Machine Learning Model Performance
The results generated from the model were compared with the data collected from the field as shown in the Figure below.
The number on the horizontal axis corresponds to the cluster number (group of 4, 2 × 2 quads) and the vertical axis shows the
magnitude of biomass in Kg/m2). The green curve shows the field value whereas the blue shows the results from the model.
The red curve indicates the difference between the model
and field results.
The above graph shows that the model and the field survey results are in general agreement in most of the locations except at a few clusters where the model seems to underestimate the biomass such as in cluster 6 and 17. Upon investigation of the field conditions, these were the locations containing large boulders, which explains why the model was unable to account for the variability caused by increase in the total surface area of the substrate and the associated increased biomass.
Similarly, few locations that show the overestimation of the biomass correspond to the areas where Ascophyllum nodosum (species of interest) was intermixed with Fucus vesiculosus, which is generally lighter in colour and is spectrally similar to the Ascophyllum nodosum.
Overall, the updated model shows the error margin of +/-6 Kg/m2 for 80% of data which is a significant improvement over +/-10 Kg/m2 for the similar range of data. From a commercial perspective, it is considered a medium range of success, but not fully capable of replacing detailed field surveys. The seaweed harvesters can utilize this output to define areas of high (>12Kg/m2), medium (8-12Kg/m2) and low (<8 Kg/m2) biomass which they can target for harvesting.
It was determined that mapping of the seaweed and the general distribution of the biomass is more crucial than the exact quantification of the biomass. Thus, the technique shows great potential for resource mapping and management that can be utilized by seaweed harvesters as well as regulatory agencies.