Welfare Assessment Protocol: Initial Development Complete

Summary

As part of our commitment to make our programs more evidence-based, last year we launched a project to develop a contextualized protocol for assessing the welfare of Indian major carps in Andhra Pradesh, India. This blog post summarises the rationale for developing a welfare assessment protocol, how it was developed, and the final output.

We applied an iterative development process to allow us to make improvements over time to a preliminary version of the protocol as we learned lessons from testing at real farms. Field work for the iterative development process ran from December 2024 through August 2025, involving testing the protocol at 73 farms across Andhra Pradesh. The impetus for developing the welfare assessment protocol was to create an evaluation and decision-making tool for FWI to help us better assess fish welfare. Going forward, we are planning to utilise this tool as part of our programming, but we are currently considering the best way to use and operationalize it.

Background

The Problem: Limitations of Using Water Quality Parameters as a Proxy For Assessing Welfare at Fish Farms

The mission of the Fish Welfare Initiative (FWI) is to improve the welfare of farmed fishes as much as possible. FWI’s current core program is the Alliance for Responsible Aquaculture (ARA). ARA ground teams collect water quality data from member farms and provide the farmers with recommendations for corrective actions in the event of key water quality parameters being outside an acceptable range, indicating that the welfare of the fishes may be compromised. 

FWI currently relies on water quality parameters as a proxy for fish welfare, with the assumption being that if water quality is poor, fishes suffer. However, relying solely on specific water quality thresholds as proxies for welfare may lead to both false positives (identifying welfare issues where none exist) and false negatives (missing actual welfare problems). Studies have shown that fishes can experience compromised welfare even when water quality parameters fall within "acceptable" ranges. A more holistic assessment of fish welfare—rather than relying solely on water quality as a proxy—is important to better assess the welfare of fishes.

The Solution: Developing a Methodology for Holistically Assessing Welfare at Aquaculture Ponds

To address the limitation of using water quality as a proxy for welfare, we set out to develop a more holistic method of assessing fish welfare which takes into account not only water quality, but other environmental, health, nutritional and behavioral indicators.  

According to the “Five Freedoms” of animal welfare, animals must be kept free from hunger or thirst, discomfort, disease, and distress, and have the freedom to express normal behaviour. Over the years, this has evolved into the concept of the “Five Domains” model of animal welfare, which focuses on four domains (environmental, health, nutritional, and behavioral) that can greatly impact the fifth affective domain (mental state). The Five Domains can be applied to fishes in aquaculture systems. Any parameter that can be measured and provides information on the extent to which fishes’ needs are being met can be considered a potential indicator of fish welfare. 

We sought to develop a welfare assessment protocol, built around the Five Domain model, that is practical, feasible, reliable, and inexpensive to utilize at all farms supported—or under consideration—by FWI to provide meaningful data to inform welfare conditions at individual fish farms. A tool which consistently produces similar results when implemented by the same or different observers—following appropriate training—could serve as a critical evaluation and decision-making tool for FWI.

The Development Process: How Our Welfare Assessment Protocol Was Developed and Field-Tested

With the goal of developing a welfare assessment protocol that is practical, feasible, reliable, inexpensive to utilize, sufficiently informative to serve as an evaluation tool, and relevant to Indian major carp (IMC) farmers in Andhra Pradesh, an iterative development process was applied. Such a process allowed us to make improvements over time to a preliminary version of the protocol as we learned lessons from testing at real farms. Field work for the iterative development process ran from December 2024 through August 2025, involving testing the protocol at 73 farms in Andhra Pradesh.

The starting point for this process was a preliminary “beta” version of the protocol, specifically developed around a set of indicators (in line with the Five Domains model) that were considered informative for holistically evaluating fish welfare in ponds, while being considered practical and feasible for FWI personnel to assess. This beta version was inspired by welfare assessment tools previously developed by FAI Farms for tilapia and grass carp. The intention was to test this beta protocol at real farms, making improvements over time as we learnt lessons from each farm, with a view to contextualizing the protocol to the realities of IMC farms in Andhra Pradesh. A two-phased iterative development process was applied, with both phases based around testing the protocol in real aquaculture ponds:

• Phase 1 focused on improving the protocol based on lessons learnt during the field testing. This aimed to make the protocol more practical to conduct, clearer for personnel to use, better contextualised for the farms supported by FWI, and, ultimately, more informative for assessing the welfare of fishes.

