For 186 of the 2320 species examined here, the cross-validated MAE produced by the phenological model was identical to that estimated using the collection dates of the specimens alone . Although these species were retained for use in PhenoForecaster, it should be noted that no climate data may be entered for these species, and the resulting predictions of flowering time consist only of a constant value reflecting an estimate of the mean observed flowering date for that species, which is not influenced by local climate conditions. Additional species will be added and models will be updated as new data or superior modeling techniques become available.Although many studies have examined patterns of phenological variation in response to local climate, few tools exist for the prediction of phenological timing under novel climate conditions. PhenoForecaster provides a free, quick, and easy-to-use software package that allows researchers of any background to quickly predict the mean flowering date of angiosperm species under novel annual conditions, or at locations where the phenology of that species has not previously been observed. Its intuitive user interface and compatibility with existing spatial climate estimation packages such as ClimateNA make phenological prediction easy to accomplish by researchers of any background without the need for extensive training or familiarity with phenoclimate modeling. It should be noted, however, that the accuracy of predictions by PhenoForecaster is variable and depends highly on the species selected for prediction. The expected accuracy of PhenoForecaster output, as reflected by the MAE value for that species, should be kept in mind when dealing with predicted MFD values generated by PhenoForecaster. Furthermore,blueberry plant size these models do not account for potential heterogeneity of phenological responsiveness among populations of a given species, but instead represent mean phenological responsiveness across all available specimens for each species.
These data were also based on models trained using phenological observations throughout North America only, and using derived estimates of local climate condition produced using ClimateNA; these estimates may exhibit some differences from ground-based observations of these climate parameters, or from estimates of these climate parameters derived using different methods. Thus, predictions of the phenology of these species outside of North America, or based on different sources of climate data, should be treated with caution. In addition, it should be remembered that PhenoForecaster models the timing of MFD only, and that the relationship of MFD to other phenophases, such as leaf-out, date of first flower, or date of last flower, may be highly variable among species and across climate gradients. These predictions should therefore be treated as dates on which the individuals of a given species are likely to be in flower where they have experienced a particular suite of climatic conditions, rather than as the onset or termination date of any specific phenophase. Where possible, we also recommend cross-checking predicted MFD values generated by PhenoForecaster against observed MFD values for that species, particularly when evaluating the phenology of a species under conditions that are outside of its historical range limits. Nevertheless, PhenoForecaster represents a freely available and powerful tool that allows any researcher to conduct rapid predictions of phenological timing under past, projected, or otherwise novel climate conditions.Many agricultural robots have been developed to perform precision farming operations and replace or augment humans in certain tasks. These robots come in two main types: I) self-propelled mobile robots, and II) robotic “smart” implements that are carried by a vehicle. Type-I robots span wide ranges of sizes and designs. Conventional agricultural self-propelled machines such as tractors, sprayers, and combine harvesters have been “robotized” over the last decade through the introduction of GPS/GNSS auto-guidance systems. These machines are commercially available today and constitute the large majority of “agricultural robots”.
They can drive autonomously in parallel rows inside fields while a human operator supervises and performs cultivation-related tasks; turn autonomously at field headlands to enter the next row; and coordinate their operations. Autonomous cabinless general purpose ‘tractor robots’ were recently introduced by several companies that are compatible with standard cultivation implements. These larger robots are designed primarily for arable farming related operations that require higher power and throughput, such as ploughing, multi-row seeding, fertilizing, and spraying, harvesting and transporting. A large number of smaller type-I special purpose mobile robots have also been introduced for lower-power applications such as scouting and weeding of a smaller number of rows at a time. Most of these robots are research prototypes introduced by various research groups. A few commercial or near-commercial mobile robots have emerged in applications like container handling in nurseries and seeding , respectively. Small robots like Xaver are envisioned to operate in teams and are an example of a proposed paradigm shift in the agricultural machinery industry, which is to utilize teams of small lightweight robots to replace large and heavy machines, primarily to reduce soil compaction.Recent review articles have discussed some of the opportunities and challenges for agricultural robots and analyzed their functional sub-systems ; summarized reported research grouped by application type and suggested performance measures for evaluation ; and presented a large number of examples of applications of robotics in the agricultural and forestry domains and highlighted existing challenges . The goals of this article are to: 1) highlight the distinctive issues, requirements and challenges that operating in agricultural production environments imposes on the navigation, sensing and actuation functions of agricultural robots; 2) present existing approaches for implementing these functions on agricultural robots and their relationships with methods from other areas such as field or service robotics; 3) identify limitations of these approaches and discuss possible future directions for overcoming them. The rest of the article is organized as follows.
