kab11k/master_thesis

Code to supplement my master's thesis titled 'Tackling the Data Deficiencies of West Africa's Small-Scale Fisheries' # Master's Thesis: Tackling the Data Deficiencies of West Africa’s Small-Scale Fisheries The code contained in this repository supplements my master's thesis titled 'Tackling the Data Deficiencies of West Africa's Small-Scale Fisheries'. # Methods ARIMA, LSTM, and Transformer models were fit to the historical domestic fleet landings of 22 countries in West Africa. Numerous exogenous variables were tested for predictive power, including: sea surface temperature (sst), chlorophyll-a concentration (chl-a), the country's average World Bank governance score (as measured across 6 different dimensions), the level of foreign landings (both in aggregate and disagreggated between reported vs. unreported landings), the domestic fleet's fishing effort by country per year, and the resulting catch per unit of effort (CPUE) by domestic fleet and year. Models were tested under 2 periods: Period 1 forecasts years 2005-2010 while Period 2 forecasts years 2013-2019. Models were evaluated against a baseline persistence model using MSE and MAE. The best performing model (ultimately, a Transformer model fit with governance and foreign landings) was used to forecast years 2020-2030. The resulting forecasts were then included in a supply and demand model in order to estimate the 2030 supply gap. # Running instructions: NOAA_data_prep.py : connects to the National Oceanic and Atmospheric Administration's data repositories and pulls down sst and chl-a data for years 1982-2021 and 1997-2022, respectively. The resulting sst data from this code can be found saved in the 'data' folder (chl files were too big to upload to github. Please reach out if needed). This code does not need to be run unless adding additional years of data and refreshing analyses. data_clean.py : contains code to clean the data for landings and exogenous variables. The resulting data from this code can be found in the 'data' folder. This code does not need to be run unless adding additional years of data and refresh …