|Paper title||From pixels to trends - How to use Earth observation data for climate change indicators in lakes|
|Form of presentation||Poster|
Climate change indicators are designed to support climate policy making and public discussions. They are important for setting, monitoring and evaluating targets and communicating changes of the investigated phenomenon. Impact indicators highlight how climate change affects certain environmental phenomena. Response indicators show how society adapts to climate change. In Germany, the Environmental Federal Agency (Umweltbundesamt) coordinates the German Adaptation Strategy to climate change. This framework comprises around 100 impact and response indicators in six clusters, i.e., health, water, land, infrastructure, economy and spatial planning/ civil protection. Indicator assessment on a national scale demands comparable data of national scale ; comparability but even availability of environmental data, however, is often challenging. Lakes, for instance, are considered as sentinels of climate change, but nation-wide data for consistent and long time series are rare.
Remote sensing of lakes experiences significant developments during the last decade. Thus, the next report of the German Adaptation Strategy aims to include remote sensing data and methods for the first time. The focus lies on four impact indicators in lakes, namely “presence of cyanobacteria” (cluster health), “beginning of spring phytoplankton bloom”, “lake water temperature” and “ice cover” (cluster water). The aim of our project is to develop an operational, retrospective processing routine based on remote sensing data for these four climate change indicators. We collected a large in-situ database for 25 lakes in Germany, for which we tested and evaluated potentially suited algorithms and sensors. We also discussed with experts and end-users the requirements on sensor-algorithms. Then, we developed different approaches to create and visualise the indicators, i.e., to obtain an easily –to-grasp figure from the remote sensing data. The results are briefly summarised below:
“Presence of cyanobacteria”:
ENVISAT MERIS and Sentinel-3 OLCI data form the data basis, Sentinel-2 is in preparation. The maximum Peak Height algorithm is used to determine presence or absence of cyanobacteria. To aggregate at lake level, we count the days with cyanobacteria presence during the season (March to October) and summer (June to September). As basis for the indicator, we set the number of days with cyanobacteria presence in relation to the number of valid image acquisitions.
“Beginning of spring phytoplankton bloom”:
ENVISAT MERIS, Sentinel-3 OLCI and Sentinel-2 MSI data form the data basis. We calculate chlorophyll-a concentrations using C2X-COMPLEX (Sentinel-2 MSI), merged algorithm derived from Maximum Peak Height following Pitarch calibration (Sentinel-3 OLCI) and C2RCC (ENVISAT MERIS) of all suited imagery acquired from March to May. The percentile 90 is used to aggregate at lake level to detect spatially variable spring blooms. From the time series, we extract the day of year and week of year at which chlorophyll-a concentration peak exceeds the 70 percentile for the first time during spring. This date then is considered as beginning of spring bloom.
“Lake water temperature”:
The Landsat 5 TM, 7 ETM and 8 TIRS thermal data form the data basis. We selected the mono-window algorithm by Sobrino/ Jimenez-Munoz combined with ERA5-Land data to retrieve lake surface water temperature. Investigation on Landsat-8 collection-2 performance is ongoing. The subsequent data analysis homogenises the results to Landsat 8 and filters outliers. The median is used to aggregate to lake level, which are then temporally averaged to monthly data. We interpolate missing monthly data if gaps are not exceeding 1 month. Gaps occur over all the year due to low revisit time of Landsat and cloud coverage. Yearly seasonal (March to October) and summer averages (June to August) are the basis for the indicator.
Landsat 8 OLI, Sentinel-2 MSI and Sentinel-1 data form the data basis. We developed sensor-specific random forest classification models to separate ice and water and mask out clouds (only optical imagery). To aggregate to lake level, we determine days when ice covers more than 80 % of the lake. Then we count the number of ice days and calculate the ratio of ice days and the number of valid image acquisitions.
Based on the above-mentioned approaches and discussions with stakeholders, we developed a framework to evaluate the data quality for the indicators. This framework indicates spatial and temporal measures of data coverage for assessing the representativeness of a value to be included in the long-term trends. Such quality measures support calculating reliable trends. Currently, we transfer the developed approaches into a retrospective, operational service for the German Environmental Agency using the cloud-processing structure of CODE-DE (National Collaborative Ground Segment). In a next step, we calculate trends and examine whether similar patterns can be derived among groups of lakes or on a national level.
Our presentation will focus on the transfer of pixel-based information into a climate change indicator, the experienced challenges, but also the new opportunities.