Local | PANORAMA

Wetland Vegetation Type Identification

Using a Gaussian function, vegetation index time series were fitted to reduce noise and extract key features. A random forest deep learning algorithm classified wetland vegetation into three types (Spartina alterniflora, Phragmites australis, Suaeda salsa). Field validation confirmed classification accuracy from 1990–2022.

Noise in raw vegetation index curves was minimized through Gaussian fitting, improving classification accuracy.
Random forest algorithms amplified interspecies spectral variance, enabling reliable feature extraction and identification.

Spectral features (e.g., vegetation moisture and structure) were crucial for increasing interspecies variance and classification precision.
Curve fitting and noise reduction significantly improved the accuracy of temporal analyses, highlighting the importance of preprocessing raw data.

Implementation

Local

Data collection

Using the Google Earth Engine (GEE) platform, Landsat TM/OLI series remote sensing data from 1990 to 2022 were collected, covering TM5, ETM+7, OLI8, and OLI9. Key spectral bands (near-infrared, red, and green light) were fused to ensure high-quality data for subsequent analysis.

Remote sensing images with ≤10% cloud cover were selected for radiometric and atmospheric corrections through batch processing.
Thematic information extraction leveraged band combinations that emphasized vegetation characteristics (e.g., NIR bands for vegetation reflectance peaks).

Limitations in resolution (spatial, temporal, and spectral) can introduce errors, necessitating robust corrections (radiometric/geometric).
Data fusion between satellites was crucial for achieving consistent long-term datasets, but this step required additional processing to align temporal and spatial resolutions.

Inception phase

Local

Academic Communication

Results were disseminated via an academic paper in Ocean-Land-Atmosphere Research (a Science Partten Journal) and shared in AAASScience WeChat Public (Official Media of American Association for the Advancement of Science in China). The findings were also included as a case study in the Yangtze River Delta Pilot Site and included in the support of major research projects on oceanography by the National Natural Science Foundation (NSFC).

By systematically integrating remote sensing data, deep learning, and ecological analysis, the project has significantly advanced wetland conservation methodologies, offering scalable solutions for biodiversity preservation, biological invasion control, and ecosystem management globally.

Documentation and dissemination of results

Local

https://spj.science.org/doi/10.34133/olar.0048

Key Drivers of Vegetation Evolution

Exploring and analysing the key drivers of vegetation spatial and temporal distribution is of practical significance for monitoring the expansion of coastal wetland vegetation, species diversity conservation and sustainable management of coastal wetlands.

Spartina alterniflora was influenced by marine factors (e.g., salinity and wave height).
Phragmites australis and Suaeda salsa were driven by rainfall, anthropogenic activities, and interspecific competition.

Understanding these factors supports better management of invasive species and promotes biodiversity conservation.

Implementation

Local

Spatial and Temporal Characteristics Analysis of Wetland Vegetation

Vegetation classification results based on the reconstruction of phenological feature time series.

Nanjing Normal University

Using spatial-temporal data, the long-term distribution characteristics of wetland vegetation were analyzed. Tools such as the landscape pattern index, migration modeling, and expansion/decline dynamics identified distinct trends:

Spartina alterniflora patches showed high aggregation and a decreasing trend.
Phragmites australis and Suaeda salsa patches displayed lower aggregation, higher fragmentation, and increasing trends.
Vegetation migration patterns revealed significant spatial and temporal heterogeneity, with vegetation distributed in bands along the terrestrial gradient.

Implementation

Local

Data Quantification and Database Establishment

A comprehensive database was created, integrating remote sensing-derived vegetation cover with key environmental, climate, and human activity data. This includes metrics such as soil salinity, sea surface temperature, seawater salinity, and aquaculture pond locations, forming a robust foundation for further analyses.

Implementation

Local

Wetland Vegetation Type Identification

A Gaussian function was applied to vegetation index time series to extract candidate features, while a deep learning algorithm identified three major vegetation types (Spartina alterniflora, Phragmites australis, and Suaeda salsa). Field validation confirmed the model's accuracy, enabling precise vegetation classification from 1990 to 2022.

Data collection

Evolve

Consulting with the community and receiving feedback

Emmily Tunuka/African People & Wildlife

Group of people sit in chairs in a community hall

Men and women attend community meetings

Emmily Tunuka/African People & Wildlife

Man stands in front of a TV showing charts

Discussing rangeland monitoring data with the community

Emmily Tunuka/African People & Wildlife

Villages participating in the Sustainable Rangeland Initiative can then come together at a Community Technology Center to share information and make collective decisions for pasture management and active restoration for the next season based on pasture quality data as well as projections of herd size and anticipated rains.

APW works closely with each village grazing committee to refine its rangeland management plan or to assist in developing one. APW follows the existing government and traditional village structures. In case such structures do not already exist, the team helps facilitate their formation, building capacity to manage rangelands.

APW has learned the importance of working not just with village-level committees but also with larger ward-level governments. Many villages in northern Tanzania share rangeland or have adjacent pastures. Thus, it is necessary to work with neighboring villages to ensure continuity in management and connectivity of ecological benefits. Since adjacent villages may compete for high-quality rangeland, cooperative management of neighboring grazing areas is imperative. As villages are added to the program, gaps in ward-level management are filled by APW and other partners, moving one step closer to ensuring connectivity in a landscape shared by people, livestock, and wildlife.

In 2020, APW began conducting harmonization meetings that bring together different stakeholders from the village level, wards, divisions, districts, regions, different ministries, parastatal institutions, NGOs among other stakeholders to discuss and streamline different agendas in regards to rangeland management in their different areas of work and also influence policy.

Documentation and dissemination of results

Review phase

Local

Verify

Group of people observe a man using a smartphone

Rangeland team member demonstrates using the mobile data collection tool

Emmily Tunuka/African People & Wildlife

Young woman types in a mobile phone while man measure grass height in a green field

Rangeland monitors collect and input data in the field

Emmily Tunuka/African People & Wildlife

Dashboard showing charts of vegetation over time and photo points

Example of a community rangeland monitoring dashboard

African People & Wildlife

Community rangeland monitors are selected by the village grazing committees to conduct monthly monitoring of the selected plots. Monitors receive training on best practices in data collection, as well as data input and interpretation protocols. Plots are located and confirmed via the Collector application for ArcGIS. Monitoring data is input into Survey123 and submitted to a cloud-based server hosted by Esri. Data collection focuses on understanding grazing quality via greenness and percentage of bare ground; grazing availability via grass height; and change in availability via percentage grazed. Monitors also record the frequency of invasive species and take a picture of the plot to report to the village grazing committee.

The data for each plot is analyzed in real time via the ArcGIS Dashboard. The rangeland monitors and grazing and pastoralist committees have access to their dashboards so that they can view their pasture quality data and trends at any time.

Monitoring

Local