The next section discusses autonomous navigation , as it is the cornerstone capability for many agricultural robotics tasks. Afterwards, sensing relating to crop and growing environment is discussed, where the focus is on assessing information about the crop and its environment in order to act upon it. Finally, interaction with the crop and its environment is discussed, followed by summary and conclusions. The operation computes a complete spatial coverage of the field with geometric primitives that are compatible with and sufficient for the task, and optimal in some sense. Headland space for maneuvering must also be generated. Agricultural fields can have complex, non-convex shapes, with non-cultivated pieces of land inside them. Fields of complex geometry should not be traversed with a single orientation; the efficiency would be too low because of excessive turning. Also,plant raspberry in container fields are not necessarily polygonal, they may have curved boundaries and may not be flat. Additionally, most agricultural machines are nonholonomic and may carry a trailer/implement, which makes computing turning cost between swaths non trivial . Finally, agricultural fields are not always flat and field traversal must take into account slope and vehicle stability and constraints such as soil erosion and compaction.Computing a complete spatial coverage of a field with geometric primitives is in principle equivalent to solving an exact cellular decomposition problem .Choset and Pignon, developed the Boustrophedon cellular decomposition . This approach splits the area into polygonal cells that can be covered exactly by linear back-and-forth motions. Since crops are planted in rows, this approach has been adopted by most researchers. A common approach is to split complex fields into simpler convex sub-fields via a line sweeping method, and compute the optimal driving direction and headland arrangement for each sub-field using an appropriate cost function that encodes vehicle maneuvering in obstacle-free headland space . This approach has been extended for 3D terrain .Existing approaches assume that headland space is free of obstacles and block rows are traversed consecutively, i.e., there is no row-skipping. These are simplifying assumption, as it has been shown that proper row sequences reduce total turning time substantially . However, dropping this assumption would require solving a routing optimization problem inside the loop that iterates through driving orientations, and many maneuvering/turning motion planning problems inside each route optimization; this would be very expensive computationally. Furthermore, all algorithms use a swath of fixed width, implicitly assuming that the field will be covered by one machine, or many machines with the same operating width. Relaxing this assumption has not been pursued, but the problem would become much more complicated. Planning could also be extended to non-straight driving patterns using nonlinear boustrophedon decompositions based on Morse functions , with appropriate agronomic, cultivation and machine constraints. Finally, as pointed out by Black more , row cultivation was historically established because it is easier to achieve with animals and simple machines. Crops do better when each plant has equal access to light, water and nutrients. Small robots could grow crops in grid patterns with equal space all around by following arbitrary driving patterns that may be optimal for the cropping system and the terrain. Hence the boustrophedon assumption could be relaxed and approximate cellular decomposition could be used to compute optimal driving patterns, where field shape is approximated by a fine grid of square or hexagonal cells. This approach has received very little attention, as field spatial planning has targeted existing large machines. An example of early work in this direction combined route planning and motion planning, with appropriate agronomic, cultivation and machine constraints .
The basic version of route planning computes an optimal traversal sequence for the field rows that cover the field, for a single auto-guided machine with no capacity constraints. This is applicable to operations in arable land, orchards and greenhouses that do not involve material transfer or, when they do, the quantities involved are smaller than the machine’s tank or storage space; hence the machine’s limited storage capacity does not affect the solution. For operations where the machine must apply or gather material in the field the maximum number of rows it can cover is restricted by its capacity; the same applies to fuel. Hence, route planning with capacity constraints is a more complicated version of the problem. When many machines operate in the same field there are two classes of operations which have different characteristics. The first class is when machines are independent of each other, i.e., they do not share any resources. In such cases, coordinated route planning is straightforward because the machines can simply work on different swaths or sub-fields of the field; possible crossings of their paths at the headlands and potential collisions can be resolved during task execution. The second class is cooperative field operations, also known as in-field logistics, which are executed by one or more primary unit/s performing the main task and one or more service unit/s supporting it/them. For example, in a harvesting operation a self-propelled harvester may be supported by transport wagons used for out-of-the field removal of harvested grain . Similarly, in fertilizing or spraying operations the auto-guided spreader/sprayer may be supported by transport robots carrying the fertilizer/sprayer for the refilling of the application unit. Agricultural tasks are dynamic and stochastic in nature. The major issues with off-line route planning are that it breaks down in case of unexpected events during operations, and it can only be performed if the “demand” of each row is known exactly in advance. For example, if a sprayer’s flow rate is constant or the crop yield is known in advance, the quantity of chemical or harvest yield of each field row can be pre-computed and optimal routing can be determined. However, yield maps are either not available before harvest or their predicted estimates based on sampling or historic data contain uncertainty. Also, robotic precision spraying and fertilizing operations are often performed “on-the go” using sensors, rather than relying on a pre-existing application map. Hence, information is often revealed in a dynamic manner during the execution of the task. Vehicle routing for agricultural vehicles is based on approaches from operations research and transportation science. Optimal row traversal for a single or multiple independent auto-guided vehicles has been modeled and solved as a Vehicle Routing Problem . This methodology was conceptually extended to include multiple identical collaborating capacity-limited machines with time-window constraints, and to nonidentical vehicles . A review of similar problems in transportation science is given in . The problem of visiting a set of known, pre-defined field locations to take measurements or samples is not an area coverage problem, and was recently modeled as an orienteering problem for non-collaborating robots, and as VRP with time windows for capacitated cooperating vehicles .