Prediction of the optical water type of lakes from catchment properties

Moritz Karl Lehmann1, Ian Hawes2, Mathew Allan3 and Kohji Muraoka2, (1)Xerra Earth Observation Institute, Raglan, New Zealand, (2)University of Waikato, Hamilton, New Zealand, (3)University of Waikato, Environmental Research Institute, Hamilton, New Zealand
Abstract:
The estimation of lake water quality attributes from satellite remote sensing is difficult due to the wide range of optically active substances and their independent variability. The choice of retrieval algorithms for water quality attributes is key for successful application of remote sensing approaches to regional lake monitoring and management. We used cluster analysis on in situ hyperspectral reflectance spectra, water chemistry measurements and expert knowledge from 75 New Zealand lakes to generate optical water types. The clusters were investigated for their relationship with catchment and geomorphology-based attributes, such as land use, land cover type, terrain slope and lake geomorphology using boosted regression tree modelling. These independent attributes were able to predict the predetermined optical water type with a success rate of up to 75% in cross validation. The regression tree model provided insights into the sensitivity of the optical water type to the catchment and geomorphology based attributes, allowing further knowledge discovery of expected and archetype optical appearance of lakes under given attributes combinations. We use the regression three model to predict the optical water type of all New Zealand lakes for which catchment data is available. A priori knowledge of this type is particularly important for lakes which are small and can only be resolved by multispectral sensors with limited spectral resolution, but spatial resolution at the scale of tens of metres (e.g., Landsat 8 OLI and Sentinel-2 MSI).