  
  

• Phase 2 focused on ensuring consistency in scoring each indicator so that the key output—the overall classification of welfare—shows limited variability when assessed by different assessors. This phase was intended to identify if changes were needed to the output from Phase 1 (i.e. the refined version of the protocol following iterative changes) to further improve how the indicators are assessed and/or scored, or inform the need for clearer guidance to assessors so that their assessments are less subjective, with the overall intention of ensuring the final assessment of welfare is independent of the person carrying out the assessment.

Staffers Mani (left) and Teja (right) assessing health indicators of IMC during a welfare assessment of a pond. This was part of Phase 2 of the development process of the welfare assessment protocol, which studied how closely the assessments of two independent assessors matched with each other. Note that this sampling procedure for assessing health indicators is stressful for the animals sampled, but our opinion is that a better understanding of the welfare of these animals outweighs the immediate stress caused.

Final Output of the Iterative Development Process

The iterative development process led to a revised protocol that is contextualized to the realities of IMC farms in Andhra Pradesh. The protocol evolved from an initial 28 indicators to a refined set of 31 indicators. Changes to the indicators (deletions, additions, modifications) and how the protocol is operationalized compared to the original beta protocol incorporated lessons from field implementation, inter-assessor reliability testing, and statistical validation.

The key points of the welfare assessment protocol are as follows:

• Each assessment is conducted by two people: a primary assessor from FWI, working in tandem with a cast-netter recruited from the locality. Such a pair can conduct two to three assessments in a single day (all in the morning).

• The key output of the assessment is the welfare status of the pond, classified as “Good”, “Fair”, or “Poor”. This is determined based on an overall welfare score (OWS), which is the final score obtained from tallying up the individual scores assigned to each of the 31 indicators evaluated during the assessment.

  
  

• An overall confidence level (OCL) is computed to indicate the level of confidence in the overall welfare score. This metric is particularly useful when some indicators can not be evaluated during the assessment. The OWS from a farm at which all 31 indicators were assessed will have a higher OCL compared to the OWS from a farm at which some indicators could not be assessed, with some indicators being considered more critical and informative than others.

  
  

• To complement the OWS, a partial welfare score (PWS) is assigned at the domain level. While the protocol is used to holistically assess overall welfare by evaluating all domains, and the key metric for evaluation purposes is the OWS, the PWS adds value by indicating the general status of each individual domain. This is useful for flagging potential concerns that may be missed if the only output were the OWS.

  
  

• Similar to the OWS, a partial confidence level (PCL) is computed for each of the four PWSs to indicate the degree of confidence in the individual PWSs.

  
  

• Two “override” systems are incorporated to prevent biologically irrational classifications. These override systems come into play when critical welfare issues are not fully accounted for in the OWS. 

  
  

  • The first override system comes into effect in the event of high mortality, which should be an immediate red flag for welfare. We consider mortality rate to be the most crucial factor influencing the welfare of farmed fishes, taking precedence over all other indicators. In our protocol, if mortality exceeds 20%, the welfare status is automatically assigned as “Poor,” regardless of the OWS.

    
    

  • The second override system takes into account critical issues at the level of individual domains, which may be masked by higher scores elsewhere during the assessment. The rationale for this override system is that a critical failure in any single domain indicates a severe welfare compromise; similarly, moderate failures across two or more domains are not conducive to good welfare regardless of how good other domains are.  

We are planning to publish the welfare assessment protocol, reporting on its development and final output, in a peer-reviewed journal. We will share complete details about the protocol once this has been published. If you are interested in viewing the tool in its current format, please contact Paul.

Our team being trained ahead of Phase 2 of the iterative development process. This phase focused on assessing if the tool can consistently produce similar results when implemented by different assessors.

Path Forward

While we are proud of the tool we have developed, there has been significant internal uncertainty about whether and how to utilize it. In particular, it remains unclear whether a more holistic welfare tool is useful for evaluating a program, such as our ARA, which only aims to improve one dimension of fish welfare—water quality.

However, we are sufficiently excited where we do wish to further test and develop our welfare assessment protocol. This is partially because we are interested in gaining a better understanding of how good—or bad—welfare is at various types of fish farms generally, as well as because we think it may be useful to have a standardized method of evaluating our programs.

To this end, we are currently planning to hire a small (1-2 person) monitoring and evaluation team in 2026, and task them with utilizing and further developing the welfare assessment protocol. We have yet to determine exactly what this team will do, but current options include using the welfare assessment protocol to:

• Compare ARA vs non-ARA farms.

• Assess ARA farms pre- and post-joining.

• Conduct needs assessments on new regions.

• Assess new interventions (e.g., feed fortification).

As always, we will keep our blog updated with our plans. For further information, see our internal monitoring and evaluation proposal.