditionally, information about forests has been collected using expensive and work-intensive field inventories, but in recent years unoccupied autonomous vehicles (UAVs) have become very popular as they represent a simple and inexpensive way to gather high resolution data of large forested areas. In addition to this trend, deep learning (DL) has also been gaining much attention in the field of forestry as a way to include the knowledge of forestry experts into automatic software pipelines tackling problems such as tree detection or tree health/species classification. Among the many sensors that UAVs can carry, RGB cameras are fast, cost-effective and allow for straightforward data interpretation. This has resulted in a large increase in the amount of UAV-acquired RGB data available for forest studies. In this review, we focus on studies that use DL and RGB images gathered by UAVs to solve practical forestry research problems. We summarize the existing studies, provide a detailed analysis of their strengths paired with a critical assessment on common methodological problems and include other information, such as available public data and code resources that we believe can be useful for researchers that want to start working in this area. We structure our discussion using three main families of forestry problems: (1) individual Tree Detection, (2) tree Species Classification, and (3) forest Anomaly Detection (forest fires and insect Infestation).

Unmanned Aerial Vehicles (UAV) are becoming an essential tool for evaluating the status and the changes in forest ecosystems. This is especially important in Japan due to the sheer magnitude and complexity of the forest area, made up mostly of natural mixed broadleaf deciduous forests. Additionally, Deep Learning (DL) is becoming more popular for forestry applications because it allows for the inclusion of expert human knowledge into the automatic image processing pipeline. In this paper we study and quantify issues related to the use of DL with our own UAV-acquired images in forestry applications such as: the effect of Transfer Learning (TL) and the Deep Learning architecture chosen or whether a simple patch-based framework may produce results in different practical problems. We use two different Deep Learning architectures (ResNet50 and UNet), two in-house datasets (winter and coastal forest) and focus on two separate problem formalizations (Multi-Label Patch or MLP classification and semantic segmentation). Our results show that Transfer Learning is necessary to obtain satisfactory outcome in the problem of MLP classification of deciduous vs evergreen trees in the winter orthomosaic dataset (with a 9.78% improvement from no transfer learning to transfer learning from a a general-purpose dataset). We also observe a further 2.7% improvement when Transfer Learning is performed from a dataset that is closer to our type of images. Finally, we demonstrate the applicability of the patch-based framework with the ResNet50 architecture in a different and complex example: Detection of the invasive broadleaf deciduous black locust (Robinia pseudoacacia) in an evergreen coniferous black pine (Pinus thunbergii) coastal forest typical of Japan. In this case we detect images containing the invasive species with a 75% of True Positives (TP) and 9% False Positives (FP) while the detection of native trees was 95% TP and 10% FP.

Insect outbreaks are a recurrent natural phenomenon in forest ecosystems expected to increase due to climate change. Recent advances in Unmanned Aerial Vehicles (UAV) and Deep Learning (DL) Networks provide us with tools to monitor them. In this study we used nine orthomosaics and normalized Digital Surface Models (nDSM) to detect and classify healthy and sick Maries fir trees as well as deciduous trees. This study aims at automatically classifying treetops by means of a novel computer vision treetops detection algorithm and the adaptation of existing DL architectures. Considering detection alone, the accuracy results showed 85.70% success. In terms of detection and classification, we were able to detect/classify correctly 78.59% of all tree classes (39.64% for sick fir). However, with data augmentation, detection/classification percentage of the sick fir class rose to 73.01% at the cost of the result accuracy of all tree classes that dropped 63.57%. The implementation of UAV, computer vision and DL techniques contribute to the development of a new approach to evaluate the impact of insect outbreaks in forest.

Temporal variations in N concentration and δ(15)N value of annual tree rings (1 year of time resolution) of two Japanese Black Pine (Pinus thunbergii) and three Japanese Red Pine (Pinus densiflora) trees under current breeding activity of the Great Cormorant (Pharacrocorax carbo) and the Black-tailed Gull (Larus crassirostris), respectively, in central and northeastern Japan were studied. Both species from control sites where no avian input occurs show negative values (δ(15)N = around -4 ‰ to -2 ‰) which are common among higher plants growing under high rainfall regimes. The δ(15)N values of P. densiflora show uniformly positive values several years before and after the breeding event, indicating N translocation that moved the absorbed N of a given growth year to tree rings of the previous year while a clear historical value of soil N dynamics was kept intact in the annual rings of P. thunbergii. Long-term N trends inferred from tree rings must take into account tree species with limited translocation rates that can retain actual N annual acquisition.

Abstract:In the last two decades the major focus of study in forest water and carbon balances in eastern Siberia has been on the effect of rain during the growing season. Little attention has been paid to the contribution of snowmelt water. The results of the present study indicate that weather conditions during the snowmelt period as well as the soil moisture conditions carried from the previous year's growing season strongly determined the water availability for the forest ecosystem at the beginning of the next growing season. In the forest-grassland intermingled ecosystem of lowland Central Yakutia, gradual snowmelt water flow from the forest into the adjacent grassland depressions increased when soil moisture was high and air temperature was low, whereas low soil moisture and high air temperatures accelerated soil thawing and consequently snowmelt water infiltration into the forest soil. We found that snow depth did not determine the volume of snowmelt water moving to the grassland depression since the thermokarst lake water level in the adjacent grassland was about 25 cm lower in 2005 than in May 2006, even though maximum snow depth reached 57 cm and 43 cm in the winter of 2004-05 and 2005-06, respectively. The contribution of snowmelt water to forest growth as well as the flow of water from the forest to the grasslands showed a strong annual variability. We conclude that warmer springs and high variability in precipitation regimes as a result of climate change will result in more snowmelt water infiltration into the forest soil when the previous year's precipitation is low while more snowmelt water will flow into the thermokarst lake when the previous year's precipitation is high.